Prediction of pharmacological activities from chemical structures with graph convolutional neural networks

Abstract

Many therapeutic drugs are compounds that can be represented by simple chemical structures, which contain important determinants of affinity at the site of action. Recently, graph convolutional neural network (GCN) models have exhibited excellent results in classifying the activity of such compounds. For models that make quantitative predictions of activity, more complex information has been utilized, such as the three-dimensional structures of compounds and the amino acid sequences of their respective target proteins. As another approach, we hypothesized that if sufficient experimental data were available and there were enough nodes in hidden layers, a simple compound representation would quantitatively predict activity with satisfactory accuracy. In this study, we report that GCN models constructed solely from the two-dimensional structural information of compounds demonstrated a high degree of activity predictability against 127 diverse targets from the ChEMBL database. Using the information entropy as a metric, we also show that the structural diversity had less effect on the prediction performance. Finally, we report that virtual screening using the constructed model identified a new serotonin transporter inhibitor with activity comparable to that of a marketed drug in vitro and exhibited antidepressant effects in behavioural studies.

De novo Prediction of Cell-Drug Sensitivities Using Deep Learning-based Graph Regularized Matrix Factorization

“…We noted that the performance of new drug sensitivity prediction is not as good as new cell line prediction, which may be due to limited expressiveness of our VAE-based representation of chemical structures. Recently, it has been shown that a family of graph neural networks can better provide a representation of chemical structures, 26 which can be explored in the future. Finally, the current representations of cellular states of cancer cells are derived using "black-box" neural networks, and interpretable deep learning models can be explored not only to achieve interpretability of our model but also to enhance its performance.…”

Feature, Function, and Information of Drug Transporter–Related Databases

“…Based on these valuable functions together with their comprehensive DT-related information, the available databases provided much-enhanced power in the research of drug metabolism and disposition. As shown in Table 5, these functions facilitated the structure-based drug design/identification , discovery of target druggability based on DT sequence (Frioux et al, 2020), disease/tissue-specific differential expression analysis (Yu et al, 2020), structure similarity search by the transported drug (Sakai et al, 2021), interplay analysis among multiple DT variabilities (Wang et al, 2021), functional analysis based on the signaling pathways (Sakil et al, 2017), functional annotation and systematic classification of DTs (Peng et al, 2021) (Hlavac et al, 2020). Overall, these customized database functions are very diverse, which are capable of promoting DT-based research on the drug ADME process.…”

“…ChEMBL database was used to identify a new inhibitor of serotonin transporter with comparable affinity to the commercial drug by structure similarity search and virtual screening (Sakai et al, 2021).…”

PlayMolecule Glimpse: Understanding Protein–Ligand Property Predictions with Interpretable Neural Networks

“…These latter methods have been reported to increase performance in some critical tasks for drug discovery, such as toxicity assessment, 5 , 6 pharmacokinetics, physicochemical property prediction, 7 − 10 and protein–ligand binding affinity prediction. 11 − 15 …”