2026 SEND Explorer Virtual User Group Meeting

Hey. Good morning, and good afternoon, everyone. My name is Melissa Cunningham. I’m the product owner of the Send Explorer application, and I wanted to welcome you to our fourth annual user group meeting. Whether this is your first or fourth meeting, I am thrilled to see each of you here today.

This is really one of our favorite days of the year. We get to see so many customers in one place, show you what’s new, and just as important to get your feedback.

As the team and I were putting this event together, there was a lot of time to reflect over the past year. And one word kept coming to mind, and that is change, change and evolution.

And this is really the theme that you will see running throughout our presentation today.

As we look at our industry and even our home life, AI has really taken over. It’s been a tool that people are using, whether it’s in the pharmaceutical industry, biotech, or even in the regulatory space, it’s impacting the way we are working. And you’ll see that in SENDEXPLORER today, where it stands, and where we’re taking it in the future.

There’s also a lot of novelty and new things happening in the nonclinical space. There’s real momentum around new approach methodologies or NAMS, utilizing your interim study analysis, looking to see if you can use virtual control animals to replace animals in your studies and also in the predictive toxicology space. And I’m sure that you are starting to see or seeing these changes within your own organizations.

My team has changed too. So Peggy and Joyce have retired. And I personally want to thank them for everything that they’ve done to get NextGen to where we are today, and for all of their assistance with each of you throughout the years.

We’ve been very fortunate though to have Samantha Martin joining us, who’s going to introduce herself very shortly.

And the product itself is evolving too. Nextgen is going to be available at the end of this month. It’s a major rewrite. It’s been a humongous undertaking, and I’m very proud of both the product and development teams who have put everything into this latest release to make it happen.

We’ve been a little bit quieter than usual because of this feat, but I’m really looking forward to the upcoming year when we can start working closely together with each of you to implement the new changes.

I do hope that today inspires you as you see the direction that the product is taking and how it can change your day to day activities as you interpret your studies.

So with me today, I have the pleasure of co speaking with Kevin Snyder, who’s the Director of Nonclinical Innovation and Emerging Technologies Jamie Gilliam, who is a Senior Product Manager on the SENDEXPLORER team, and Samantha Martin, who is a Product Manager on the SENDEXPLORER team as well.

I do have a few housekeeping topics to discuss though before we get to our content. Our aim is to wrap up on time and stick to our schedule.

We ask that everybody mute their lines. We do hope that you will ask questions. So on the upper right hand side of GoToWebinar, there is a question feature.

In the off chance that you experience technical issues, please use the chat feature, also on the upper right hand side, and the marketing team will try to assist you. We’ll be recording our session today, and we will make that link available to you about a week from now via email. And we do love feedback. So whether it’s questions about the content that you’re seeing today or anything else, please feel free to reach out, and we hope that you’ll fill out the survey at the conclusion of today’s presentation.

So a little bit about today’s agenda. Samantha is going to walk us through seven four o, so what’s happening and what the impact will look like. Jamie is going to give a great walkthrough of the next gen workflow and some of the new features that will be available to you.

Kevin will give a talk called, is it real? Is it relevant? Is it safe? Contextualizing your toxicology data with the SENDEXPLORER data lake. I’ll follow it up with upgrade details and a little bit more about our road map for version 13.1 and version 14. The team is gonna gather together to take questions, and then I’ll do a very brief wrap up.

But for those of you on the call that are a little bit newer to the Sitara family, I wanted to speak a little bit about our mission, and that is to solve complex challenges in drug discovery and development through scientific expertise, advanced biosimulation, and innovative technologies.

So we’re very proud to have 2,600 companies that we support and 23 global agencies that have adopted our technology.

We have over twenty years of industry leadership and innovation experience, and fifteen hundred employees call Sitara home, four thirty of which have their PhD. And we’re spread throughout 33 companies across the world. It’s a really great place to work.

But more importantly, I am very proud that we are able to partner with each of you to bring lifesaving medicines to our patients in need. So more than 90% of all novel drugs approved by the US FDA since 2014 were supported either through our services or technology.

And then finally, I wanted to just share some of our applications that we have and services. So at the center of this Explorer, or I like to think it is.

But really, Sitara offers a fully integrated suite of solutions that span the entire drug development life cycle from discovery through launch and post market.

And this slide really illustrates how our platforms and consulting services really work together to accelerate development and improve decision making.

On the left hand side of the screen, you see Sitara IQ. And this is an AI powered QSP platform, and it helps to predict biological responses.

We also have ChemAxon, which is a chem informatics suite of tools with a design hub, and it fully integrates into D360. This is an application that can aggregate all of your research data in one place.

But as we start moving into the preclinical space that we’re all much more familiar with, we start seeing deeper modeling and simulation with our SimShip platform. And for those of you on the call that are DMPK scientists, we have Phenex, which is our PKPD platform for analysis and modeling, and that’s moved to the cloud this year, which has been very exciting for that team.

We also have Pinnacle 21, and this can help validate your data before it goes into SEND explorer or before a submission.

We also have global submit to help with submissions and coauthor. I did wanna call them out because this can really help facilitate your report writing and even creation of CTD tables. So it can be a huge time saver for those tasks that take a lot of time.

We have two applications that aren’t shown on this slide as well that are quite helpful in the preclinical space. We have something called Tox Studio, where you can predict toxicological outcomes using in silico technologies, and this can help benefit streamlining your workflows, and also reducing the number of animals used in testing.

And we also have DIDB, which is the Drug Interaction Database. It’s the industry’s largest scientifically curated database to assess drug interactions and safety.

So if you’re interested in any of the different applications or services that we offer, you can get in contact with your sales rep. Satara.com also has a lot more information, or I personally am very happy to put you in contact with the right colleague.

So with that, I am going to turn it over to Samantha. She’s going to run through our first polling question of the day and then discuss Zen four o.

Thank you.

Hello, everybody. The first polling question for the day is where is your organization and its preparation for Send four point o?

Whether you started planning, you haven’t started planning, you have no idea what Send is, and this is all foreign, give you a couple of seconds to answer the question.

Okay, it looks like most people are still assessing the impact on your current SEND data sets. Some haven’t started planning and some are actively updating their processes. Good.

So again, hi, everybody. My name is Samantha Martin. I am a manager of product management for SunExplorer. I’ve been with SIRTAR for about five months, and my background is in veterinary, science and research along with years of experience in SEND data sets and submissions. And I will be discussing the next major release of the SEND IG four point o.

For those of you who do not know, SEND stands for the standard for exchange of nonclinical data. SEND specifies a way to collect and present nonclinical data in a consistent format and is one of the required standards for data submission to the FDA. Version 4.0 is the largest revision of the guide since SEND was introduced.

It changes the domains you submit, many of the variables inside them, and the underlying approach to how non standard data is represented. My goal today is to give you a clearer picture of differences from version 3.1.1 and what that means for your studies and your data pipelines. One note to keep in mind is this overview is based on the current 4.0

revision history, public review comments are still being worked through, and some items are still marked as in progress. So because of this, some details may shift before final publication.

Here’s the scale of the four point o changes in five numbers. Nine new domains are being added, which significantly broadens the kind of nonclinical data SEN can carry. Two domains are being deprecated.

Roughly 40 new variable concepts are being introduced, many of which are added to several domains at once, creating numerous new domain level variable entries. For those of you, again, that don’t know, a variable is a specific column or data field within a domain that represents a measured, counted, or descriptive attribute of a study subject.

On the other hand, 49 existing variables are being removed, and around 200 variable labels have been standardized. This will allow the variables to be more consistent across the standard.

And beyond the numbers, there are several structural changes that we will also cover.

One of the most important changes in 4.0 is the expansion of the scope the SEND model will now be able to cover.

Until now, SEND has been strongest for the core general toxicology endpoints, body weights, clinical observations, clinical pathology, and macroscopic and microscopic pathology. Data from many specialty assessments either didn’t fit the standard or had to be forced bit through custom domains. Version 4.0 changes that. Along with nine new domains, it includes an appendix that defines the scope of the standard.

That appendix sets out which study types and endpoints the Cdisc SEND team considers competently modeled. So what’s new? SEND will expand into immunology with two new domains. Self phenotype findings covers immunophenotyping data.

