Advancements in neural machinery have led to a wide range of algorithmic solutions for molecular property prediction. Two classes of models in particular have yielded promising results: neural networks applied to computed molecular fingerprints or expert-crafted descriptors and graph convolutional neural networks that construct a learned molecular representation by operating on the graph structure of the molecule. However, recent literature has yet to clearly determine which of these two methods is superior when generalizing to new chemical space. Furthermore, prior research has rarely examined these new models in industry research settings in comparison to existing employed models. In this paper, we benchmark models extensively on 19 public and 16 proprietary industrial data sets spanning a wide variety of chemical end points. In addition, we introduce a graph convolutional model that consistently matches or outperforms models using fixed molecular descriptors as well as previous graph neural architectures on both public and proprietary data sets. Our empirical findings indicate that while approaches based on these representations have yet to reach the level of experimental reproducibility, our proposed model nevertheless offers significant improvements over models currently used in industrial workflows.

The synthesis of complex organic molecules requires several stages, from ideation to execution, that require time and effort investment from expert chemists. Here, we report a step toward a paradigm of chemical synthesis that relieves chemists from routine tasks, combining artificial intelligence–driven synthesis planning and a robotically controlled experimental platform. Synthetic routes are proposed through generalization of millions of published chemical reactions and validated in silico to maximize their likelihood of success. Additional implementation details are determined by expert chemists and recorded in reusable recipe files, which are executed by a modular continuous-flow platform that is automatically reconfigured by a robotic arm to set up the required unit operations and carry out the reaction. This strategy for computer-augmented chemical synthesis is demonstrated for 15 drug or drug-like substances.

Computer assistance in synthesis design has existed for over 40 years, yet retrosynthesis planning software has struggled to achieve widespread adoption. One critical challenge in developing high-quality pathway suggestions is that proposed reaction steps often fail when attempted in the laboratory, despite initially seeming viable. The true measure of success for any synthesis program is whether the predicted outcome matches what is observed experimentally. We report a model framework for anticipating reaction outcomes that combines the traditional use of reaction templates with the flexibility in pattern recognition afforded by neural networks. Using 15 000 experimental reaction records from granted United States patents, a model is trained to select the major (recorded) product by ranking a self-generated list of candidates where one candidate is known to be the major product. Candidate reactions are represented using a unique edit-based representation that emphasizes the fundamental transformation from reactants to products, rather than the constituent molecules’ overall structures. In a 5-fold cross-validation, the trained model assigns the major product rank 1 in 71.8% of cases, rank ≤3 in 86.7% of cases, and rank ≤5 in 90.8% of cases.

Computer-aided synthesis planning (CASP) is focused on the goal of accelerating the process by which chemists decide how to synthesize small molecule compounds. The ideal CASP program would take a molecular structure as input and output a sorted list of detailed reaction schemes that each connect that target to purchasable starting materials via a series of chemically feasible reaction steps. Early work in this field relied on expert-crafted reaction rules and heuristics to describe possible retrosynthetic disconnections and selectivity rules but suffered from incompleteness, infeasible suggestions, and human bias. With the relatively recent availability of large reaction corpora (such as the United States Patent and Trademark Office (USPTO), Reaxys, and SciFinder databases), consisting of millions of tabulated reaction examples, it is now possible to construct and validate purely data-driven approaches to synthesis planning. As a result, synthesis planning has been opened to machine learning techniques, and the field is advancing rapidly. In this Account, we focus on two critical aspects of CASP and recent machine learning approaches to both challenges. First, we discuss the problem of retrosynthetic planning, which requires a recommender system to propose synthetic disconnections starting from a target molecule. We describe how the search strategy, necessary to overcome the exponential growth of the search space with increasing number of reaction steps, can be assisted through a learned synthetic complexity metric. We also describe how the recursive expansion can be performed by a straightforward nearest neighbor model that makes clever use of reaction data to generate high quality retrosynthetic disconnections. Second, we discuss the problem of anticipating the products of chemical reactions, which can be used to validate proposed reactions in a computer-generated synthesis plan (i.e., reduce false positives) to increase the likelihood of experimental success. While we introduce this task in the context of reaction validation, its utility extends to the prediction of side products and impurities, among other applications. We describe neural network-based approaches that we and others have developed for this forward prediction task that can be trained on previously published experimental data. Machine learning and artificial intelligence have revolutionized a number of disciplines, not limited to image recognition, dictation, translation, content recommendation, advertising, and autonomous driving. While there is a rich history of using machine learning for structure-activity models in chemistry, it is only now that it is being successfully applied more broadly to organic synthesis and synthesis design. As reported in this Account, machine learning is rapidly transforming CASP, but there are several remaining challenges and opportunities, many pertaining to the availability and standardization of both data and evaluation metrics, which must be addressed by the community at large.

The task of learning an expressive molecular representation is central to developing quantitative structure-activity and property relationships. Traditional approaches rely on group additivity rules, empirical measurements or parameters, or generation of thousands of descriptors. In this paper, we employ a convolutional neural network for this embedding task by treating molecules as undirected graphs with attributed nodes and edges. Simple atom and bond attributes are used to construct atom-specific feature vectors that take into account the local chemical environment using different neighborhood radii. By working directly with the full molecular graph, there is a greater opportunity for models to identify important features relevant to a prediction task. Unlike other graph-based approaches, our atom featurization preserves molecule-level spatial information that significantly enhances model performance. Our models learn to identify important features of atom clusters for the prediction of aqueous solubility, octanol solubility, melting point, and toxicity. Extensions and limitations of this strategy are discussed.

We demonstrate molecular similarity to be a surprisingly effective metric for proposing and ranking one-step retrosynthetic disconnections based on analogy to precedent reactions. The developed approach mimics the retrosynthetic strategy defined implicitly by a corpus of known reactions without the need to encode any chemical knowledge. Using 40 000 reactions from the patent literature as a knowledge base, the recorded reactants are among the top 10 proposed precursors in 74.1% of 5000 test reactions, providing strong quantitative support for our methodology. Extension of the one-step strategy to multistep pathway planning is demonstrated and discussed for two exemplary drug products.

We present a supervised learning approach to predict the products of organic reactions given their reactants, reagents, and solvent(s).

Reaction condition recommendation is an essential element for the realization of computer-assisted synthetic planning. Accurate suggestions of reaction conditions are required for experimental validation and can have a significant effect on the success or failure of an attempted transformation. However, de novo condition recommendation remains a challenging and under-explored problem and relies heavily on chemists’ knowledge and experience. In this work, we develop a neural-network model to predict the chemical context (catalyst(s), solvent(s), reagent(s)), as well as the temperature most suitable for any particular organic reaction. Trained on ∼10 million examples from Reaxys, the model is able to propose conditions where a close match to the recorded catalyst, solvent, and reagent is found within the top-10 predictions 69.6% of the time, with top-10 accuracies for individual species reaching 80–90%. Temperature is accurately predicted within ±20 °C from the recorded temperature in 60–70% of test cases, with higher accuracy for cases with correct chemical context predictions. The utility of the model is illustrated through several examples spanning a range of common reaction classes. We also demonstrate that the model implicitly learns a continuous numerical embedding of solvent and reagent species that captures their functional similarity.