Do Chemformers Dream of Organic Matter? Evaluating Transformer Models for Synthesis Prediction

Explore a conference talk that delves into the application of transformer models for synthesis prediction in pharmaceutical research. Learn how language models like transformers have been integrated into drug discovery processes, tackling tasks such as property prediction, molecular optimization, and reactivity predictions. Discover the efforts to train and implement transformer models for synthesis predictions in a production platform used daily by chemists. Examine the challenges encountered when training on large reaction data sets and compare the performance of these models with existing approaches for product prediction and retrosynthesis. Gain insights into how transformer models trained on diverse reaction sets can outperform current models. Understand the remaining obstacles preventing full adoption of transformer models in synthesis prediction. The speaker, Samuel Genheden, Associate Director at AstraZeneca and leader of the Deep Chemistry team, shares his expertise in computational methods, multiscale approaches, and AI-assisted retrosynthesis planning, offering a comprehensive look at the intersection of artificial intelligence and pharmaceutical chemistry.