Immunogenicity specimen assessment covers antidrug antibody, or ADA, data and similar assessments. These were formerly in the LB domain, and together they now have clear guidance on where data should be represented.

Genetic toxicology in vivo brings the in vivo genotoxicity results into SEND.

The new nervous system findings domain gives a home for neurobehavioral data, including the functional observational battery data. Bob data has historically been among the hardest to standardize because there are a mix of observations, counts, and scored assessments, and every lab records them a little differently.

With NB, alongside the new scoring scales domain for documenting scoring systems, FOB results can be submitted in a consistent, reviewable format.

The new ophthalmic examinations and skin test results domains are dedicated to targeted eye and dermal assessments. These were formally put into the CL domain.

Pharmacokinetics input documents the concentration data and transformations behind your PK parameters, which makes the path from PC to PP more easily traceable. And the nonstandard variables domain replaces the use of supplemental qualifier domains completely.

Finally, two domains have been removed: body weight gain and tumor findings, along with a tumor. Xbt mapping appendix.

These changes allow for more of your study data to be submitted in a standard form, which means more of it can be reviewed with standard tools.

So what about changes to existing domains?

Exposure changes significantly and can now represent doses that were not administered. Clinical observations now explicitly covers assumptions of qualitative food and water consumption. For example, changes in appetite are presented in the CL domain, while qualitative food and water consumption has been moved to the FW domain. Any results that are not part of routine clinical observations and are from specifically targeted tests like I just mentioned must now be reported in their respective domains and not in CL.

Demographics moves to numeric age ranges and adds an example for subjects who were never dosed. Organ measurements now supports in life measurements. And across the safety pharmacology domains, you’ll see new variables, continuous vital signs data and arrhythmia count guidance.

Microscopic findings is substantially expanded with new sections for carcinogenicity studies, sexual maturity, reproductive cycle phase evaluation, and targeted staining, each with its own assumptions and examples. Masses are now linked across domains using Mass ID with new follow-up record conventions.

For PK, transformations move to the new PI domain, and there are new examples for biodistribution in gene therapy studies and tissue concentrations.

The trial design domains now have more support for multiple test articles, and there are new examples for Latin square designs with and without washout and for split scoop designs.

On the data model, the guide has been rewritten in ways that will change how your teams read and apply it. Wherever the guide supports a conformance rule, should is now must. So practices your teams treated as recommendations may now very much well be requirements.

The domain specification tables have also been redesigned to align with STTM version three point o, and each domain now opens with an overview table so you know exactly what a domain is for before you look at a single variable. The guide has also been restructured for easier navigation and the file extension is no longer restricted to XPT format. It is important to note that the FDA’s requirements may still differ from C DISC in terms of what format is allowable for submission.

In section four of the new guide, the general assumptions has gained a set of new topics that answer some longstanding questions. This section’s main purpose is to provide guidance when it comes to dealing with complex data scenarios.

These include how to use the sequence number, how to represent multiple values in one variable, a much more detailed description of how to create a custom domain amongst many others. More information on section four can be found directly in the implementation guide.

Four appendices will provide new reference material, but two may directly stand out to you and your needs. Appendix A combines the trial summary and trial set codes and adds a structured description for treatment vehicles.

So what does this mean for CROs and sponsors? Appendix A allows for one standard approach to trial summaries and trial sets and uses such an approach across multiple datasets. Vehicle composition is also then described the same way everywhere. This means study setup is faster, there are fewer sponsor specific exceptions, it is easier to compare control groups and treatment arms across studies, and there’s less rework when validation findings come back.

Appendix C defines the scope of version 4.0, meaning which study types and endpoints the C disc send team considers competently modeled. This helps you plan and can give clearer references for scoping at much earlier study stages, like protocols.

CROs and sponsors will know upfront which endpoints and studies are in scope of send, out of scope, or which may need use of custom domains. This can help with contracts and data transfer agreements, as well as reduce surprises on time for submission.

SEND 4.0 isn’t a change to one data set or one domain. Every change together reaches almost every part of how SEND data will be produced, checked, and delivered.

Mapping specifications, programs, data transfers, and authoring software all need to support 4.0’s new domains, variables, and code lists. The same updates are required for the viewers, reports, and scripts.

Updated send conformance rules and core designations will change your validation reports, so plan time to resolve new findings.

QC checklists, comparison programs, and training need to account for the new content, and drop checks for retired domains and variables.

Define XML will have several updates, so get familiar with the new Define XML appendix. And finally, update your SOPs and templates. For some time, you will likely support both versions side by side, where 3.1.1 is for ongoing studies and 4.0 is for new ones.

Every SEND process will touch four point o. So our recommended next step, start with a gap assessment of your current SEND outlets against four point o. Talk with your teams about their readiness timelines. Update your mapping specifications and templates. Pilot a 4.0 on a representative study before you need it for submission, and keep an eye on regulatory acceptance dates so that your transition is timed to when agencies will require or accept the newest version. We’re happy to support you with any of these steps.

And how will those of us at Sartara be ready? We are very involved in the Cdisc Send team, including any and all revisions to the IG. We have been closely monitoring all the content changes to make sure that we stay informed. In terms of Zendesk Explorer, some of the changes that come with 4.0 may change some of the visualizations you are used to, especially in terms of baseline data. Our development team and product team are working together to ensure that Zendesk can be prepared to handle the new domains, variables, and processes.

So when is Send four point o expected? As of right now, Send four point o is estimated to be published at the end of q one for twenty twenty seven.

And while 4.0 develops and finalizes, we plan to keep you all updated with any Send 4.0 news and how it will affect Send Explorer and your datasets.

As a reminder, please note that any information I’ve shared today is still based on that current SendIG 4.0 revision history, and some details are still subject to change before final publication. For more information on SEND 4.0, feel free to head over to the C DISC public wiki page where you can read more on the upcoming changes and expectations. The link included on this slide will be available once this presentation is shared.

And now I will hand you guys over to Jeanie.

Thank you, Samantha.

I’m to share my screen here.

Okay.

Good morning, good afternoon, everyone, and thank you for joining. I’m Jamie Gillum, the senior manager of product management for Sun Explorer here at Sutara. If you were in attendance at last year’s UGM, we gave you a preview of the new features and the automated endpoints of interest functionality. We’re very proud of the progress that we’ve made in the past year. So during this session, I will talk about the enhancements, especially with the endpoints of interest workflow. Then I’m going to finish with a live demo so that you’re able to see the new navigation in action.

So let’s get started then. The concept of an endpoint of interest is present in Send Explorer version 12. This is a special feature that allows users to flag specific endpoints while you’re reviewing the data. The biggest benefit in version 12 is the ability to filter by the endpoints of interest in different visualizations and summaries.

And this really focuses your attention on the important aspects of the data.

SENDexplorer NextGen introduces some very exciting enhancements, and I’m gonna highlight those both in the slides and during the demonstration.

I’m gonna walk through what a data review process might look like if you’re using SendExplorer and the endpoints of interest functionality.

So currently, though, version 12 allows for manual flagging of numeric endpoints and tissues of interest. SENDEXPLORER Next Gen retains all of that existing functionality, but it also allows for manual flagging of clinical observation findings from either the ClinObs time course or the subject view. For growth and histopathology data, not only can tissues be flagged, but now specific findings for a tissue can also be flagged.

All the user approved endpoints of interest for a single study will still appear on the study overview page and also on the multi study endpoints of interest grid view. And so that’s still available where you’ll see the global list of all of the endpoints of interest that have been flagged for all of the studies within your SundExplorer application.

In addition to those, Send Explorer Next Gen has a new Endpoints of Interest details page, which we’ll see in more detail in a minute here in the demo.

But a major part of the new functionality that we’re the most excited about is the ability to auto flag the endpoints of interest. So the user can execute permissive statistical analysis that will automatically flag endpoints of interest.

This analysis is done for numeric clinical pathology and organ way data, and then also for your gross and histopathology data. Any endpoint that meets the established criteria will be marked as pending for a further scientific review. The automated analysis methods do assume that the study design is a repeat dose toxicity study and that the sponsor group codes are ordered by ascending dose with the control group first. The auto flagging is not yet recommended for other study designs or if there are multiple treatments used in the study.

