SVDTI: Stacked variational autoencoder with SMILES-based drug representations for identifying drug-target interaction

The rapid identification of novel drug–target interactions (DTIs) remains a critical challenge in drug development, as traditional experimental methods are both resource-intensive and time-consuming. Motivated by the need to accelerate drug discovery and reduce experimental costs, computational stra...

Full description

Saved in:
Bibliographic Details
Published in:Neurocomputing (Amsterdam) Vol. 661; p. 131837
Main Author: Ha, Jihwan
Format: Journal Article
Language:English
Published: Elsevier B.V 14.01.2026
Subjects:
Drug
Drug-target interaction
Neural collaborative filtering
Protein
Stacked variational autoencoder
Target
Drug
Target
Neural collaborative filtering
Stacked variational autoencoder
Drug-target interaction
Protein
ISSN:0925-2312
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The rapid identification of novel drug–target interactions (DTIs) remains a critical challenge in drug development, as traditional experimental methods are both resource-intensive and time-consuming. Motivated by the need to accelerate drug discovery and reduce experimental costs, computational strategies have emerged as powerful alternatives, leveraging advanced algorithms and data-driven approaches to predict potential DTIs efficiently. In this paper, we introduce a novel method, which employs a stacked variational autoencoder (SVAE) to efficiently predict drug–target interactions, with the goal of enhancing the understanding and identification of these crucial relationships in drug discovery. This model leverages protein sequences and drug chemical properties as input features. It employs a stacked variational autoencoder (SVAE) with Long Short-Term Memory (LSTM) networks to map high-dimensional data into compact, informative low-dimensional vectors. The LSTM architecture captures temporal dependencies in protein sequences, thereby enhancing the model's ability to encode complex patterns. Next, the feature representation is fed into a neural collaborative filtering (NCF) model. This model combines the linear characteristics of matrix factorization with the nonlinear representation power of a multi-layer perceptron (MLP) to generate the final prediction, thereby improving the accuracy of DTI prediction. As a result, in comparison to existing state-of-the-art methods for DTIs prediction, our model demonstrates remarkable improvements in predictive performance. These findings highlight the capability of the proposed model to effectively integrate diverse sources of information for predicting DTIs, addressing critical challenges in drug discovery and offering a robust and efficient framework that contributes valuable perspectives to the field.
ISSN:0925-2312
DOI:10.1016/j.neucom.2025.131837