Large-Scale Chemical Language Representations Capture Molecular Structure and Properties

TL;DR

This paper introduces MoLFormer, a transformer-based model trained on 1.1 billion molecules' SMILES sequences, which captures molecular structure and properties effectively, outperforming existing models on multiple property prediction tasks.

Contribution

The paper presents a large-scale, efficient transformer model for molecular embeddings trained on massive unlabelled data, demonstrating superior performance and structural understanding over prior models.

Findings

MoLFormer outperforms existing baselines on multiple benchmarks.

The model captures spatial relationships between atoms.

It effectively predicts various molecular properties.

Abstract

Models based on machine learning can enable accurate and fast molecular property predictions, which is of interest in drug discovery and material design. Various supervised machine learning models have demonstrated promising performance, but the vast chemical space and the limited availability of property labels make supervised learning challenging. Recently, unsupervised transformer-based language models pretrained on a large unlabelled corpus have produced state-of-the-art results in many downstream natural language processing tasks. Inspired by this development, we present molecular embeddings obtained by training an efficient transformer encoder model, MoLFormer, which uses rotary positional embeddings. This model employs a linear attention mechanism, coupled with highly distributed training, on SMILES sequences of 1.1 billion unlabelled molecules from the PubChem and ZINC datasets. We show that the learned molecular representation outperforms existing baselines, including supervised and self-supervised graph neural networks and language models, on several downstream tasks from ten benchmark datasets. They perform competitively on two others. Further analyses, specifically through the lens of attention, demonstrate that MoLFormer trained on chemical SMILES indeed learns the spatial relationships between atoms within a molecule. These results provide encouraging evidence that large-scale molecular language models can capture sufficient chemical and structural information to predict various distinct molecular properties, including quantum-chemical properties.