A geometric deep learning framework for drug repositioning over heterogeneous information networks

Abstract

Drug repositioning (DR) is a promising strategy to discover new indicators of approved drugs with artificial intelligence techniques, thus improving traditional drug discovery and development. However, most of DR computational methods fall short of taking into account the non-Euclidean nature of biomedical network data. To overcome this problem, a deep learning framework, namely DDAGDL, is proposed to predict drug-drug associations (DDAs) by using geometric deep learning (GDL) over heterogeneous information network (HIN). Incorporating complex biological information into the topological structure of HIN, DDAGDL effectively learns the smoothed representations of drugs and diseases with an attention mechanism. Experiment results demonstrate the superior performance of DDAGDL on three real-world datasets under 10-fold cross-validation when compared with state-of-the-art DR methods in terms of several evaluation metrics. Our case studies and molecular docking experiments indicate that DDAGDL is a promising DR tool that gains new insights into exploiting the geometric prior knowledge for improved efficacy.

SiSGC: A Drug Repositioning Prediction Model Based on Heterogeneous Simplifying Graph Convolution

“…As for the biological knowledge of diseases, it can be extracted from the Medical Subject Heading thesaurus. 38 The autoencoder model 39 is also utilized to reduce dimension to avoid the curse of dimensionality, the dimension of all initial features being set as 64, and we straightly used the processed features provided by Zhao et al 22 Feature Learning from a Heterogeneous Graph. Heterogeneous graphs have been broadly used to describe complex systems through a natural data structure.…”

“…19−21 Moreover, deep learning methods can overcome the limitation of low-level representation of machine-learning methods. 22,23 Yang used the graph embedding method of Mashup to extract drug and disease features from 15 networks, which are utilized to generate prediction by random forest. 24 SkipGNN model uses the node2vec algorithm to initialize the raw features, which are further fed into graph convolution networks to generate the improved features and make the prediction through the improved features.…”

“…Before SiSGC learns the features from the graph, biological knowledge information is used as the initial feature to narrow down the search space. Inspired by the work of Zhao et al, 22 we utilized a local-smoothing strategy 29 to achieve the adaptive selection of information diffuse distance, which can avoid feature overfitting. The similarity feature with SVD dimensionality reduction is fully considered as the compensation of multisource information to further improve the performance of our model.…”

“…The method of DDAGDL is proposed by Zhao et al to face the problem, but the straight use of denoised similarity information is neglected, which brings general performance to DDA prediction. 22 To address these issues, we propose a novel framework, SiSGC, for predicting DDAs. In detail, the heterogeneous graph containing drug, protein, and diseases is constructed.…”

“…However, the methods in this category have limited ability to capture higher-level features. With the increasing development of deep learning techniques in the fields of speech recognition and computer vision, many researchers in computational biology gradually pay attention to these techniques. − Moreover, deep learning methods can overcome the limitation of low-level representation of machine-learning methods. , Yang used the graph embedding method of Mashup to extract drug and disease features from 15 networks, which are utilized to generate prediction by random forest . SkipGNN model uses the node2vec algorithm to initialize the raw features, which are further fed into graph convolution networks to generate the improved features and make the prediction through the improved features .…”