While these automated endpoints of identifying…

While these automated methods of identifying pending endpoints of interest don’t cover all study designs and data types yet, users can still take advantage of the work flow by manually flagging those endpoints.

So the auto flagging for LB and OM numeric endpoints applies an overall ANOVA followed by a two tailed Dunnet’s test.

And the comparison is between the control group and the high dose group. This is done separately for males and females, and it’s only done at the last time point for which data are present prior to the start of a recovery phase.

Additionally, numeric endpoints are auto flagged if the group mean values are strictly increasing or strictly decreasing across all dose groups, again assessed separately for males and females.

For the MA and MI pathology domains, the auto flagging algorithm is going to calculate a metric separately for the control group and the high dose group. This is done for each combination of tissue and finding for subjects with a terminal disposition.

If the calculated metric for the high dose group is at least 1.5 times the calculated metric for the control group for a given sex, then the associated finding is flagged as a finding of interest for that occurrences of a finding and the high dose group has at least one occurrence, that finding will be flagged for review.

And it is important to note that the calculated metric used for auto flagging does not consider findings in any dose group other than the first and the last or the control and the high dose group. And it doesn’t look at findings associated with subjects with any disposition other than terminal sacrifice.

Okay, so there’s just a few other important details I wanted to share before moving on to the demo. This slide shows the endpoints of interest workflow.

It’s important to note that the auto flagging process is not automatically generated during the study upload, but it is executed on demand by the user. And this is intentional for several reasons, one of which is there may be legacy studies for which users do not want to auto generate those endpoints of interest. But once a user clicks generate to initiate the analysis, the endpoints that meet the criteria are marked with the status of pending. As part of your review process, you can opt to approve, ignore, or simply leave endpoints as pending for further review.

There’s also the ability to add clarifying free text comments to any of the endpoints.

And during the review, the endpoints can also be manually added. What I haven’t mentioned yet is the AI summarization piece, and this is the last step in the workflow, and it will generate a summary based on a prompt and the approved endpoints of interests. So the endpoints of interests workflow can be done as soon as you start receiving data for an ongoing study, all the way through to report generation and beyond, maybe when you want to start doing cross study analysis.

Okay, so then I am going to move on to the demo portion here.

This is the NextGen Send Explorer dashboard that you see.

So Send Explorer NextGen brings a refreshed interface, which includes a new navigation panel for moving between studies and visualizations.

In this demo, I will highlight the navigation changes, but primarily I will focus on the endpoints of interest process. You will see how the pending endpoints are flagged automatically, and then how the approved endpoints, along with the prompt, result in an AI summary. As we go, I really encourage you to think about where this process fits into your existing data review workflow and where it will save you time.

Upon logging in, the dashboard is still the landing page. Users can access the studies using the hyperlink in the Study ID column.

Each of the columns can be filtered and sorted. You can do more advanced filtering using other fields, and you can filter for, you know, studies that share a common interest. But new on the dashboard is the Select column here on the left hand side.

And this column is going to allow me to check the box for a study or a group of studies, and then certain visualizations will allow filtering for just those studies that are selected on the dashboard. This is a time saver if you are reviewing multiple studies for maybe a specific compound or a program. I’ll go ahead and select a couple here just so we can kind of take a look at it in a minute.

You may have noticed that the options in my top ribbon have changed, and the Multi Study and Visualization tabs are no longer present. By using the three bars here on the left hand corner, the new navigation menu appears. We have links to the home page and the dashboard pages at the top. Then we have tabs for single study and multi study. Right now, the focus is on the single study tab, and you can see the study pull down menu. At the top will be the last five studies that were recently viewed in the system.

Any studies that I had selected on the dashboard are present in the Dashboard section, and then there is a complete list of the studies that are available in this particular database. The filter field allows for a search of the available studies, and as I type, it is going to limit my options, and I’m going to limit to the INDS06. And when a study is selected, then my navigation menu is populated with links to the Overview page here, the new Endpoints of Interest details page, followed by all of the different visualizations, the group summary and listings, and then at the bottom here, the metadata domains. Now these are all of the views that are applicable for the selected study.

So if I just take a look quickly at the multi Study tab, this displays all of the Multi Study enabled visualizations. So we’ll first take a look then at the overview page for INDS-six.

And the Overview page presents the metadata about the study and these metadata are sourced from the SEND datasets. This page has undergone a redesign, including reordering of several sections, although first we still see the key study information followed by the Notes, Images, and Links section.

But instead of what is the test subject information in version 12, we now see a redesigned study table. And this table reflects SendExplore’s interpretation of the study design, and it’s intended to correspond with the experimental design table that appears in study protocols. So this table identifies the planned groups, the treatment, dose levels, route of administration, as well as the planned disposition events.

The next section then is the Disposition Summary table.

This table presents the actual disposition information, and by default, it is displaying the mortality counts that you see here, and those are present for each disposition event for the entire study duration. The grouping is by sex. We can see males and females, and then within that, the group and dose level. If I wanted to view the breakdown by study day or study week, I can use the time option, and I’ll first select by day. And now we can see on the left hand side the study day for each of the disposition events. And if I wanted to see them by study week, it will then roll that up into the study weeks.

To display the counts for just unscheduled deaths, I can use the toggle at the top for unscheduled.

And then in the totals drop down here, I’m gonna select both. And now I have two total rows at the bottom of the table, one with the count, and then the other is the percent of the animals that had an unscheduled death in each group for each study week.

Okay. So let me use the reset button here to return it to the default, and I want to switch to the Survival option.

Using the Survival option, the death statuses no longer appear, and my count has now switched to cumulative. If I select a time of by week and choose percent instead of count, now the table shows the percent of animals that are alive for each of my groups during each of the study weeks.

So this is a finalized study, so we would expect zero percent survival in the last week. However, this is a useful view for users who are looking at ongoing studies.

Overall, this table really does provide many options for customization, and both of these tables can be copied to your clipboard and then pasted into another document for use.

The remainder of the overview page is the same as version 12 with the additional study metadata information. If we take a look at the right side here, we see the grid for the endpoints of interest, and this is the next topic that we’ll take a look at.

And as I mentioned, the endpoints of interest functionality has been vastly improved in the next gen version of Send Explorer, and it’s been enhanced with an AI component. So from my navigation panel, I’m going to select a new study.

Go ahead and take a look at this study, and I’m gonna go to the endpoints of interest page.

And you can see here that the auto generation has not yet been done for this particular study because the auto generation process is initiated by the user. This is done by clicking the Generate option. So the auto flagging process will be applied to clinical pathology and organ weight data. Keep in mind numeric data for other endpoints can be manually flagged. Once the analysis has finished, by default we’re seeing the results for the numeric endpoints, and each of them has a pending status.

The grid is showing me the domain and the test codes, and then I have separate columns for males and females here.

If we take a look at these columns, this column will show the result of the Dunitz test. And then hovering over a particular cell, I can see even more details about that particular test.

It’s a little touchy on the screen here. There we go.

The next column then is our trend column. And when the group mean values are strictly increasing or decreasing, there will be arrows in this column to show the trend. Now these trends may sometimes be more informative because the analysis is being done across all of the dose groups.

The high dose group means are shown as fold of control, so you can see that separately here for males and females. And then the last two columns are Edit and Status. As you’re scrolling through the grid initially to get a sense of what endpoints stand interesting, you can use the links in the Edit column to approve, ignore, or edit a finding.

The information in the Status column lets me know if an endpoint has been approved and by whom, or if it’s still pending or ignored.

As I’m analyzing these endpoints of interest, I have the ability to drill down to the actual data.

So these test codes that you see here are hyperlinked to the summary by time point view. And we’ll go ahead and take a look at one here. I’m gonna open it in a different browser window.

And this is going to be the summary by time point for the LB domain or the clinical pathology data. And drilling down to the summary data allows me to then look across multiple time points for all of my groups. I can review calculated change from baseline measures if those are present. I can look for outliers, and then I could also drill down to individual subject values if necessary. So in this example, the MCHC endpoint was flagged as a pending endpoint of interest based on the analysis on day 90, which is the end of the dosing period for this study. And now I can look at the data in the recovery phase to determine if there is still an effect.

