Interpretable Drug Target Prediction Using Deep Neural Representation

Abstract

The identification of drug-target interactions (DTIs) is a key task in drug discovery, where drugs are chemical compounds and targets are proteins. Traditional DTI prediction methods are either time consuming (simulation-based methods) or heavily dependent on domain expertise (similarity-based and feature-based methods). In this work, we propose an end-to-end neural network model that predicts DTIs directly from low level representations. In addition to making predictions, our model provides biological interpretation using two-way attention mechanism. Instead of using simplified settings where a dataset is evaluated as a whole, we designed an evaluation dataset from BindingDB following more realistic settings where predictions of unobserved examples (proteins and drugs) have to be made. We experimentally compared our model with matrix factorization, similarity-based methods, and a previous deep learning approach. Overall, the results show that our model outperforms other approaches without requiring domain knowledge and feature engineering. In a case study, we illustrated the ability of our approach to provide biological insights to interpret the predictions.

Predicting Drug Protein Interaction using Quasi-Visual Question Answering System

“…BindingDB is a public database of experimentally measured binding affinities, focusing chiefly on the interactions of small molecules and proteins. In our experiments, we use the customized BindingDB dataset constructed by [Gao et al., 2018] for head-to-head comparisons. The dataset contains 39, 747 positive examples and 31, 218 negative examples from bindingDB.…”

“…Here, we report the RE scores at 0.5%, 1%, 2%, and 5% FPR thresholds. For BindingDB dataset, we also report the accuracy following [Gao et al , 2018].…”

“…We further assess our model on the BindingDB dataset. We compare our model with four base-lines: 1) Similarity-based method Tiresias [Fokoue et al , 2016]; 2) DBN [Wen et al , 2017], a deep learning method using middle-level features from predefined molecular finger-prints and protein descriptors; 3) E2E [Gao et al , 2018] using GCN and LSTM to process drug molecules and protein high-level information (Gene Ontology annotations) respectively; 4) GNN [Tsubaki et al , 2018].…”

“…In addition, though a few representation learning studies predicted DPIs based on protein sequence or their functional annotations [Tsubaki et al , 2018; Gao et al , 2018; Öztürk et al., 2018], their accuracies were limited with a lack of protein structural information.…”

“…With the recent increase in protein structural data and protein-ligand interaction datasets, there is a rapid progress in machine learning-based methods [Ragoza et al , 2017; Tsubaki et al , 2018; Gao et al , 2018]. Usually, the prediction is treated as a task of binary classification by integrating information of ligands, proteins, and their interactions in a unified framework.…”