DL models now learn from less data; new ensemble framework boosts pharma ELN prediction robustness

marwinsegler · x · 2026-09-22

The author summarizes the team's findings: deep learning models can now learn from much less data, a new ensemble framework improves robustness, and fine-tuning on pharma electronic lab notebook (ELN) data proves feasible. Extensive analyses are in the paper, and model code and weights are open-sourced on GitHub.

The community is invited to hack the code, fine-tune, tweak, and add new models, and to report mispredictions so the model can be improved for everyone.

Related event: Microsoft's RetroChimera retrosynthesis model published in Nature and open-sourced(5 posts)→

Original post →

More from Research

Research channel →