I will come back to the summary views in a minute to talk a little bit more about some of the navigation changes. However, before we go back to the endpoints of interest details page, I wanted to point out how manually flagging endpoints can happen in this particular view. You can see there is a flag here, and this is going to allow me to approve the test as an endpoint of interest or edit one that’s already been approved.

Back on my Endpoints of Interest details page, at the top of the page, there is a Clinical Observations tab.

Now, the clinical observations and points of interest will not be auto generated. Findings must be flagged by the user, and they can be flagged in the ClinObs time course or subject view. But once a finding has been selected, it will then display on this page. At the top of the grid here, you see a link, and this option is going to allow me to view the ClinObs time course visualization. So we’ll go ahead and open that up.

And this initial view is all of the findings for this particular study with separate plots for each of the findings by sex. Now, the Findings dropdown here is where I can flag my findings of interest, and that’s done by clicking this flag icon in line with a specific finding. So we’ll go ahead and click a couple here just to add that there. Okay? And then at the top of this box is a filter option.

So I can choose this flag to then filter for just those findings of interest, and now the plot has been limited to findings of interest. And it’s a very useful way of focusing in on key aspects of the study.

So I’m going to use the navigation to go back to my endpoints of interest for this particular study, and if I go to the Clinical Observations tab, you can then see those two findings that I had added in the time course visualization.

Next at the top, we have the pathology findings tab. Now the system automatically generated an initial set of findings of interest.

The algorithm determined if there was an increase in either the incidence and or severity in the high dose group that is significantly different from the control group. Those that met the criteria are listed here as pending.

The grid presents the different tissues and then the findings within each tissue separate for males and females. I can use the edit column again to approve, ignore, or edit each row, and then the status will be present here.

And any of those that are approved are going to be shown on the Study Overview page, and they will be part of the filtering in several visualizations.

The links at the top of this particular tab will open the severity heat map, either for macroscopic findings or the microscopic findings, and that’s going to allow us to see more details about the specimens and findings as a whole. So clicking the link here for the view of the severity heat map for histopathology, we’ll go ahead and take a look at this.

I now have a view that’s limited to the findings and tissues that have been either approved or are still pending as endpoints of interest. So in my left hand panel under the data selection options are the tissues of interest, and you can see the approved and the pending tissues are selected. I simply need to check the not identified, and this option then shows all of the histo data for this study. And then I can scroll through the screen here and take a look at the additional findings and tissues.

I could then flag the findings or tissues of interest manually. If I wanted to flag the tissue of heart, I can do that by clicking on the flag icon that you see here. Or if I wanted to choose this particular finding of infiltration, I could also select this manually here. And any change that you make on this particular visualization is going to be reflected on the EOI Pathology page.

So at this point, I want to walk through a different study that represents the workflow after a user has reviewed all of the data and has approved and added some endpoints of interest and some annotations to those endpoints. So I’m going to change to the INDS-five study, and we’ll go ahead and look at the endpoints of interest page here.

So if you take a look then at my status column, you can see those that have been approved. There’s one still pending and several that are ignored. We can sort these columns, and now all of the approved rows sort to the top. I could also apply a filter and simply look at just all of those that are approved. If I were to navigate back to the study overview page, any endpoint that is marked approved is going to appear on that page.

So I’ve mentioned a couple of times the edit feature, and a really… This is a really nice feature because you can click the edit link here, and now we see a pop up, and that is going to allow you to add additional text to the notes field, and you can see that the additional text has already been added to this particular endpoint.

The approved status is also very important here.

Just remember, any of these that are marked as approved are going to be part of the text that’s analyzed by the AILLM. So once the review is complete and all of these selections are approved, you can click the Summarize tab here.

What we will see then in the left panel is the prompt instructions. And the default text is applied to all studies.

It can be edited globally for all studies if you have administrator privileges, or it can be edited here at the study level. And this particular prompt is instructing the LLM to generate the summary for this study that is formatted and styled in accordance with the ICH M4S guidance for Module 2.6.6.

Also, embedded in this prompt, if I continue to scroll here, are going to be some formatting instructions, and at the bottom we will find some study specific metadata here that it is sourcing from the study, and then the dose level details.

In the endpoints of interest summary dropdown is all of the approved endpoints of interest, along with the statistical information and then any user annotations. And this could also be edited if necessary. When you’re ready, you would click Summarize with AI, and then the prompt and the endpoint summary details are going to be analyzed by the AI LLM, which returns a summary to the page. And I’ll go ahead I’ve already generated it just for the purposes of today’s demonstration. So this has generated the AI summary, and you could then copy this into Word for further review and editing.

Okay. So now that we’ve looked at the endpoints of interest workflow, I do wanna look a little more at the navigation. I looked briefly at the LB summary by time point view, and I wanna took… Talk a little bit more about the summary view to really highlight several of the enhancements.

If we take a quick look back at the navigation panel, the group summaries and listings are here in the middle, and these allow us access to the individual data grids and then the summary views for each data type for the selected study. The data type that different visualizations are going to be a default view for each of the different data types. So for the LB here, we see the default view of the summary by time point.

And instead of the configuration options across the top, like you’re used to in version 12, we now see configuration options here in the left hand panel. And this redesign is a lot more consistent with other visualizations across the application.

Also in the left hand panel here, the domains pull down menu allows you to move quickly to another domain.

You could also change the data view to either the data listing, which is the individual data grid. In this instance, we could go to the summary by test code, summary by group, and then you have your pivot views of the individual grid itself.

You can then modify the group by options here. You can select or deselect the stats that are shown on the grid. You can turn on the reference range highlighting. And then we also have an endpoint filtering section where you can filter for endpoints of interest, or you could additionally select a filter for control ranges, statistical analysis filters, and category filters as applicable.

Another nice feature of this new panel is that it can be collapsed. So using the two arrows here, I can open and close that panel to take advantage of the screen space that I have.

Now the functionality of the grids that you see is essentially the same, and there are columns of metadata here. There’s still the option to add, remove, or reorder the columns.

The ability to save presets still exists, but the option has been moved here to the top of the panel. And just a reminder, the presets are a great time saver because they allow you to essentially bookmark a visualization specific configuration for future use. And presets are available on almost all of the visualizations in Send Explorer. So keep this in mind as you’re looking at other visualizations throughout system because if there’s a particular visualization and there are specific customizations that work best for you, this functionality will save you time setting up that particular visualization in the future.

Melissa, can you let me know how I’m doing this time?

Give you a little time check. We just have a couple more minutes to wrap up.

Okay. Thanks. Thought maybe that was the case. So I do wanna show one last feature here for the scatterplot time course, and we’ll go ahead and navigate to the multi study tab and the scatterplot time course visualization.

I already have a visualization that is configured here. It’s part of the presets that I have saved for the demo because I was afraid of running out of time.

But what I really want to highlight here is that in version 12, the ability…

You only have the ability to plot one single endpoint. And now in SendExplorer NextGen, we can plot more than one endpoint. So this particular view has been set up to display the RDW endpoints for the dog study INDS06 and the rat study INDS05, along with the RBC values for the same dog and rat study. And across all of my plots, there is a shared x axis and study days, and then within each of the study days are the dose groups.

My y axes are optimized for the data in each plot. Our coloring that you see here is by the dose categories of control, low, mid, and high. And this configuration allows me to visualize the scatterplot effect of the individual values, as well as the mean distribution shown by the box and whisker plots. And so taking a look at these studies, you know, we can kind of get a sense of the dose related trend for both of my endpoints here.

One last thing I want to point out then on the navigation panel is this option here. It’s an additional filter option. And this has the apply dashboard selections.

So this would allow me to quickly select and filter for any studies that I had selected on the dashboard. If you remember, these are the two studies that I had actually selected. So those would be part of my filter. So again, just kind of this time saving measure here.

So the two takeaways for this particular one is the enhancement of allowing you to plot more than one endpoint and then giving you the ability to quickly filter for specific studies.

