The story of machine learning in general, and its application to molecular design in particular, has been a tale of evolving representations of data. Understanding the implications of the use of a particular representation -- including the existence of so-called `activity cliffs' for cheminformatics models -- is the key to their successful use for molecular discovery. In this work we present a physics-inspired methodology which exploits analogies between model response surfaces and energy landscapes to richly describe the relationship between the representation and the model. From these similarities, a metric emerges which is analogous to the commonly used frustration metric from the chemical physics community. This new property shows state-of-the-art prediction of model error, whilst belonging to a novel class of roughness measure that extends beyond the known data allowing the trivial identification of activity cliffs even in the absence of related training or evaluation data.

Deep learning models for anticipating the products of organic reactions have found many use cases, including validating retrosynthetic pathways and constraining synthesis-based molecular design tools. Despite compelling performance on popular benchmark tasks, strange and erroneous predictions sometimes ensue when using these models in practice. The core issue is that common benchmarks test models in an in-distribution setting, whereas many real-world uses for these models are in out-of-distribution settings and require a greater degree of extrapolation. To better understand how current reaction predictors work in out-of-distribution domains, we report a series of more challenging evaluations of a prototypical SMILES-based deep learning model. First, we illustrate how performance on randomly sampled data sets is overly optimistic compared to performance when generalizing to new patents or new authors. Second, we conduct time splits that evaluate how models perform when tested on reactions published years after those in their training set, mimicking real-world deployment. Finally, we consider extrapolation across reaction classes to reflect what would be required for the discovery of novel reaction types. This panel of tasks can reveal the capabilities and limitations of today's reaction predictors, acting as a crucial first step in the development of tomorrow's next-generation models capable of reaction discovery.

Drug discovery is inherently challenged by a multiple criteria decision making problem. The arduous path from hit discovery through lead optimization and preclinical candidate selection necessitates the evolution of a plethora of molecular properties. In this study, we focus on the hit discovery phase while beginning to address multiple criteria critical to the development of novel therapeutics to treat Mycobacterium tuberculosis infection. We develop a hybrid structure- and ligand-based pipeline for nominating diverse inhibitors targeting the β-ketoacyl synthase KasA by employing a Bayesian optimization-guided docking method and an ensemble model for compound nominations based on machine learning models for in vitro antibacterial efficacy, as characterized by minimum inhibitory concentration (MIC), and mouse pharmacokinetic (PK) plasma exposure. The application of our pipeline to the Enamine HTS library of 2.1M molecules resulted in the selection of 93 compounds, the experimental validation of which revealed exceptional PK (41%) and MIC (19%) success rates. Twelve compounds meet hit-like criteria in terms of MIC and PK profile and represent promising seeds for future drug discovery programs.

The discovery of therapeutic molecules is fundamentally a multi-objective optimization problem. One formulation of the problem is to identify molecules that simultaneously exhibit strong binding affinity for a target protein, minimal off-target interactions, and suitable pharmacokinetic properties. Inspired by prior work that uses active learning to accelerate the identification of strong binders, we implement multi-objective Bayesian optimization to reduce the computational cost of multi-property virtual screening and apply it to the identification of ligands predicted to be selective based on docking scores to on- and off-targets. We demonstrate the superiority of Pareto optimization over scalarization across three case studies. Further, we use the developed optimization tool to search a virtual library of over 4M molecules for those predicted to be selective dual inhibitors of EGFR and IGF1R, acquiring 100% of the molecules that form the library's Pareto front after exploring only 8% of the library. This workflow and associated open source software can reduce the screening burden of molecular design projects and is complementary to research aiming to improve the accuracy of binding predictions and other molecular properties.

By drawing on similarities between energy landscapes and model response surfaces we gain new insight into model performance, even in the absence of data.

Quantitative structure-property relationships (QSPRs) aid in understanding molecular properties as a function of molecular structure. When the correlation between structure and property weakens, a dataset is described as "rough," but this characteristic is partly a function of the chosen representation. Among possible molecular representations are those from recently-developed "foundation models" for chemistry which learn molecular representation from unlabeled samples via self-supervision. However, the performance of these pretrained representations on property prediction benchmarks is mixed when compared to baseline approaches. We sought to understand these trends in terms of the roughness of the underlying QSPR surfaces. We introduce a reformulation of the roughness index (ROGI), ROGI-XD, to enable comparison of ROGI values across representations and evaluate various pretrained representations and those constructed by simple fingerprints and descriptors. We show that pretrained representations do not produce smoother QSPR surfaces, in agreement with previous empirical results of model accuracy. Our findings suggest that imposing stronger assumptions of smoothness with respect to molecular structure during model pretraining could aid in the downstream generation of smoother QSPR surfaces. Pretrained molecular representations are often thought to provide smooth, navigable latent spaces; analysis by ROGI-XD suggests they are no smoother than fixed descriptor/fingerprint representations.

