Upon the completion of the Human Genome Project, the lead investigator, Dr. Francis Collins remarked:
Science is a voyage of exploration into the unknown. We are here today to celebrate a milestone along a truly unprecedented voyage, this one into ourselves. Alexander Pope wrote, “Know then thyself. Presume not God to scan. The proper study of mankind is man.” What more powerful form of study of mankind could there be than to read our own instruction book?
The critical technique used in the Human Genome project—DNA sequencing—is a disruptive technology. The ability to sequence an individual person’s genome is likely to substantially impact how we use medications to treat patients.
Untangling the genotype-phenotype relationship
How does the behavior of cells, tissues, organs, and organisms emerge from interactions between the genome and environment? Currently, this question is mostly addressed by looking for statistical associations between the full genome sequence and phenotypic traits. Genome-wide association studies (GWAS) can lead to discovering genetic loci associated with different phenotypes including increased disease susceptibility. While impressive and certainly valuable, GWAS does not answer how a patient’s genetic polymorphisms contribute to increased disease susceptibility. Without knowing the underlying mechanisms of these associations, it’s difficult to use this information to design therapeutic interventions. Also, we are frequently interested in explaining genome-environment-phenotype interactions involving factors such as exposure to drugs and/or toxins, diet, or exercise. But, studying these more complex interactions using an approach based solely on statistical association is challenging.
Mechanistic simulation of the genotype-phenotype relationship
Mechanistic modeling is an alternative to statistical approaches that can yield greater understanding and predictive power. I’ve spent most of my academic career performing mechanistic modeling of the genotype-phenotype relationship.
Fortunately, we know a lot about the molecular biology of the cell. PubMed contains millions of articles describing individual interactions between molecular components of the cell. How can we represent this knowledge as a computer model capable of simulating the dynamics of a cell’s molecular components? The molecular machinery of the cell knows how to express the genome in the context of a particular environment. If we could reverse engineer this machinery in the form of a computer model, we could use it to predict the phenotype arising from the interaction of the environment and a genotype. For a given genetic polymorphism, the model would simulate the dynamic response to environmental conditions.
The rise of PBPK
Modeling the entire molecular biology of cells is a daunting task. So how can we create large-scale mechanistic models with predictive power sufficient for drug developers? The acceptance of physiologically-based pharmacokinetic (PBPK) modeling by industry, academia, and regulators gives me reason to be optimistic.
PBPK is a mechanistic approach for describing the dynamics of drug absorption, distribution, metabolism and elimination in physiological compartments. The variables of the model are the drug concentrations in physiological compartments. The dynamics of these variables are modeled by the system of ordinary differential equations using compartment volumes, blood flows and partition coefficients as parameters. These parameters are based on the human physiology literature and in vitro assays rather than estimated for each study. The model is then used to simulate the dynamics of the drug’s concentration at the site of action.
The whole-body PBPK models used today contain “models within models.” The physiological compartments are subdivided into smaller compartments and new, intra-organ flows are defined. Certara’s Simcyp Simulator contains mechanistic models of the gut, lung, kidney, brain and liver. The general method of building this large-scale model is the same as building models of intracellular networks: literature knowledge and in vitro assay data are represented by a computational model.
Use of large-scale, mechanistic models by industry, regulators
Performing simulations using PBPK models let us use this knowledge to predict the behavior of the system. In fact, predictions based on PBPK simulations have been accepted by regulators as the sole evidence for assessing certain types of drug-drug interactions (DDIs). Thus, we can bypass performing clinical trials to quantify these DDIs.
And this example isn’t an isolated case. Around 100 label claims have now been informed by PBPK simulations. To me, this suggests that it’s possible to build large-scale, literature-based mechanistic models with predictive power sufficient for the most stringent application: regulatory submission.
Extending mechanistic modeling to account for all human genes
The whole-body PBPK model of the Simcyp Simulator accounts for about 20 genes encoding drug metabolism enzymes and transporters. Genetic polymorphisms in the population of interest can be input into the software and used to simulate the PK variability expected in particular trial. However, there is huge gap between this mechanistic model and incorporating the full genetic code of the patient― around 30,000 genes. How can we extend mechanistic models to account for all genes and interpret ~omics data (genomics, metabolomics, proteomics, etc.) by mechanistic simulation rather than statistical association?
One answer involves building mechanistic models of molecular networks operating in intracellular space. Genome scale metabolic networks (GSMNs) are models that account for thousands of genes. To learn how GSMNs can be used to analyze ~omics data, please watch a webinar I gave on the topic. Let me know what you think in the comments section!