Drug‐Drug Interaction Discovery: Kernel Learning from Heterogeneous Similarities

Abstract

We develop a pipeline to mine complex drug interactions by combining different similarities and interaction types (molecular, structural, phenotypic, genomic etc). Our goal is to learn an optimal kernel from these heterogeneous similarities in a supervised manner. We formulate an extensible framework that can easily integrate new interaction types into a rich model. The core of our pipeline features a novel kernel-learning approach that tunes the weights of the heterogeneous similarities, and fuses them into a Similarity-based Kernel for Identifying Drug-Drug interactions and Discovery, or SKID3. Experimental evaluation on the DrugBank database shows that SKID3 effectively combines similarities generated from chemical reaction pathways (which generally improve precision) and molecular and structural fingerprints (which generally improve recall) into a single kernel that gets the best of both worlds, and consequently demonstrates the best performance.

Similarity-based machine learning support vector machine predictor of drug-drug interactions with improved accuracies

“…In terms of the resources forin silico DDI extraction or prediction, they commonly consist of textual and structural data, in which textual data can refer to the literatures, electronic health records (EHRs) or comments in social media, while chemical, molecular and pharmacological properties are included in structural data . However, the former is more likely to be used for detecting ADEs from the reported literatures, EHRs or comments in social media .…”

“…Similarity‐based methods can also be applied in a well‐developed kernel method, which makes it promising for combining similarities and features to mimic the interactions between drug pairs. For example, Devendra et al developed a novel kernel‐learning approach using the different similarities between molecules, structures, phenotype and genomics to mine DDI, which proved to be effective . Instead of employing drug structures in DDI prediction, Ferdousi et al constructed a method based on drug functional similarities consisting of carriers, transporters, enzymes and targets .…”

“…For example, Devendra et al developed a novel kernel-learning approach using the different similarities between molecules, structures, phenotype and genomics to mine DDI, which proved to be effective. 4 Instead of employing drug structures in DDI prediction, Ferdousi et al constructed a method based on drug functional similarities consisting of carriers, transporters, enzymes and targets. 12 Ruifeng et al built large-scale drug-protein interaction profiles with protein-chemical linked data from the STITCH database to construct the prediction models for drug pairs that are likely to induce adverse drug reactions via synergistic DDI, and then evaluated the performance using the TWOSIDES database.…”

A Machine Learning Framework for Predicting Drug-drug Interactions

“…Existing computational methods could be roughly classified into three categories, i.e. similarity-based methods [5][6][7][8], networks-based methods [9][10][11][12][13] and machine learning methods [14][15][16][17][18][19][20][21][22]. Similarity-based methods directly calculate similarity scores between drug profiles to infer DDIs.…”

“…In the last decades, machine learning has attracted increasing attention in inferring drug-drug interactions [14][15][16][17][18][19][20][21][22]. Most of the machine learning methods focus on data integration to increase the accuracy of DDI prediction.…”