Deep learning has become a powerful and frequently employed tool for the prediction of molecular properties, thus creating a need for open-source and versatile software solutions that can be operated by non-experts. Among current approaches, directed message-passing neural networks (D-MPNNs) have proven to perform well on a variety of property prediction tasks. The software package Chemprop implements the D-MPNN architecture, and offers simple, easy, and fast access to machine-learned molecular properties. Compared to its initial version, we present a multitude of new Chemprop functionalities such as the support of multi-molecule properties, reactions, atom/bond-level properties, and spectra. Further, we incorporate various uncertainty quantification and calibration methods along with related metrics, as well as pretraining and transfer learning workflows, improved hyperparameter optimization, and other customization options concerning loss functions or atom/bond features. We benchmark D-MPNN models trained using Chemprop with the new reaction, atom-level and spectra functionality on a variety of property prediction datasets, including MoleculeNet and SAMPL, and observe state-of-the-art performance on the prediction of water-octanol partition coefficients, reaction barrier heights, atomic partial charges, and absorption spectra. Chemprop enables out-of-the-box training of D-MPNN models for a variety of problem settings in a fast, user-friendly, and open-source software.

pyscreener is a Python library that seeks to alleviate the challenges of large-scale structure-based design using computational docking. It provides a simple and uniform interface that is agnostic to the backend docking engine with which to calculate the docking score of a given molecule in a specified active site. Additionally, pyscreener features first-class support for task distribution, allowing users to seamlessly scale their code from a local, multi-core setup to a large, heterogeneous resource allocation.

High-throughput virtual screening is an indispensable technique utilized in the discovery of small molecules. In cases where the library of molecules is exceedingly large, the cost of an exhaustive virtual screen may be prohibitive. Model-guided optimization has been employed to lower these costs through dramatic increases in sample efficiency compared to random selection. However, these techniques introduce new costs to the workflow through the surrogate model training and inference steps. In this study, we propose an extension to the framework of model-guided optimization that mitigates inference costs using a technique we refer to as design space pruning (DSP), which irreversibly removes poor-performing candidates from consideration. We study the application of DSP to a variety of optimization tasks and observe significant reductions in overhead costs while exhibiting similar performance to the baseline optimization. DSP represents an attractive extension of model-guided optimization that can limit overhead costs in optimization settings where these costs are non-negligible relative to objective costs, such as docking.

In molecular discovery and drug design, structure-property relationships and activity landscapes are often qualitatively or quantitatively analyzed to guide the navigation of chemical space. The roughness (or smoothness) of these molecular property landscapes is one of their most studied geometric attributes, as it can characterize the presence of activity cliffs, with rougher landscapes generally expected to pose tougher optimization challenges. Here, we introduce a general, quantitative measure for describing the roughness of molecular property landscapes. The proposed roughness index (ROGI) is loosely inspired by the concept of fractal dimension and strongly correlates with the out-of-sample error achieved by machine learning models on numerous regression tasks.

Structure-based virtual screening is an important tool in early stage drug discovery that scores the interactions between a target protein and candidate ligands. As virtual libraries continue to grow (in excess of 108 molecules), so too do the resources necessary to conduct exhaustive virtual screening campaigns on these libraries. However, Bayesian optimization techniques, previously employed in other scientific discovery problems, can aid in their exploration: a surrogate structure–property relationship model trained on the predicted affinities of a subset of the library can be applied to the remaining library members, allowing the least promising compounds to be excluded from evaluation. In this study, we explore the application of these techniques to computational docking datasets and assess the impact of surrogate model architecture, acquisition function, and acquisition batch size on optimization performance. We observe significant reductions in computational costs; for example, using a directed-message passing neural network we can identify 94.8% or 89.3% of the top-50 000 ligands in a 100M member library after testing only 2.4% of candidate ligands using an upper confidence bound or greedy acquisition strategy, respectively. Such model-guided searches mitigate the increasing computational costs of screening increasingly large virtual libraries and can accelerate high-throughput virtual screening campaigns with applications beyond docking.

An enantioselective, radical-based method for the intramolecular hydroamination of alkenes with sulfonamides is reported. These reactions are proposed to proceed via N-centered radicals formed by proton-coupled electron transfer (PCET) activation of sulfonamide N–H bonds. Non-covalent interactions between the neutral sulfonamidyl radical and a chiral phosphoric acid generated in the PCET event are hypothesized to serve as the basis for asymmetric induction in a subsequent C–N bond forming step, achieving selectivities of up to 98:2 er. These results offer further support for the ability of non-covalent interactions to enforce stereoselectivity in reactions of transient and highly reactive open-shell intermediates.