Molecular optimization by capturing chemist’s intuition using deep neural networks

Abstract

A main challenge in drug discovery is finding molecules with a desirable balance of multiple properties. Here, we focus on the task of molecular optimization, where the goal is to optimize a given starting molecule towards desirable properties. This task can be framed as a machine translation problem in natural language processing, where in our case, a molecule is translated into a molecule with optimized properties based on the SMILES representation. Typically, chemists would use their intuition to suggest chemical transformations for the starting molecule being optimized. A widely used strategy is the concept of matched molecular pairs where two molecules differ by a single transformation. We seek to capture the chemist’s intuition from matched molecular pairs using machine translation models. Specifically, the sequence-to-sequence model with attention mechanism, and the Transformer model are employed to generate molecules with desirable properties. As a proof of concept, three ADMET properties are optimized simultaneously: logD, solubility, and clearance, which are important properties of a drug. Since desirable properties often vary from project to project, the user-specified desirable property changes are incorporated into the input as an additional condition together with the starting molecules being optimized. Thus, the models can be guided to generate molecules satisfying the desirable properties. Additionally, we compare the two machine translation models based on the SMILES representation, with a graph-to-graph translation model HierG2G, which has shown the state-of-the-art performance in molecular optimization. Our results show that the Transformer can generate more molecules with desirable properties by making small modifications to the given starting molecules, which can be intuitive to chemists. A further enrichment of diverse molecules can be achieved by using an ensemble of models.

Lib-INVENT: Reaction Based Generative Scaffold Decoration for in silico Library Design

“…This nature of the modelling objective affects the choices made when preparing the validation set. While it is common to randomly hold out a portion of the training data and use these for evaluations 32 , the sliced dataset used here does not lend itself well to this approach. To be able to fairly judge the generalization ability of the model on previously unseen scaffolds, it is necessary to ensure that the validation scaffolds are not present in the training dataset.…”

Transformer Neural Network-Based Molecular Optimization Using General Transformations

“…Following [25], the SMILES representation of molecule and the Transformer model from NLP is used in our study. The Transformer is trained on a set of molecular pairs together with the property changes between source and target molecules.…”

“…Figure 1 shows an example of source and target sequences which are fed into the Transformer model during training. More details can be found in [25].…”

“…Given a set of molecular pairs {(X, Y, Z)} where X represents source molecule, Y represents target During testing, given a new (X, Z) ∈ X × Z, the model will be expected to generate a diverse set of target molecules with desirable properties [25].…”

“…Jin et al [17,23,24] utilized molecular graph representations and viewed the molecular optimization problem as a graph-to-graph translation problem. He et al [25,26] instead utilized the stringbased representation, the simplified molecular-input line-entry system (SMILES) [27] and employed the machine translation models [28,29] from natural language processing (NLP). They trained machine translation models (Transformer and Seq2Seq) to mimic the chemist's approach of using matched molecular pairs (MMPs) [30,31] where two molecules differ by a single chemical transformation.…”