My notes on agentic AI, drug discovery, and teaching
-
Virtual cells for target discovery, perturbation models, and benchmarks
Virtual cell models try to predict how cells respond to genetic and chemical perturbations. A practitioner's guide to what they can do for target discovery, what the benchmarks say, and where the current limits still matter.
-
Synthetic lethality and combination targets: ML methods for finding drug pairs that work together
Synthetic lethality turns a drug-target problem into a target-pair problem. Which ML methods find the pairs? CRISPR screens, DAISY, SynLethDB, and GNN link prediction.
-
Chemical and Biological Data Repositories for AI Drug Discovery
A maintained catalog of publicly accessible chemical and biological data repositories for machine learning in drug discovery, with notes on the data-quality properties to verify before training a model.
-
Knowledge graphs, network medicine, and the first end-to-end AI-discovered drug: a target-discovery case study
Rentosertib is the first end-to-end AI-discovered drug to show a Phase 2 efficacy signal. This post examines the knowledge graph that identified TNIK and the program that followed.
-
Likelihood of approval, phase transitions, and the crowded-vs-abandoned map of therapeutic areas
Phase I likelihood of approval is 6.7% industry-wide. Oncology, rare disease, and biomarker-stratified programs tell different stories. A 2026 map.