So that concludes the demo portion then of my presentation, and I’m going to go back to a polling question, and you’ll be able to answer this polling question, and then I’m going to turn it over to Kevin next.

Okay. Great. Thank you so much, Janie. That was very compelling. So if you could be so kind as to fill out if you think that this would be useful using the new workflow for your data review process, that would be super great.

Okay. Thank you everybody for filling out the survey. I think what Kevin shares next might make you consider how useful this will be. So with that, I’m going to turn it over to Kevin. Thank you so much.

Alright.

I am…

There we go. Alright.

Alright. Thanks for that, Melissa. And I’m going to start my presentation. I’m Kevin Snyder, director of nonclinical innovation and emerging technologies.

And the title of my talk is, is it real? Is it relevant? Is it safe? Contextualizing your toxicology data with SENDEXPLORER data lake.

So my talk is really about how we can leverage the new SENDEXPLORER data lake to answer these three questions.

So this is kind of an outline of where the talk is going. We’re gonna start with talking about is it real, and the answer… The approach we’re gonna take to that is to discuss how historical control data can be visualized and analyzed via a Sendiver R package.

And this is help… Can help determine whether something is treatment related… A finding is treatment related or incidental.

They can provide robust historical control reports for all sorts of purposes. And then I’ll talk about some future directions about how we may be able to leverage Bayesian inference and incorporate that into our analyses to better interpret underpowered studies and take a more empirical approach to resolving treatment relatedness of effects in context of historical information.

And we talked about, is it relevant?

I’m gonna discuss a bit of the endpoints of interest that Jamie just just shared about and talk about some future directions, how we can expand that by applying LLM assisted extraction of endpoints of interest from the study reports and how that can be used to differentiate class effects from novel findings. And I’ll end this section with some work that I’m doing with Hughes and predictive modeling and how that could be incorporated and utilized by the data lake.

And last question, is it safe?

Contextualizing risk via tox navigator, another R package. This one’s a… Has a R Shiny user interface, and it’s a a great tool for visualizing no adverse effect levels in context of exposure margins for different dose levels.

You can look across multiple studies at a time and see what are the key findings across studies and at what doses are we seeing these effects, and this can help interpret these findings with respect to clinical safety margins.

And the foundation on all of this, we’re kind of building all of this on top of the Send Explorer data lake. So I wanted to start with the motivation for, you know, why did we decide to build this Send Explorer data lake feature. And Send Explorer, as you know, provides high quality send data analysis and visualization capabilities for toxicology workflows.

However, the Send Explorer data lake enables a broader set of analytical applications beyond current SEMID Explorer functionality. These include kind of bleeding edge new approaches, like Bayesian analysis of treatment effects using historical control data, large scale clustering of tox studies for class profiling, application of predictive modeling, which could be used to automate characterization of target organs of toxicity, potentially develop Qsar models for general tox endpoints, or translate toxicology studies into human risk of adverse events.

We can also leverage Data Lake to work with the LLM supported integration of study report conclusions into SENData.

And the the idea here… Well, actually, I’ll I’ll get into the architecture, and then I’ll explain how you can utilize this. So currently, the architecture of Send Explorer is set up that you take your raw XPT files, you ETL, load them into the Send Explorer database, and the user interface kinda sits on top of that, and and you can run run Send Explorer, you know, on on the data in this this clean format.

What we’re doing new with the data lake is we’re taking all the study data from the database, and we’re pushing it into… Getting a copy or an archive of it in Parquet files in an s three bucket that you’ll have access to.

And this is really only applies to our really larger clients, but it is very useful for for larger clients. There there is a technical limitation on how many studies can be active in the SEN Explorer database, and it’s in, you know, five to 10,000. It’s it’s large.

But… So very very large clients like the FDA have kind of hit this limit, and this new data lake architecture facilitates the ability to keep only the studies that you’re using actively. And there’s kids… There’ll be thousands of studies, but you can limit the studies that you’re using in the the data mart, but keep copies of everything in the data lake, and you can easily archive and unarchive them because they’re stored in the same format as the database in these Parquet files.

But what’s really exciting about having this data lake is that this also opens up the opportunity to apply data science techniques and and analysis to all of your cleaned and and merged Parquet format send data. And so I’ll be talking about some some data processing packages that we’ve we’ve developed. Right now, they’re in R. We may expand to Python if that’s a need for clients, and how you can leverage these for custom advanced analytics and predictive modeling methods.

And then these can be built out into… For experimental use cases with with custom web apps and user interfaces to the Shiny application, which… Or package, which works for R and Python.

And as these use cases mature, we can then move them into the Send Explorer user interface. So this is kind of a interactive sandbox for for bleeding edge approaches where we can work with you to develop custom analysis. We can show you things that that we’re working on kinda before they get into the next release.

And and and, like… And then as those… The most successful of of these experimental approaches can then be later incorporated into official releases and integrated with all the other great functionality of the Send Explorer UI.

And so Sartara is developing some of these initial packages, which I’ll be talking about. And then we can work with you or potentially with the community through Fuse or taking work with with open source packages developed by Fuse or other toxicology open source data communities, and then kind of build those and adapt them to run off the data lake to to meet our clients’ needs and and then eventually get get these functionalities into the user interface.

And one last thing, this also allows us to integrate our data from Send Explorer with other Certara applications.

One example of this that we’re exploring is integrating with so that we could pull the predicted… For first in human studies, you know, for for IND enabling tox studies, we could pull the predicted PK exposure that we expect to see in humans into SENDEXPLORER from SimSip, and then represent doses as exposure margins against the predicted human exposure. We should also integrate with PHOENIX to get the actual exposures. And so that’s something that that Sartara’s been been putting a lot of effort into, and we’ll be excited to share that with you as that develops.

So… See. And now I’m gonna talk about from from Data Lake to… Or is it… Just kind of… This is the… A fun example of of a custom analysis that you could do with Data Lake just to give… Get Will’s training.

So this is just a a quick analysis that I I built up, just clustering the lab test results. I created a little dilly panel of of liver function tests and just looked at… These are just the studies that are in our kind of… In our test database. So these are…

And and you can… But you can see even with these these limited number of of test studies that the the effects of of these treatments on on these liver enzymes is is differentiated from… By the different treatments. And so you can see INDS one two three and drug b had more of a bilirubin effect, and they kind of cluster out over here. Whereas INDS zero two, I think this was compound b, tend to have more of an LFT increase effect. So these are the kinds of custom visualizations that that you can build or we can work with you to build.

So now I’m gonna get into these three questions starting with, is it real? Contextualizing findings with historical evidence. So this is talking about an R package that was originally developed while I was at FDA in collaboration with BioceleRate and FUSE. It’s called SendDigger, and it helps build a a SQLite database of of your send datasets and then has some querying functions for historical control analysis. So we’ve adapted this R package. We’ve renamed it send diver for the data lake, and now you can read from the Parquet files. No need to build your own database, and it will generate historical control reports.

And these can help you show you, you know, what is normal, get a sense of what are incidental findings, or what might be treatment related effects.

So there’s one function primarily that runs the whole analysis pipeline, and this filters your study so you can define, you know, what population of control animals you want to include in your historical report. You can filter by study design. You can grab the control animals, obviously, then you can get specific animals based on species, strain, sex.

You can filter for specific ages, and this will be the the age at which the finding was observed.

And then you can filter for only the test results that you’re interested in, And then it will run statistics to make these… This data consumable. It’ll give you the the number of animals, the the median mean, standard deviation, min, and quantile ranges, And these can be flexibly set by the application.

You can threshold so that you only report results for endpoints with a a certain number of animals. So you wanna have enough animals that’s a meaningful pool of animals to to report on.

It’ll aggregate per animal, and this will produce a HTML report or a PDF report.

And it also has a efficient caching functionality to it so that if you want to run multiple queries of the same population with slight slight tweaks, it will reload. It’s already queried halfway through this list, and you just change the test codes you’re interested in. It’ll just start from here and then run the rest of it, and so that makes acquiring more efficient.

This is an example HTML report. So you can see in this example, we’re looking at alkaline phosphatase and serum for Spragdolly rats aged eight weeks to 36.

And you can see that it splits it by sex. It also provides a report for combined sexes.

