C5T5: Controllable Generation of Organic Molecules with Transformers

Daniel Rothchild, Alex Tamkin, Julie Yu, Ujval Misra, Joseph E. Gonzalez

Keywords: molecular modeling, sequence modeling, conditional sequence modeling, drug discovery

Abstract: Methods for designing organic materials with desired properties have high potential impact across fields such as medicine, renewable energy, petrochemical engineering, and agriculture. However, using generative models for this task is difficult because candidate compounds must satisfy many constraints, including synthetic accessibility, intellectual property attributes, ``chemical beauty'' (Bickerton et al., 2020), and other considerations that are intuitive to domain experts but can be challenging to quantify. We propose C5T5, a novel self-supervised pretraining method that works in tandem with domain experts by making zero-shot select-and-replace edits, altering organic substances towards desired property values. C5T5 operates on IUPAC names---a standardized molecular representation that intuitively encodes rich structural information for organic chemists but that has been largely ignored by the ML community. Our technique requires no edited molecule pairs to train and only a rough estimate of molecular properties, and it has the potential to model long-range dependencies and symmetric molecular structures more easily than graph-based methods. We demonstrate C5T5's effectiveness on four physical properties relevant for drug discovery, showing that it learns successful and chemically intuitive strategies for altering molecules towards desired property values.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2108.10307/code)