Improving reaction prediction through chemically aware transfer learning

2025/01/01 by Angus B. Keto, Taicheng Guo, Nils Gönnheimer +3 · 1 voice
Chemistry · Computer Science · Engineering · #Advanced Chemical Sensor Technologies #Computational Drug Discovery Methods #Various Chemistry Research Topics

paper · doi:10.1039/d4dd00412d

openalex publication_date 2025/01/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/06/22

Abstract

Pretraining of NERF models on chemically related mechanisms significantly improves the performance compared to pretraining by larger, mechanistically dissimilar reaction datasets.

Citations

• A Survey on Transfer Learning
• BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension

Discussions

• In our next featured article, Wiest et al. show the data efficiencies that can be achieved when using models pre-trained on datasets of mechanistically related reactions, as opposed to larger sets of [bsky, 2 points, 0 comments]

Related

• Pre-training Transformers for Molecular Property Prediction Using Reaction Prediction
• Understanding and improving transferability in machine-learned activation energy predictors
• A Self-feedback Knowledge Elicitation Approach for Chemical Reaction Predictions
• Toward On-the-Fly Prediction of Reaction Energetics for High-Throughput Screening
• MolKD: Distilling Cross-Modal Knowledge in Chemical Reactions for Molecular Property Prediction