ABSTRACT – MECHANISTIC PHARMACODYNAMIC MODELING FOR DRUG COMBINATIONS
Most industries simulate design options before implementation, but this is rarely possible in the pharmaceutical
and medical industries. An important gap is unbiased drug combination response predictions, which is
experimentally impractical. A long-term vision of our lab is improving drug development and precision medicine
by building “mechanistic pharmacodynamic models” that can simulate drug combination responses. Such
models infuse pharmacology concepts with physics and engineering approaches to describe causal, quantitative,
and dynamic mechanisms underlying drug response. A foundational premise is that capturing (i) mechanistic,
causal network structure, (ii) dose-response, (iii) dynamics, and (iv) cell-cell variability is necessary to improve
many combination response predictions. Here, we study how drug combinations affect single-cell
proliferation and death fates by merging theoretical and experimental innovation. The first project builds
on our recent and one of the most comprehensive mechanistic models for regulation of single-cell proliferation
and death dynamics. We will leverage our involvement with a recent LINCS consortium effort that generated a
deep molecular characterization of perturbation response dynamics, including dose responses to 8 drugs. We
will integrate network biology with mechanistic models using new approaches to obtain candidate models that
are consistent with this dataset, and experimentally test drug combination predictions for the 8 drugs. This will
for the first time address the prediction of a comprehensive set of drug combination responses across varied
mechanisms of action relying on causal biochemical reasoning and also identify novel mechanisms of signaling
and drug response through iterative model refinement and experimental validation. The second project builds
on our recently developed experimental approach for fluorescence multiplexing called MuSIC. We propose that
MuSIC can enable high-dimensional genetic interaction screening in single mammalian cells, which is not yet
possible but would be transformative. We will test the approach by evaluating genetic interactions between a
recently curated set of 667 gene targets of 1,578 FDA-approved drugs. This work will nominate new network
structures not only for use in the first project, but also more generally. The third project also leverages the
above mechanistic model but pivots across cell lines with Cancer Cell Line Encyclopedia data for 1,132 cell lines
and 24 drugs. An innovative and foundational feature of our model is that it ingests multi-omic data to create a
cell line-specific context through “initialization”. We will generate 1,132 model variants with cell line-specific
profiles and evaluate predictive capacity for single and prioritized drug combination responses. This project will
establish performance of the current models, identify critical modeling gaps for improving predictions, suggest
new potentially effective drug combinations, and elucidate mechanisms underlying synergy. Overall, these
projects will produce next-generation pharmacodynamic models that move towards filling the drug combination
prediction gap that hinders drug development and precision medicine.