Molecule Attention Transformer (Deep Learning Paper Explained)

Molecule attention transformer (MAT) is an interesting application of the transformer architecture on chemistry domain. MAT shows the very promising and intriguing results on the several molecule property prediction tasks and also paves the way for more adaptation of deep learning approaches in chemistry research. This video introduces the model architecture and analyzes why it is so effective.

The paper: https://arxiv.org/abs/2002.08264 The official implementation: https://github.com/gmum/MAT

Abstract Designing a single neural network architecture that performs competitively across a range of molecule property prediction tasks remains largely an open challenge, and its solution may unlock a widespread use of deep learning in the drug discovery industry. To move towards this goal, we propose Molecule Attention Transformer (MAT). Our key innovation is to augment the attention mechanism in Transformer using inter-atomic distances and the molecular graph structure. Experiments show that MAT performs competitively on a diverse set of molecular prediction tasks. Most importantly, with a simple self-supervised pretraining, MAT requires tuning of only a few hyperparameter values to achieve state-of-the-art performance on downstream tasks. Finally, we show that attention weights learned by MAT are interpretable from the chemical point of view. ... https://www.youtube.com/watch?v=9qSEFFqdXdA