The third CACHE challenge results have been published in the Journal of Chemical Information and Modeling. I contributed as one of the 23 participating teams.

This challenge targeted macrodomain 1 (Mac1) of SARS-CoV-2 Nsp3 — a promising antiviral drug target. Teams collectively proposed 1739 ligand candidates using a broad range of design strategies. Over 85% of the designed molecules were chemically novel, yet the best experimentally confirmed hits were structurally similar to previously reported compounds, largely preserving the adenine-mimicking chemotype seen in known Nsp3-Mac1 ligands.

The winning workflows again reinforced patterns seen in CACHE #1 and #2: two of the top-performing teams used physics-based screening to train machine-learning models for rapid large-library screening, while four others relied exclusively on physics-based approaches. Pharmacophore searches and fragment growing strategies also featured among the successful workflows.

Although the confirmed actives are structurally conservative, they provide new structure–activity relationship insights that can inform future antiviral development. The broader message from three consecutive CACHE challenges is consistent: multiple, methodologically distinct design strategies can efficiently converge on similar potent molecules, and hybrid physics/ML pipelines are a reliable component of high-performing workflows.