AI in chemistry: 13 records so far

Strongest gains

1. AI trained on tiny molecules matches costly DFT energy calculations on ones up to 54 atoms

   An AI model trained on molecules with up to 8 heavy atoms kept chemical accuracy against reference DFT calculations on organic molecules up to 54 atoms, at force-field computational cost.

   Benchmark or lab evaluation · Evidence: High · 76/100 · TRV-2026-0206

2. AI read 355 papers to pick a light-driven catalyst: it made hydrogen peroxide twice as fast

   An AI-guided literature search picked a light-driven catalyst that made hydrogen peroxide at 1.6 mM per hour, twice the 0.7 mM per hour of a comparison catalyst, reaching 82.3 mM after 72 hours.

   Controlled or validated comparison · Evidence: Moderate · 68/100 · TRV-2026-0383

3. AI model predicts how long 61 hydrocarbons take to clear a gas chromatograph, R2 of 0.993

   In a benchmark study using 608 data points, an adaptive boosting support vector regression model predicted retention times for 61 C1-C12 hydrocarbons across 73 gas chromatography settings, scoring R2 of 0.992-0.993 on validation and testing sets.

   Benchmark or lab evaluation · Evidence: Moderate · 52/100 · TRV-2026-0903