And you can use these quantile ranges to give reliable reference ranges.

And and then the animal counts are are shown so you know how many animals were used for for each of these analyses.

The future direction that we’re excited about taking this is to utilize this historical control information as a Bayesian prior to support analysis of treatment relatedness of effects. So a common problem, in large animal studies. You may have a finding in one out of three control animals and two out of three treated animals, and it may be ambiguous whether this is an actual finding or not. And something that we experienced toxicologists do in their head is that you think about how common is this finding. The intuition being if it’s a it’s a rare finding, then then maybe this this was a fluke that you got one in the control animal, and seeing two and treated is actually quite significant. If it’s a common finding, say, that this finding has a fifty percent background incidence, then you’re likely gonna see one in… One out of three in one group and two out of three in another group with no treatment effect involved.

And so Bayesian inference is kind of a way of formalizing that intuition into an empirical approach.

So the first thing you need to do is match historical controls to get the right population to to pull from for for your prior distribution, and then formulate that distribution into into a prior and estimate the the posterior probability that your treated groups are higher than the posterior population of your control groups.

And so these incidence counts can be calculated from from Sundiver, and these are just some kind of visualizations of what this would look like with probability distributions. So the dotted line is your prior distribution. This would be your posterior for control and for treated.

This would be in the case where the the background finding is quite rare. And so you can see that there’s a… There’s not so much overlap between the controlled posterior and the treated posterior, and and and this backs up that intuition that, you know, this is likely… More likely to be treatment related if the the background rate is quite low. But we can actually put some numbers to this, and then this can be helpful for applying at scale. You you might not have time to do all this intuition in your head if you wanted to run this across multiple studies, across multiple findings. This gives you a method of in… Of performing that intuitive task of evaluating a finding in context of historical information, but you can run that at scale.

Then this is just showing the same information when the background finding would be 45% incidents. And you can see here that there’s a lot of overlap between these distributions, and the the calculated probability that treatment related rule would be about 76%, which you’d probably wanna set a higher threshold if you’re doing some automated approach to weed out what are the true treatment related effects versus ambivalent or unlikely true results.

There is actually a… This was developed by one of my postdocs at FDA, Youssef Ali. I think he’s he’s still at FDA.

And he he developed a a proof of concept, our Shiny web application that will implement this… These calculations for you. And so right now, you’d be able to calculate your historical reference ranges from from the data lake and and from the Sundiver package, and then you could you could… You know, if you have a experienced team, you could kinda feed that into this application. You can strip out the calculation from here and implement it on your own. But this is also something that we’re planning to do, you know, as as we take time to develop these these functionalities for our clients.

A direction we’re we’re thinking is to set up this connection for you to be able to feed background information from Sendiver into this Bayesian approach.

The next question that we wanted to answer is, is it relevant? So interpreting, you know, which endpoints matter?

You know, SEND data is very dense. It has lots of detailed information, and it can be difficult to distill, you know, what are the signals from the noise and what are the taxicologically relevant signals for the statistically significant signals. And that’s where the endpoints of interest workflow that we’ve built into Send Explorer over the past couple of releases, really, can be can be leveraged to to help you in… Interpret the… And keep track of the… Which findings are relevant, and then use that information to understand class profiles and and such.

And so as as Jamie showed, we’ve implemented auto flagging of potential endpoints of interest. The scientists can then review, ignore, add notes, and… Or manually select their own endpoints that that, for whatever reason, were not detected by our statistical analysis.

And then we can generate these narrative summaries, which can be fed into… Potentially into your module two summaries.

And as Jamie showed, this this process, the auto flagging will give you a list of potential endpoints of interest. The… And feature direction I’d note, especially for clients who have a large amount of data, it should be possible to leverage this Bayesian inference scheme into the statistical detection of these so we can do a bit more more nuanced detection and a bit…

Improve the identification of potential treatment related effects based on the historical control context, which will just improve the fidelity of that work.

From there, once we’ve established what are the endpoints we’re interested in, we can generate these text summaries.

Any… You can add notes into the prompt, and those will be incorporated. For example, you can add notes about animals that that died early on the study and and what happened there, things like that.

And this text is then editable, and you can use it for, like I said, for module two summaries or based on on different prompt summarizations. We’re working on down the road being able to summarize… Compare and contrast across multiple studies. Now that you’ve identified what are the important key findings in multiple studies, we can generate narrative summaries to explain what’s similar and what’s different.

But something that we’re we’re excited about as a a future direction would be also to use large language models in AI to extract endpoints of interest from the study report document.

And the idea here is we could run… We could kind of automate automate the process of of reviewing endpoints of interest by extracting endpoints of interest from the study report, extracting endpoints of interest statistically from the data, and taking the intersection of these, because endpoints that are found in both cases are going to be likely to be be truly significant and interesting endpoints, and this allows false positives from either method to kinda fall out on the sides.

This probably won’t have the the same fidelity as manual review, but what’s nice about it is this approach, you’d be able to scale this and run this across all of your studies and get a sense of where are the important endpoints of interest across across all of the studies for which you have SEND data and study reports.

And then this information could enable you to do… To understand the the toxicological profiles associated with a particular pharmacological target or or therapeutic modality or chemical class based on all the information you have. And… But… And pulling in the endpoints of interest, it gives you an extra filtering, you know, and especially pulling in from the narrative summary, you’re really capturing that expert knowledge in a way that just pulling in the raw data, you know, you may find that there’s there’s similar… Similarities among the raw data in in different studies, but here, we’re focusing just on the endpoints that we know matter and signals that are really important.

And then if we have this information, when we we run a new study within a class or therapeutic modality, we can differentiate. Are the findings in this study, do they match with the profile that we’ve seen from… For similar compounds, or are we seeing novel toxicity? Because that will drive differences in interpretation and decision making of how we approach moving this compound forward. If it’s on target or class effects, we likely have a sense of the clinical relevance of these findings. We may have clinical experience of how do these findings translate, and we know kind of how to move forward with them, whereas if it’s truly a novel finding, we probably need to be more cautious and really take it seriously, try to resolve, is this finding real?

And if it is real, be a bit more careful and and and cautious as we move to humans.

And this is just saying, yeah, we can derisk class effects. We can flag novel toxicity.

Now I’d like to talk about a a FUSE project that we’re working on… That I’ve been working on for for several years, developing predictive modeling approaches using SENData.

And I’m bringing this up because the outputs of this project could be useful and leveraged by SendExplorer data like clients.

So the the initial phase of this project was kind of a show and tell phase where you wanted to share what methods, what are different different organizations doing with their SEND data and predictive modeling with their toxicology data. We kind of wrapped that up after a year and a half or two years of of of show and tell sharing, and we started this working phase where we’re trying to… We kind of looked at what are the what are the topics and methods that were presented during that show and tell phase, and we picked two methods and are developing… Initially, two methods and are developing best practices for how to implement these, and we’ll kind of work our way through all the different kind of steps in that predictive modeling project process and and develop best practices with the goal that at the end, we can develop these kind of end to end workflows to to solve real problems and and leverage predictive modeling in useful useful ways.

And as these community vetted models are developed, we can then operationalize them within the SENEXT Explorer data lake.

So the… Kind of the motivating example for this project was from a pilot project I did while I was at FDA, where at at FDA, we had… Reviewers were able to flag their target organs as they reviewed studies, and we built a a machine learning random forest based model to try to retrain the model on studies that were labeled with, for example, liver as a target organ, and we wanted to see if the model could learn what were the endpoints in the in the raw data that were associated with the reviewer flagging liver as a target organ. And once that model’s trained, then as new data came in, we could run that model and make predictions about, you know, how likely is a reviewer to flag liver as a target. And it kinda gives you almost like Netflix suggests these are shows you might be interested in. The idea here was to build a model that could could predict these are organs you might be interested in based on the raw data.

And so sub team one in the the FUSE project were really kind of hammering out this workflow and and trying to to mature it from proof of concept into something operational. The first step is to make sure that we’re we’re normalizing the data. We’re taking this this granular raw send data and converting it into a a form that can be interpreted by a machine learning model.

