COATI: Multimodal Contrastive Pretraining for Representing and Traversing Chemical Space

Abstract

Creating a successful small molecule drug is a challenging multiparameter optimization problem in an effectively infinite space of possible molecules. Generative models have emerged as powerful tools for traversing data manifolds composed of images, sounds, and text and offer an opportunity to dramatically improve the drug discovery and design process. To create generative optimization methods that are more useful than brute-force molecular generation and filtering via virtual screening, we propose that four integrated features are necessary: large, quantitative data sets of molecular structure and activity, an invertible vector representation of realistic accessible molecules, smooth and differentiable regressors that quantify uncertainty, and algorithms to simultaneously optimize properties of interest. Over the course of 12 months, Terray Therapeutics has collected a data set of 2 billion quantitative binding measurements of small molecules to therapeutic targets, which directly motivates multiparameter generative optimization of molecules conditioned on these data. To this end, we present contrastive optimization for accelerated therapeutic inference (COATI), a pretrained, multimodal encoder-decoder model of druglike chemical space. COATI is constructed without any human biasing of features, using contrastive learning from text and 3D representations of molecules to allow for downstream use with structural models. We demonstrate that COATI possesses many of the desired properties of universal molecular embedding: fixed-dimension, invertibility, autoencoding, accurate regression, and low computation cost. Finally, we present a novel metadynamics algorithm for generative optimization using a small subset of our proprietary data collected for a model protein, carbonic anhydrase, designing molecules that satisfy the multiparameter optimization task of potency, solubility, and drug likeness. This work sets the stage for fully integrated generative molecular design and optimization for small molecules.

The Potential of Neural Network Potentials

“…For example, this can be used to fit interatomic potentials to experimentally obtainable structural data . It is also an important component of modern generative models for drug discovery . Finally, this idea is also being applied to DFT functional development, where all of the key parameters used to define a DFT functional are treated as differentiable parameters that can be optimized with respect to some loss …”

From Latent Dynamics to Meaningful Representations

“…The ability to learn meaningful and useful representations of data is a key challenge toward applying machine learning (ML) and artificial intelligence (AI) to various real-world problems. On one hand, a useful representation extracts and organizes the discriminative information from data to support effective ML for downstream tasks. − In the realm of chemistry, such representations offer promising avenues for enhancing predictions of molecular properties, forecasting reactions, predicting pharmacological activities, facilitating exploration of vast chemical spaces, and accelerating computational simulations. , On the other hand, a meaningful representation is often more interpretable, which is fundamental to helping humans trust AI. Recently, a variety of representation learning methods have been proposed based on the idea of autoencodinglearning a mapping from high dimensional data to a low dimensional representation or latent space which is able to approximately reconstruct the original data .…”