Large pretrained models such as GPT-3 have had tremendous impact on modern natural language processing by leveraging self-supervised learning to learn salient representations that can be used to readily finetune on a wide variety of downstream tasks. We investigate the possibility of transferring such advances to molecular machine learning by building a chemical foundation model, ChemBERTa-2, using the language of SMILES. While labeled data for molecular prediction tasks is typically scarce, libraries of SMILES strings are readily available. In this work, we build upon ChemBERTa by optimizing the pretraining process. We compare multi-task and self-supervised pretraining by varying hyperparameters and pretraining dataset size, up to 77M compounds from PubChem. To our knowledge, the 77M set constitutes one of the largest datasets used for molecular pretraining to date. We find that with these pretraining improvements, we are competitive with existing state-of-the-art architectures on the MoleculeNet benchmark suite. We analyze the degree to which improvements in pretraining translate to improvement on downstream tasks.

ChemBERTa-2: Towards Chemical Foundation Models

Recent advances in generative models have made exploring design spaces easier for de novo molecule generation. However, popular generative models like GANs and normalizing flows face challenges such as training instabilities due to adversarial training and architectural constraints, respectively. Score-based generative models sidestep these challenges by modelling the gradient of the log probability density using a score function approximation, as opposed to modelling the density function directly, and sampling from it using annealed Langevin Dynamics. We believe that score-based generative models could open up new opportunities in molecule generation due to their architectural flexibility, such as replacing the score function with an SE(3) equivariant model. In this work, we lay the foundations by testing the efficacy of score-based models for molecule generation. We train a Transformer-based score function on Self-Referencing Embedded Strings (SELFIES) representations of 1.5 million samples from the ZINC dataset and use the Moses benchmarking framework to evaluate the generated samples on a suite of metrics.

Score-Based Generative Models for Molecule Generation

We propose a framework using normalizing-flow based models, SELF-Referencing Embedded Strings, and multi-objective optimization that efficiently generates small molecules. With an initial training set of only 100 small molecules, FastFlows generates thousands of chemically valid molecules in seconds. Because of the efficient sampling, substructure filters can be applied as desired to eliminate compounds with unreasonable moieties. Using easily computable and learned metrics for druglikeness, synthetic accessibility, and synthetic complexity, we perform a multi-objective optimization to demonstrate how FastFlows functions in a high-throughput virtual screening context. Our model is significantly simpler and easier to train than autoregressive molecular generative models, and enables fast generation and identification of druglike, synthesizable molecules.

FastFlows: Flow-Based Models for Molecular Graph Generation

GNNs and chemical fingerprints are the predominant approaches to representing molecules for property prediction. However, in NLP, transformers have become the de-facto standard for representation learning thanks to their strong downstream task transfer. In parallel, the software ecosystem around transformers is maturing rapidly, with libraries like HuggingFace and BertViz enabling streamlined training and introspection. In this work, we make one of the first attempts to systematically evaluate transformers on molecular property prediction tasks via our ChemBERTa model. ChemBERTa scales well with pretraining dataset size, offering competitive downstream performance on MoleculeNet and useful attention-based visualization

Bringing Atomistic Deep Learning to Prime Time

Differentiable physics provides a new approach for modeling and understanding the physical systems by pairing the new technology of differentiable programming with classical numerical methods for physical simulation. We survey the rapidly growing literature of differentiable physics techniques and highlight methods for parameter estimation, learning representations, solving differential equations, and developing what we call scientific foundation models using data and inductive priors. We argue that differentiable physics offers a new paradigm for modeling physical phenomena by combining classical analytic solutions with numerical methodology using the bridge of differentiable programming.