And the goal here is, yeah, extraction of safety signals from send format data to support predictive modeling. This is kind of normalization, z scoring for continuous data, scoring for kind of severity incidents, scoring for for categorical data. There’s nuances here, like categorical versus numeric, time series data, dose response, and statistical… We’re also exploring statistical filtering of of data, which which could involve traditional frequentist approaches or or Bayesian approaches.

There’s a… In parallel, we have another sub team working on kind of the other end of this, of the the target organ… Labeling end, and the approach they’re working on is to making study report conclusions available for data mining. And there is a paper from Genentech. So Leonard Anger from Genentech is leading this subteam. They had a great paper from Genentech recently describing a a workflow to to implement the extraction of kinda key findings or, yeah, the… Whether the the study director or the the narrative say were the key findings from the study report.

This is all open source available on GitHub, and we’re we’re working to implement this workflow into… To run of… On on… Potentially, on our clients’ PDFs, and then integrate that information into the the data lake in in the Send Explorer.

A really high high powered, a bit future, but ambitious direction that we’d really love to to take this would be to apply this predictive modeling approach to support weight of evidence assessments for carcinogenicity study waivers.

So the latest guidance… ICH guidance for carcinogenicity ICH s one b has… It gives a kind of a carve out that if if you think that your compound is very unlikely to cause carcinogenic… Have carcinogenic potential, you can write up a a weight of evidence argument based on these criteria over here to to argue that, you know, we don’t need to run the two year rat study. We can do the six month mouse transgenic study and and be done with it.

And the process for this is is still…

I I feel like it’s still kind of being worked out. It’s a bit hand wavy, you know, coming up with your your arguments, and it’s a bit hand wavy on the regulator side in terms of what are they gonna accept. And that creates a lot of uncertainty because if you you submit an argument to a regulator and, you know, you’re waiting six months perhaps for them to get back to you with feedback, if they disagree with you, well, now you’re six months behind and you still have to connect that to your rat study. And so we think through…

By applying predictive statistical modeling, we we may be able to add some empiricism and some certainty into this approach, because you’ll note that some of the criteria that this weight of evidence is based on is our endpoints that are present within SEND datasets, and we have SEND datasets for our chronic studies. We also have SEND datasets for our CAR C studies. And so it should be possible to to build predictive models that that truly understand empirically what is the relationship between certain histopath findings and incidence of tumorigenicity in in CAR C studies.

And, you know, as initially, we’d like to build this around SEND data, but as…

If that’s successful, we can then expand that to include other data such as secondary pharmacology or target biology.

Sotara has tools for analyzing secondary pharmacology. There’s a a data standard being developed by the Pistoy Alliance for in vitro secondary pharmacology data, which would support these kinds of workflows in the future. And then even for for target biology, QSP mechanistic modeling approaches may also be incorporated into into this kind of modeling support for weight of evidence.

The last question, is it safe? Connecting tox studies or toxicity findings to clinical exposure. So here I’m talking about an R package called ToxNavigator. This was something we derive from an R package developed at FDA, a tox summary.

So you may be familiar, we’ve presented this both publicly through FUSE and I think at the last BGM.

So this this tox summary package… Now it’s the the tox navigator package with connectivity to the data lake allows you to plot exposure margins relative to clinical exposure for your various doses so you can compare, you know, across multiple studies. So here’s rat study. Here’s a dog study. These are the the relative exposure margins associated with the different dose levels, And then you can also note your key findings and the the kind of subjective severity of those to to get a sense of how those findings are are potentially getting worse or or more severe at higher doses.

And this, yeah, this allows you to to visualize multiple studies in a single plot.

We’ve also developed a new hermatogram visualization, which is the same information but plotted. Now we have the actual exposure on the on the… Kinda log scaled on the x axis, and we have these dotted lines representing the human start dose, anticipated or or real exposure we saw at the human start dose, and and maybe the the maximum dose that we plan to dose in phase one. And then we can provide for each dose the actual… The the range, what was the lowest AUC we saw in animals at that dose level was the highest, and you… So you can see kind of the full range of what exposure did we see in animals and how does that… And then what findings did we see at those exposure levels and how do those compare to the the clinical disc.

We can also generate this information in tables where we can describe the the exposure margins of different findings, you know, across and which studies did we observe these findings in. This can be exported to Word, Excel, CSV. So the tax navigator package, it… We’ve adapted this to be able to pull most of this information from the data lake, you can pull the the nonclinical dose levels, the TK information. Even in terms of selecting your key findings, it’ll pull in the endpoints of interest that you’ve selected as kind of a starting point, and you can select from from that drop down which endpoints you wanna include in this analysis.

Then the scientists can select the findings that they think are truly important for this analysis.

You can indicate the severity or adversity at each dose level and the no adverse effect level. And also supply the clinical doses and PK, although, as I said earlier, we’re looking to be able to pull this information directly from SimChip or Phoenix for clients who have those tools as well. And this will output these visualizations of findings relative to clinical exposure as well as tabular representations of the exposure margins at different findings or even to have study level safety margins.

So… And, yeah, the endpoints of interest from the AI workflow can populate that finding list.

So Sorry.

I just wanted to do a quick time check.

Yep. I’m wrapping it up. Alright.

So we covered, you know, is it real? So we can use historical information to… And and Bayesian inference in the future to really parse out, you know, what are the real signals in our data. Is it relevant? We can use our expert knowledge to kind of label endpoints of interest along with maybe statistical flagging to to bubble up things we we need to look at, but we can store that that that human knowledge as endpoints of interest or extract it from the narrative study reports and then feed that into predictive models.

And is it safe? We can then compare…

You know, we see all these findings in tox studies, but how do they how do they… What are the margins between the exposure at which we saw these findings in animals and the exposure at which we see that we were going to expose humans, and how much margin do we have of of safety?

And the foundation of all of this, you know, the SENDEXPLORER data, like, can can support answering all of these questions.

So last, I just wanted to end with how we can we can go from these tox profiles to actually inform better development decisions. Right? We can we can use the the historical control Bayesian inference to actually derisk irrelevant findings.

We can characterize class effects and, you know, distinguish, you know, what is a novel tox from a class effect, which we could potentially derisk based on our clinical experience. We can contextualize liabilities across across programs, and and, last, ideally, we’ll be able to to waive unnecessary studies through kind of empirical modeling support for weighted evidence arguments.

And with that, I’m gonna turn it over to Melissa to handle our next polling question.

And feel free to put your questions into the question… Not the chat, but the question q and a part of the GoToMeeting.

K. Thank you, Kevin. Great presentation. I’m really excited to see the direction we can take utilizing the SUNDEXPLORER data.

Okay. If you wanna just take a moment to fill out the next poll question once it pops up.

It’s gonna pop up.

Okay. Maybe it’s not going to work.

If you wanna put into the chat if there’s one of these features that maybe you’re not utilizing so much within, Send Explorer, that would be… Okay. Here it goes.

If you can put this in, it is really helpful as we start working on future directions in the Send Explorer application, whether it’s deprecating a feature or actually adding more resources to simplify or add new features. So this is really helpful. K.

Alright. We’ll just give it one more second to give people time. K. So this is really helpful. I appreciate everybody taking a moment to fill that out.

Okay, so we’re gonna talk a little bit about the roadmap. So Jamie gave you a very nice overview of some of the features that are gonna be in version 13.

And I really wanna talk about this because it might impact the direction you go with upgrading your Send Explorer.

So in 13, we have the automated endpoints of interest, AI summarization, the multiple endpoints in the scatterplot time course, simplified navigation, dashboard filtering within the visualization, and then there are some bug fixes and other enhancements. So this is just very high level. This is going to be available after October 30.

However, you may need to wait for version 13.1.

And the biggest impact is really the embryo fetal development summarization.

So for any of those domains that are incorporated into the EFD data, you’ll not be able to do the group summarization in version 13. So the underlying data will still exist in SENDEXPLORER via the data grids.

However, we are putting this back into 13.1 and revamping it a little bit.

We do plan to bring the study schedule back, and we also want to enable multiple endpoint of interest AI prompts. So right now what you saw Jamie demonstrate allows one prompt to be put into the system. A user can overwrite that, of course, but we wanna enable multiple prompts.

