Original title: Chemical and geometric representation fidelity improves drug--target affinity prediction
Yixiao Li, Yining Qian, Yefan Chen, Zenghui Chen +4 · September 2026
Medicines work by binding to specific target proteins in the body, but computer models struggle to predict how tightly they attach. Before testing interactions, computers must translate molecules into digital descriptions called representations. Many tools compress or smooth out these descriptions, accidentally wiping out tiny chemical details and exact physical shapes that determine how well a drug binds. To solve this, researchers built ReGeoDTA. It preserves chemical differences across drug molecules and keeps the smooth, real-world geometric shapes of proteins. Tested across three benchmark datasets, this method consistently made affinity predictions more accurate. When researchers deliberately degraded the representations, predictions got worse. Making the prediction models more complex could not recover the lost data.
You are baking a cake and measuring ingredients with a scoop that only measures whole cups. The half-cup of flour that would make the cake light is silently rounded away, and no amount of careful mixing later can put it back. Here, the system keeps the full chemical and structural detail of a drug and protein at the very start, so the later steps see everything they need.
When a drug candidate binds to its target protein, how well they fit predicts whether the drug works. This paper shows that the way computers encode that fit changes accuracy more than fancier prediction algorithms do. Better representation could help researchers rank promising drug candidates before costly experiments, saving time and money.
The abstract does not state specific limitations. It focuses on representation fidelity for affinity prediction and reports improvements on three benchmark datasets, but it does not discuss whether these gains hold outside those datasets, how much computational cost the preservation adds, or whether the approach works for other interaction types such as binding kinetics.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Chemical and geometric representation fidelity improves drug--target affinity prediction — Yixiao Li, Yining Qian, Yefan Chen et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.