One drug-design model handles merging, dual-targeting, and selectivity without retraining.

Medicines work by fitting into biological targets. To do this, small-molecule drugs rely on three-dimensional interactions, like how shapes match and electrical charges pull together. Researchers wanted a universal way to design these molecules based on those interactions. They created ShEPhERD–2, a computer model that generates 3D molecular structures. It works by taking an interaction profile—a blueprint containing target shape, electrostatic features, and specific interaction points—and building molecules that match it. The team found that ShEPhERD–2 designs low-strain, drug-like molecules faster and more accurately than the older model. Users can prioritize key contact points, lock in specific molecular pieces, and combine multiple profiles. Without needing extra training, this single tool succeeded at tasks like dual-target design and modality hopping.

Why it matters

Drug discovery is slow and costly because each problem usually needs a customized tool. ShEPhERD-2 shows that a single model, driven by a 3D interaction profile, can perform several design tasks without retraining. This could let researchers test more ideas faster and with fewer resources.

Questions answered