Multi-omics data harmonisation for the discovery of COVID-19 drug targets

Background: COVID-19 pandemic has caused a high number of deaths globally. Despite the volume of experiments performed, biology of the virus is not yet fully understood. Translatomics and proteomics allow users to capture large quantities of high dimensional SARS-CoV-2 data and it is possible to extract valuable information by analysing each omics data separately. However, integrating multiple modalities can yield information hidden in individual omics analyses, while enhancing robustness and reproducibility. In a review we showed that existing multi-omics approaches are vulnerable to information loss or restricted to specific experimental set-ups, supporting our hypothesis.  Methods: In contrast to existing approaches, our pipeline unified proteome and translatome data at individual protein levels, ensuring minimum information loss during integration. Using a multivariate approach, we accounted for the high dimensional structure of the data, and identified features in each omics data which discriminate strongly between biological classes. We further exploited this subset of features to identify strong correlations within and between omics data types. From these strongly correlated features, we then identified potentially medically relevant biological targets for drug development. To assess their association with the virus, we compared our findings against publicly available datasets. An ensemble of computational docking analyses was also carried out for all identified drug-target combinations. To further narrow down this list, a pharmacokinetic analysis was used to assess drug efficiency and toxicity in humans and results were ranked. Results: We report 54 candidate drug targets, with 51 passing the pharmacokinetics criteria. Amsacrine, ceritinib and crizotinib that target DNA topoisomerase2-alpha and tyrosine kinases return the highest molecular docking score across four separate algorithms. Thus, using a multi-omics harmonic approach we report a high confidence list of computationally screened drugs for further biological validation. Our pipeline is open-access and generalisable to future datasets with two or more modalities.