And also those fold controls that were added via the comments that are fed into the AI summarization, we wanna automate that so an end user doesn’t need to manually enter those. So just at a very high level, this is what we’re planning for 13, and this is going to be due out later this winter.

But as we look ahead to version 14 and beyond, we have some exciting new features that we’re planning to introduce. So over on the left hand side, this is where we are. So for the most part, 13.1 is locked in. But I do hope that we can bring in Tax Navigator. So Kevin showed that it will be available via the data lake in version 13, but I would like to put that into the interface. There’s a lot of excitement around that.

We would also like to enable doing analysis on multi studies using the endpoints of interest. Again, now, it’s just for a single study.

Samantha gave an overview of SUN 4.0, so you, I think, can understand there are going to be huge differences. So we need to start working on the initial planning for a new data model, as well as perhaps a different way to visualize the study data. So this is going to be a very big effort that will keep us busy. And then finally, Sattara is very dedicated to integrating our framework together. And Kevin talked a little bit about some of the exciting plans we have in store as well.

So along the way, we do plan to do bug fixes and new user requests, which would need to be prioritized.

So I’m estimating that we will be back on schedule to have new versions yearly, and I’m expecting this to be at the end of two thousand twenty seven.

But you might be asking yourself, how can you help shape the future of SendExplorer? So moving forward, I would love to have a little more interaction with all of our users, and I’d like to start getting more input. And the input could be on a variety of different topics that you feel strongly about. It might be Send Explorer’s intimidating.

Can we help simplify the interface, which we’ve worked really hard at doing, but maybe there’s something else we can do. Might be enhancing what you just saw with Talks Navigator and points of interest enhancements. It could be anything. There might be a missing visualization that you feel strongly about.

And so I do feel like this can be very low effort, so we’d love help with brainstorming sessions.

Once we’ve done those, it would just be reviewing the requirements, making sure we’ve captured those appropriately, and then the ability to look at early prototypes with us and provide feedback.

So if this is something that sounds interesting and like something you would really like to participate in, if you could fill out the survey at the end and maybe just note what you’re interested in, that would give us the ability to reach out to you.

But switching gears a bit. So I wanna talk to you about how we’re going to prioritize the upgrades.

So as of 10/30/2013, NextGen is gonna be available along with the release notes. The release notes will go into great detail about exactly every enhancement that is going to be available.

And this is going to be for new customers and existing.

And we would be happy to provide those to you. But our plan is to really start doing the upgrades in WAVs, starting from the simplest to the most complex.

So simplest are our SoundExplorer Plus customers.

So these customers do not have any integrations. They typically will not perform user acceptance tests on their test environments. There’s a smaller user base typically, and so change management is just gonna be simpler. It’s just a couple of people that need to show up for training and act on things.

So wave two are what I consider more moderate complexity. This is where we start introducing our SendExplorer warehouse customers.

Study counts could be from two studies all the way up to eight fifty studies. You will not have a development environment that we need to worry about, and there’s no integrations with additional platforms.

And then finally, we get to wave three, which is the most complex.

So this is going to be our warehouse, or customers who are also planning to take Data Lake as well. You might have a lot of customizations or integrations along the way that we’ve done for you.

The user base might be a lot larger. You might have users scattered throughout the world that we need to take into consideration. Your study counts are typically much higher, over eight fifty studies. You also may be somebody who does user acceptance testing across multiple environments. It’s not a given, but you might.

And you may also need to do additional planning. So we’ve tried talking to customers over the past year, but this may require a project or bringing in additional people to do these upgrades.

And then finally, you may be doing EFD studies and feel very strongly that you want to wait for version 13.1. So I’m not capturing every single scenario here, but hopefully you get an idea of where your company falls in terms of the different waves.

But during the upgrade, I wanted to just give a very high level understanding of like what you can anticipate. And this is really for wave one and two customers.

So we will reach out to you and there will be agreement on an upgrade plan. And we’re not just going to perform these because there are very big changes to the application.

So you’ll approve the dates. We’ll confirm the training dates with you as well. The day of the upgrade, Send Explorer won’t be available for between two to eight hours. This is when we do backups and upgrade.

We restart the system. The Satara team will do quality checks and make sure test and production are back up and working. You’ll receive a notification, you’ll get our documentation, and then the site is gonna be available to use. But after the upgrade, we will do training with you to cover the new navigation and endpoints of interest workflow.

And for Plus customers, the data upload and ingestion is moving, so there will be brief training to cover that.

And then as usual, ad hoc training will always be available as needed to make sure you’re successful.

For wave three though, this is gonna look pretty similar. However, the big thing that we need to do is to coordinate each environment for upgrade one step at a time, really going at your pace more.

So we will still be contacting you, getting approval for upgrades. We will start targeting training dates. The downtime is going to look very similar. Our goal, if you have a dev environment, would to be try to do dev and test at the same time.

We’ll still give you the quality checks, and then a decision will be made. Will you upgrade to production or not? So hopefully the answer is yes, And then we’ll follow the same steps as dev and test. And then just like with wave one and two, we will be providing training.

This might look a little bit different though. So if you have a very large user base, there may need to be multiple sessions.

So we’re dedicated to making sure all of your end users are trained and ad hoc training will also be available to you.

So just a few words about the training. So our goal is to perform this virtually just like we typically do. We will review the new navigation and points of interest workflow and any of the other new wonderful features that we have shown today.

The Plus customers, again, they will have a brief training on how to upload their studies. We’re estimating that the entire training will take between sixty and ninety minutes.

The sessions will be recorded and shared with you, and the training materials will also be provided. So just as a helpful reminder, people forget we have a help center in SendExplorer, even in version 12, But we are updating all of our training materials and Quick Sheets and knowledge base articles. So those will also be available in version 13 as well.

I know this is a whirlwind to think about. But anyhow, hopefully this gives you a high level understanding of what to anticipate as we start contacting each of you.

So before we get to q and a, I would just like to ask one more polling question, because hopefully you’re still awake. And that’s really, are there specific enhancements that you would like to see in SendExplorer in the future? So if you have any ideas of what you would like to see, please feel free to actually use the chat feature. And I will be reviewing those at the conclusion of this meeting today. So feel free to pop those in.

Okay.

So we’ve set aside a little time to go over q and a. And I was just checking out the question section. And right now, I don’t see any questions.

So that either means the information was very straightforward today, or maybe you just haven’t had time to formulate some questions.

Okay. It looks like…

Okay. This is an enhancement. So more customizable options for plotting data. Okay.

So can definitely look into that.

Okay. I’ll just give a second in case people are typing their questions. And if not, we can conclude today’s event.

Okay. So I’m not seeing anything coming through. So just keep going so we can end a little bit on the earlier side.

So we would love for you to each continue to interact with us. So I have on this next slide just a couple of our in person conferences that we will be attending. So Jamie, Kevin, and myself will be at ACT at Booth 719 in San Antonio in November. Kevin is going to give a couple of talks on the sixteenth.

I will give a talk with a Sitara toxicologist on the seventeenth. So if you’re attending this meeting, we would love for you to stop by just to say hello.

We will somehow be going back to San Antonio in March to SOT. So again, love to say hello to each of you if you’re planning to attend. Kevin has a couple virtual events that are planned on October 28 and December 3. So if you’d like more information, we’d be happy to share those details with you.

And then finally, I just wanted to wrap up the twenty twenty six SoundExplorer User Group meeting, and say a very special thank you to each of you, our customers, for making Sitara SoundExplorer your trusted partner in preclinical visualization and analysis.

We really do look forward to working with you in the upcoming year and hearing your feedback about our latest version.

I did wanna give a special thank you to the SendExplorer development team who are not here with us today.

So Paul, Joe, Paul Lewis, Dan, Ben, and Arthur, thank you for everything that you’ve done to make today and October 30 available. And the marketing team, Brian, Marcella, Erica, and Jessica for making this event possible today.

So if you could just take a moment though to complete our survey at the end as you’re logging off, that would be very helpful. We will mail you a link to the recording. And again, please do stay in touch with us. We love to hear from each of you. And that’s a wrap for this year. Thank you again for your attention, and I hope today was very informative for you. Thanks.