Reducing Drug Discovery and Development Risk with Mechanistic Modeling

What It Covers

John Burke, cofounder and CEO of Applied BioMath, explains how mechanistic modeling replaces traditional allometric scaling in drug development. Using differential equations to simulate biological systems from in vitro through human dose prediction, the approach reduces late-stage attrition, accelerates IND submissions, and helps companies prioritize which drug candidates to advance.

Key Questions Answered

• •Early-stage candidate triage: Apply mechanistic modeling at project inception to rank-order drug concepts before committing resources. When evaluating 10 candidates with funding for only 2–3, modeling can identify which require the fewest experiments, have the most developable biophysical properties, and carry the lowest uncertainty in human dose predictions — potentially freeing budget for an additional program.
• •Sensitivity analysis as experiment prioritization: Mechanistic models use nonlinear differential equations where some parameters, when varied across thousands of simulations, produce negligible output changes, while others cause drastic shifts. Identifying these high-sensitivity parameters — such as protein synthesis rate constants — directs teams toward the specific experiments that resolve go/no-go decisions rather than generating broad datasets.
• •Justified higher starting doses for Phase 1: Traditional allometric scaling applies conservative safety factors that often produce homeopathic starting doses. Mechanistic models incorporating target expression levels, cell numbers, and binding kinetics can demonstrate to the FDA a more precise safety window, supporting a higher, pharmacologically relevant starting dose and compressing dose-escalation timelines.
• •Translational fidelity through layered model building: Build mechanistic models sequentially — in vitro, then mouse, then nonhuman primate — recapitulating each dataset using zero, first, and second-order reactions with minimal parameter fitting to avoid overfitting. This single connected model carries mechanistic assumptions across species, reducing translational uncertainty more reliably than dotted-line allometric extrapolation between species body weights.
• •Post-IND mechanistic PopPK for Phase 1b/2 design: After first-in-human data is available, integrate real clinical measurements back into the mechanistic model rather than switching immediately to standard population PK. This hybrid approach characterizes parameter distributions based on mechanism before patient numbers are large enough for purely statistical PopPK, enabling more precise recommended Phase 2 dose selection and trial design.