No abstract

Efficient design and screening of the novel molecules is a major challenge in drug and material design. This paper focuses on a multi‐stage pipeline, in which several deep neural network models are combined to map discrete molecular representations into continuous vector space to later generate from it new molecular structures with desired properties. Here, the Attention‐based Sequence‐to‐Sequence model is added to “spellcheck” and correct generated structures, while the oversampling in the continuous space allows generating candidate structures with desired distribution for properties and molecular descriptors, even for a small reference datasets. We further use computer simulation to validate the desired properties in the numerical experiment. With the focus on the drug design, such a pipeline allows generating novel structures with a control of Synthetic Accessibility Score and a series of metrics that assess the drug‐likeliness. Our code is available at https://github.com/SoftServeInc/novel-molecule-generation.

Drug discovery pipelines typically involve high-throughput screening of large amounts of compounds in a search of potential drugs candidates. As a chemical space of small organic molecules is huge, a "navigation" over it urges for fast and lightweight computational methods, thus promoting machine-learning approaches for processing huge pools of candidates. In this contribution, we present a graph-based deep neural network for prediction of protein-drug binding affinity and assess its predictive power under thorough testing conditions. Within the suggested approach, both protein and drug molecules are represented as graphs and passed to separate graph sub-networks, then concatenated and regressed towards a binding affinity.The neural network is trained on two binding affinity datasets-PDBbind and data imported from RCSB Protein Data Bank. In order to explore the generalization capabilities of the model we go beyond traditional random or leave-cluster-out techniques and demonstrate the need for more elaborate model performance assessmentsix different strategies for test/train data partitioning (random, time-and propertyarranged, protein-and ligand-clustered) with a k-fold cross-validation are engaged.

Chemical yield is the percentage of the reactants converted to the desired products. Chemists use predictive algorithms to select high‐yielding reactions and score synthesis routes, saving time and reagents. This study suggests a novel graph neural network architecture for chemical yield prediction. The network combines structural information about participants of the transformation as well as molecular and reaction‐level descriptors. It works with incomplete chemical reactions and generates reactants‐product atom mapping. We show that the network benefits from advanced information by comparing it with several machine learning models and molecular representations. Models included logistic regression, support vector machine, CatBoost, and Bidirectional Encoder Representations from Transformers. Molecular representations included extended‐connectivity fingerprints, Morgan fingerprints, SMILESVec embeddings, and textual. Classification and regression objectives were assessed for each model and feature set. The goal of each classification model was to separate zero‐ and non‐zero‐yielding reactions. The models were trained and evaluated on a proprietary dataset of 10 reaction types. Also, the models were benchmarked on two public single reaction type datasets. The study was supplemented with analysis of data, results, and errors, as well as the impact of steric factors, side reactions, isolation, and purification efficiency. The supplementary code is available at https://github.com/SoftServeInc/yield-paper.