Identifying a possibility of alternative drug targets for plasma proteins using the database of biologically-informed causal estimates

Valeriia Haberland [1], Benjamin Elsworth [1], Yi Liu [1], Jie Zheng [1], Pau Erola [1], Denis Baird [1], Matt Lyon [1], Philip Haycock [1], Gibran Hemani [1], Tom R. Gaunt [1] 1 MRC Integrative Epidemiology Unit (IEU), Bristol Medical School, University of Bristol, Oakfield House, Oakfield Grove, Bristol, BS8 2BN, UK

Word limit: 2000 characters

Keywords: Mendelian Randomization, protein-protein interactions, alternative drug target, graph-based database

The clinical trial success rates are traditionally quite low e.g., some studies cite the rates between around 2% and somewhat 8%, depending on the disease area. It has been mentioned in many sources that Mendelian Randomization (MR) analysis, which estimates the genetically-informed causality, has since contributed to improving the success rates. The problem that affects the low success rates is also confided in the biological complexities that sometimes prevent a strongly identified drug target to be used by drug. Hence, the alternative drug targets, although potentially less influential, could be a solution in such scenarios. In this work, we attempt to estimate whether the success rate could increase if we account for the protein-protein interactions (PPI), assuming proteins to be our main focus as a drug target. For this purpose, we extract precomputed MR estimates and PPI information from the graph-based database EpiGraphDB ( which combines data from multiple publicly available sources, for example, IntAct and STRING PPI data. Here, we compare the MR results with regard to their significance and effect size between the main identified target (protein) and potentially alternative target which interact with each other for a specific disease or risk factor. Then, we also compare these results to the MR results of the randomly chosen sample of the unrelated proteins for the same disease or risk factor. Our preliminary results show promising potential that we plan to expand towards the more complex PPI network-related models.