PubChem atom environments

Improving the quality of chemical language model outcomes with atom-in-SMILES tokenization

Tokenization is an important preprocessing step in natural language processing that may have a significant influence on prediction quality. This research showed that the traditional SMILES tokenization has a certain limitation that results in tokens failing to reflect the true nature of molecules. To address this issue, we developed the atom-in-SMILES tokenization scheme that eliminates ambiguities in the generic nature of SMILES tokens. Our results in multiple chemical translation and molecular property prediction tasks demonstrate that proper tokenization has a significant impact on prediction quality. In terms of prediction accuracy and token degeneration, atom-in-SMILES is more effective method in generating higher-quality SMILES sequences from AI-based chemical models compared to other tokenization and representation schemes. We investigated the degrees of token degeneration of various schemes and analyzed their adverse effects on prediction quality. Additionally, token-level repetitions were quantified, and generated examples were incorporated for qualitative examination. We believe that the atom-in-SMILES tokenization has a great potential to be adopted by broad related scientific communities, as it provides chemically accurate, tailor-made tokens for molecular property prediction, chemical translation, and molecular generative models.

Retrosynthetic reaction pathway prediction through neural machine translation of atomic environments

Designing efficient synthetic routes for a target molecule remains a major challenge in organic synthesis. Atom environments are ideal, stand-alone, chemically meaningful building blocks providing a high-resolution molecular representation. Our approach mimics chemical reasoning, and predicts reactant candidates by learning the changes of atom environments associated with the chemical reaction. Through careful inspection of reactant candidates, we demonstrate atom environments as promising descriptors for studying reaction route prediction and discovery. Here, we present a new single-step retrosynthesis prediction method, viz. RetroTRAE, being free from all SMILES-based translation issues, yields a top-1 accuracy of 58.3% on the USPTO test dataset, and top-1 accuracy reaches to 61.6% with the inclusion of highly similar analogs, outperforming other state-of-the-art neural machine translation-based methods. Our methodology introduces a novel scheme for fragmental and topological descriptors to be used as natural inputs for retrosynthetic prediction tasks.

Stereo-Aware Extension of HOSE Codes

Descriptions of molecular environments have many applications in chemoinformatics, including chemical shift prediction. Hierarchically ordered spherical environment (HOSE) codes are the most popular such descriptions. We developed a method to extend these with stereochemistry information. It enables distinguishing atoms which would be considered identical in traditional HOSE codes. The use of our method is demonstrated by chemical shift predictions for molecules in the nmrshiftdb2 database. We give a full specification and an implementation.

Weighted Averaging Scheme and Local Atomic Descriptor for pKa Prediction Based on Density Functional Theory

As a continuation of our work on developing a density functional theory-based pK_a predictor, we present conceptual improvements to our previously published shell model, which is a hierarchical organization of pK_a training sets and which, in principle, covers all chemical space. The improvements concern the way the studied chemical compound is associated with the data points from the training sets. By introducing a new descriptor of the local atomic environment which foregoes dependence on chemical bonding and connectivity, we are able to automatically locate molecules from the training set that are most relevant to the proton dissociation equilibrium under study. This new scheme leads to the prediction of a single pK_a value weighted across multiple training sets and thus patches a defect disclosed in the formulation of our previous model. Using the new parametrization approach, the pK_a prediction gets rid of outliers reported in previous applications of our approach, eliminates ambiguity in interpreting the results, and improves the overall accuracy. Our new treatment accounts for multiple conformations both on the level of energetics and parametrization. Illustrative results are shown for several types of chemical structures containing guanidine, amidine, amine, and phenol functional groups, and which are representative of practically important large and flexible drug-like molecules. Our method's performance is compared to the performance of other previously published pK_a prediction methods. Further possible improvements to the organization of the training sets and the potential application of our new local atomic descriptor to other kinds of parametrizations are discussed.