Development and Application of a Data-Driven Reaction Classification Model: Comparison of an Electronic Lab Notebook and Medicinal Chemistry Literature

Abstract

Reaction classification has often been considered an important task for many different applications, and has traditionally been accomplished using hand-coded rule-based approaches. However, the availability of large collections of reactions enables data-driven approaches to be developed. We present the development and validation of a 336-class machine learning-based classification model integrated within a Conformal Prediction (CP) framework to associate reaction class predictions with confidence estimations. We also propose a data-driven approach for “dynamic” reaction fingerprinting to maximize the effectiveness of reaction encoding, as well as developing a novel reaction classification system that organizes labels into four hierarchical levels (SHREC: Sheffield Hierarchical REaction Classification). We show that the performance of the CP augmented model can be improved by defining confidence thresholds to detect predictions that are less likely to be false. For example, the external validation of the model reports 95% of predictions as correct by filtering out less than 15% of the uncertain classifications. The application of the model is demonstrated by classifying two reaction data sets: one extracted from an industrial ELN and the other from the medicinal chemistry literature. We show how confidence estimations and class compositions across different levels of information can be used to gain immediate insights on the nature of reaction collections and hidden relationships between reaction classes.

Enhancing reaction-based de novo design using a multi-label reaction class recommender

“…The process starts with a set of classified reactions. We use the hierarchical reaction classification system SHREC [13] which is compatible with the reaction vector approach and is summarised briefly below. SHREC is distributed across four hierarchical levels ranging from general reaction categories to increasingly more specific subclasses.…”

“…Thus, each entry in the training set represents one or more starting materials and a multi-label classification represented as a binary vector formation reactions. The reaction classifier was trained on data extracted from the USPD which was originally annotated with NameRxn classes, and a table showing the mapping of the original NameRxn labels to the four-level SHREC is shown in the Supporting Information of Ghiandoni et al [13]. Note that the four-level hierarchical labelling in SHREC is not exhaustive in terms of nomenclature due to its bias towards the USPD and NameRxn.…”

“…A set of 26,757 single-step reactions was extracted from Journal of Medicinal Chemistry (JMC) publications covering the period January-September 2018, using Reaxys. The reactions were cleaned using the same protocol as applied to the USPD (24,606 were retained); reaction vectors were calculated; the reactions were classified using SHREC and all entries associated with a high prediction probability were retained, based on a minimum credibility score of 0.25 as described in Ghiandoni et al [13]. 16,582 entries remained at this stage.…”

“…We have demonstrated a high degree of success in reproducing the known product(s) of a reaction based on known reactants and their associated reaction vectors [12]. In a recent publication we describe the generation of reaction vectors from the 1.8 million reactions extracted from the US patent literature in the development of a data-driven reaction classification method called SHREC [13]. SHREC takes a reaction vector as input and outputs a reaction class.…”

Mapping the Space of Chemical Reactions using Attention-Based Neural Networks

“…Unfortunately, the limited set of reaction classes used (only 50 most important ones) makes it difficult to judge how this algorithm would perform on a more comprehensive set. Recent work by Ghiandoni et al 13 introduced an alternative hierarchical classification scheme and random forest classifier for reaction classification. However, the model requires reaction center information as input, which limits the applicability of the method.…”