LatentRxnFlow predicts reactions as continuous latent trajectories via Conditional Flow Matching from reactant-product pairs. Offers SOTA USPTO accuracy with trajectory diagnostics and uncertainty estimation. Enables error mitigation and reliable predictions.
Key Points
- 1.No mechanistic supervision needed
- 2.Geometric uncertainty signals
- 3.Gated inference fixes failure modes
Impact Analysis
Enhances transparency and trustworthiness in reaction prediction for discovery workflows. Prioritizes predictable outcomes, reducing validation needs.
Technical Details
Time-dependent dynamics in latent space anchored at products. Spectral analysis localizes errors without discrete steps.