Reproduce “General-purpose foundation models surpass specialized ones on chemical property prediction tasks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Stepan N. Ostarkov, Gregorii A. Semakin, Ivan A. Bespalov, Michael G. Medvedev
The study tests whether prompted general-purpose language models can predict molecular properties from few labeled examples. Models receive molecule representations and measured values, then predict properties for new molecules. On QM9 εLUMO, GPT-6 Astra reaches R² 0.85 with 60 examples, versus 0.53 for the best chemical foundation model; Astra leads across five regression tasks, but not BACE classification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.