Home▸AI/ML Drug Discovery Infrastructure▸REINVENT 2D: Policy-Gradient Molecule Graph

REINVENT 2D: Policy-Gradient Molecule Graph

Watch a REINVENT-style policy-gradient agent build drug-like molecules token by token on a 2D molecular graph, scored against a target pocket size and heteroatom balance, and reinforced with the REINFORCE gradient over training batches. Drag to pan, scroll to zoom.

AI/ML Drug Discovery Infrastructure2DAdvanced60 FPS📱 Mobile-adapted⇄ 3D version

A compact policy learns to grow drug-like molecules one atom at a time — carbon, nitrogen or oxygen, branching, ring closure or stop — the same token-by-token generation REINVENT uses on SMILES strings. Every finished molecule is scored against a target pocket size and heteroatom balance, and the REINFORCE policy gradient pushes the token probabilities toward the choices that produced above-average reward. Watch the molecular graph grow live on a pannable, zoomable 2D canvas, batch by batch, as the reward history climbs and the policy's action-probability bars sharpen.

⚙ Under the hood

Watch a REINVENT-style policy learn to grow drug-like molecules token by token on a pannable, zoomable 2D graph, scored against a target pocket size and heteroatom balance and reinforced batch by batch with the REINFORCE policy gradient. A live bar chart shows the action policy sharpening as it trains.

reinforcement learningdrug discoverymolecular generationREINVENTpolicy gradientcanvas 2D

2D · HTML5 Canvas 2D · 60 FPS target · runs fully client-side, no install