June 2026
Model Performance Series
Astra AI on Enzymes
Enzymes are the most heavily exploited drug-target class in medicine — from kinase to protease inhibitors — and a distinctive computational target: defined by the chemistry they perform rather than a shared fold, with catalysis at a handful of active-site residues. This report shows how Orbion's Astra AI Suite characterizes them from sequence alone — function, topology, PTM sites, binding pockets, and thermostability — measured against public experimental ground truth across 6,304 catalytic proteins.
95.3% Recognized as Enzymatic
82% Pocket Success
ρ=0.93 ΔTm (T4 Lysozyme)
ρ=0.93 Thermostability Correlation
The suite's flagship result. On a deep mutational scan of T4 lysozyme — the canonical stability reference — predicted vs. experimental ΔTm reach ρ 0.93 (n=315); across the full set, ρ 0.88 with 90% directional accuracy on strong-effect mutations. Rank order and direction are reliable.
95.3% Recognized as Enzymatic
Function is the defining question for enzymes, and the EC head answers it — 95.3% recognized as enzymatic, distributed across all seven EC top-classes in proportions that track nature. GO molecular-function recovered in the top-5 for 94.2%.
82% Pocket Success
62% ligand recall and 82% pocket-success across 2,318 enzymes with co-crystal data. Strongest on the chemically well-defined catalytic centres — dioxygenases, kinase ATP sites, FAD/NAD flavoenzymes.
Have a question?
Whether you want to understand our models, test a specific protein, or just ask how Orbion applies to your research — send us your question and we'll get back to you. Or, if you'd prefer a live walkthrough, book a demo.
