# DeepDTA: Deep Drug-Target Binding Affinity Prediction

**Authors:** Hakime Öztürk, Elif Ozkirimli, Arzucan Özgür  
**arXiv:** [1801.10193](https://arxiv.org/abs/1801.10193)  
**Date:** 2018-01-30  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1801.10193v2/

## Key points

- Identifying new drug-target interactions is crucial in drug discovery
- Most computational methods focus on binary classification of DT interactions
- Protein-ligand interactions involve a continuum of binding strength values, known as binding affinity, which is challenging to predict accurately
- Advanced learning techniques such as deep learning architectures can be used to predict binding affinities with the increase in available affinity data in DT knowledge-bases
- This study proposes a deep-learning based model that uses only sequence information of both targets and drugs to predict DT interaction binding affinities
- The proposed model uses 1D representations of targets and drugs with convolutional neural networks (CNNs)
- The model achieved the best Concordance Index (CI) performance compared to two state-of-the-art methods: KronRLS algorithm and SimBoost, by constructing high-level representations of a drug and a target via CNNs.
- This study presents a novel approach for predicting DT interaction binding affinities using deep learning architecture with only sequence information.

## Summary

In the field of drug discovery, identifying new drug-target (DT) interactions is crucial. While many computational methods have been proposed to predict DT interactions, most focus on binary classification - determining whether a DT pair interacts or not. However, protein-ligand interactions involve a continuum of binding strength values, known as binding affinity, which remains a challenge to predict accurately. With the increase in available affinity data in DT knowledge-bases, advanced learning techniques such as deep learning architectures can be used to predict binding affinities. This study proposes a deep-learning based model that uses only sequence information of both targets and drugs to predict DT interaction binding affinities. Unlike previous studies that use either 3D structures of protein-ligand complexes or 2D features of compounds for prediction, this work models protein sequences and compound 1D representations with convolutional neural networks (CNNs). The results show that the proposed deep learning-based model using 1D representations of targets and drugs is an effective approach for drug target binding affinity prediction. The authors also compare their model's performance against two state-of-the-art methods: KronRLS algorithm and SimBoost. In one larger benchmark dataset, their model achieved the best Concordance Index (CI) performance by constructing high-level representations of a drug and a target via CNNs. This outperformed both KronRLS algorithm and SimBoost. Overall, this study presents a novel approach for predicting DT interaction binding affinities using deep learning architecture with only sequence information. The results demonstrate its effectiveness compared to other state-of-the-art methods and highlight its potential application in drug discovery research.