Evan Feinberg

Evan Feinberg is a molecular machine-learning researcher and co-founder and CEO of Genesis Molecular AI. His contributions include collaborative research on PotentialNet, a family of graph-convolution models for predicting molecular properties, and public arguments for stricter accuracy standards in protein–ligand structure prediction.[2][3][1]ⁿ¹

In a July 1, 2026 Latent Space interview with Genesis CTO Sergey Edunov, Feinberg described diffusion as an enabling development: “We sort of had to wait for the right primitive to get created, and that turned out to be diffusion…” Genesis’s PEARL model places a ligand while also adjusting the protein around it, addressing the flexibility involved when the two bind. Feinberg called 3D structure prediction a site of particularly innovative diffusion research.[1]

Feinberg challenged the use of 2 angstrom RMSD as a benchmark threshold for a good predicted pose. “If your model is sitting at 1.8, 1.9 Angstrom RMSD, that’s slop, most likely,” he said, arguing instead for a one-angstrom threshold to capture the molecule’s core placement and interactions. The interview article illustrates the concern with an aromatic ring: reversing its orientation can leave a prediction looking plausible while changing the interactions it represents. His proposed threshold expresses a judgment about medicinal-chemistry usefulness, rather than a universal accuracy rule.[1]

Feinberg and Edunov also argued that Genesis’s predictive models had become accurate enough to support agentic drug-discovery loops. Their internal system, SAPPHIRE, was described as reasoning about molecular poses, forming hypotheses, reading literature, using internal tools and proposing candidates for the next iteration. Their argument linked the usefulness of these workflows to the reliability of the molecular predictions on which successive decisions depend.[1]

Before this work on diffusion and agents, Feinberg studied applied physics at Yale and earned a Ph.D. in computational biophysics at Stanford. In Vijay Pande’s Stanford lab, he conducted AI and simulation research that included PotentialNet. He founded Genesis in 2019 to further develop and apply that research, connecting the company’s origins to his earlier work on molecular prediction.[2]

The PotentialNet paper, first submitted to arXiv on March 12, 2018 and revised on October 22, presented graph convolutions designed for protein–ligand binding-affinity prediction. Feinberg co-authored it with Debnil Sur, Zhenqin Wu, Brooke E. Husic, Huanghao Mai, Yang Li, Saisai Sun, Jianyi Yang, Bharath Ramsundar and Vijay S. Pande. The team also introduced a Regression Enrichment Factor to measure early enrichment in chemical predictions and a cross-validation strategy based on structural-homology clustering to assess generalizability.[3]

In a follow-up paper submitted to arXiv on March 28, 2019, Feinberg, Robert Sheridan, Elizabeth Joshi, Vijay S. Pande and Alan C. Cheng applied PotentialNet deep featurization to ADMET property prediction. Their approach learned features relevant to each chemical task from molecular graphs, with atoms as nodes and bonds as edges, rather than relying on fixed molecular fingerprints. Using cross-validation and prospective analyses, the authors reported improved ability to extrapolate into new regions of chemical space.[4]