Evaluating the robustness of connectivity methods to noise for in silico drug repurposing studies

Tham, Nevin and Langley, Sarah R. ORCID: https://orcid.org/0000-0003-4419-476X 2022. Evaluating the robustness of connectivity methods to noise for in silico drug repurposing studies. Frontiers in Systems Biology 2 , 1050730. 10.3389/fsysb.2022.1050730

PDF - Published Version Available under License Creative Commons Attribution. Download (2MB)

Abstract

Drug repurposing is an approach to identify new therapeutic applications for existing drugs and small molecules. It is a field of growing research interest due to its time and cost effectiveness as compared with de novo drug discovery. One method for drug repurposing is to adopt a systems biology approach to associate molecular ‘signatures’ of drug and disease. Drugs which have an inverse relationship with the disease signature may be able to reverse the molecular effects of the disease and thus be candidates for repurposing. Conversely, drugs which mimic the disease signatures can inform on potential molecular mechanisms of disease. The relationship between these disease and drug signatures are quantified through connectivity scores. Identifying a suitable drug-disease scoring method is key for in silico drug repurposing, so as to obtain an accurate representation of the true drug-disease relationship. There are several methods to calculate these connectivity scores, notably the Kolmogorov-Smirnov (KS), Zhang and eXtreme Sum (XSum). However, these methods can provide discordant estimations of the drug-disease relationship, and this discordance can affect the drug-disease indication. Using the gene expression profiles from the Library of Integrated Network-Based Cellular Signatures (LINCS) database, we evaluated the methods based on their drug-disease connectivity scoring performance. In this first-of-its-kind analysis, we varied the quality of disease signatures by using only highly differential genes or by the inclusion of non-differential genes. Further, we simulated noisy disease signatures by introducing varying levels of noise into the gene expression signatures. Overall, we found that there was not one method that outperformed the others in all instances, but the Zhang method performs well in a majority of our analyses. Our results provide a framework to evaluate connectivity scoring methods, and considerations for deciding which scoring method to apply in future systems biology studies for drug repurposing.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Biosciences
Publisher:	Frontiers Media
ISSN:	2674-0702
Date of First Compliant Deposit:	24 August 2023
Date of Acceptance:	21 November 2022
Last Modified:	25 Aug 2023 06:18
URI:	https://orca.cardiff.ac.uk/id/eprint/161841

Actions (repository staff only)

Edit Item