English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Motivation: Machine learning based prediction of compound-proteininteractions (CPIs) is important for drug design, screening and repurposingstudies and can improve the efficiency and cost-effectiveness of wet labassays. Despite the publication of many research papers reporting CPIpredictors in the recent years, we have observed a number of fundamental issuesin experiment design that lead to over optimistic estimates of modelperformance. Results: In this paper, we analyze the impact of several importantfactors affecting generalization perfor-mance of CPI predictors that areoverlooked in existing work: 1. Similarity between training and test examplesin cross-validation 2. The strategy for generating negative examples, in theabsence of experimentally verified negative examples. 3. Choice of evaluationprotocols and performance metrics and their alignment with real-world use ofCPI predictors in screening large compound libraries. Using both an existingstate-of-the-art method (CPI-NN) and a proposed kernel based approach, we havefound that assessment of predictive performance of CPI predictors requirescareful con-trol over similarity between training and test examples. We alsoshow that random pairing for gen-erating synthetic negative examples fortraining and performance evaluation results in models with bettergeneralization performance in comparison to more sophisticated strategies usedin existing studies. Furthermore, we have found that our kernel based approach,despite its simple design, exceeds the prediction performance of CPI-NN. Wehave used the proposed model for compound screening of several proteinsincluding SARS-CoV-2 Spike and Human ACE2 proteins and found strong evidence insupport of its top hits. Availability: Code and raw experimental resultsavailable at https://github.com/adibayaseen/HKRCPI Contact:Fayyaz.minhas@warwick.ac.uk

Insights into performance evaluation of com-pound-protein interaction prediction methods