Search papers, labs, and topics across Lattice.
This study introduces PG-EVIKAL, an innovative approach that enhances molecular property predictions by leveraging neighbor fusion at test time, which utilizes the measured labels of structurally similar training molecules without the need for retraining. By incorporating a learned property-distance metric to prioritize neighbors based on property relevance, PG-EVIKAL significantly outperforms baseline evidential models, achieving a median RMSE reduction of 19.4% across 16 molecular datasets. The method also improves prediction calibration and adapts dynamically to new data in sequential-assay scenarios, showcasing the practical utility of evidential uncertainty in refining predictions.
Neighbor fusion at test time can refine molecular property predictions by leveraging the most relevant training data, achieving substantial accuracy gains without retraining.
A trained molecular property model can be refined at test time by correcting each prediction with the measured labels of the most similar training molecules, a retraining-free procedure we call neighbor fusion; evidential neural networks make it principled by using their aleatoric and epistemic uncertainty to parameterize a Bayesian update. Our main contribution, PG-EVIKAL, learns a property-distance metric to re-rank structurally similar neighbors by their property relevance before fusion, building on EVIKAL (scalar Kalman filter) and GP-EVIKAL (Gaussian process variant handling correlated neighbors). Evaluated on 16 molecular datasets, PG-EVIKAL reduces RMSE relative to the evidential model baseline on 14 of them, with a median reduction of 19.4%, and improves calibration; in sequential-assay scenarios it further incorporates newly measured molecules, refining predictions as they arrive without retraining. This work demonstrates that evidential uncertainty decomposition is not merely a calibration objective but an actionable inference resource that enables test-time refinement of molecular property predictions.
