Template-Free Retrosynthesis with Graph-Prior Augmented Transformers

This article describes a novel approach to retrosynthesis, a crucial task in chemistry, using transformer models. The use of graph-based priors is a key element, likely improving the model's understanding of chemical structures and reactions. The 'template-free' aspect suggests an advancement over traditional methods that rely on predefined reaction templates. The ArXiv source indicates this is a pre-print, so the results and impact are yet to be fully assessed.