Omics technologies have the potential to revolutionize our understanding of individual patients and their response to medical treatments. However, the complexity of omics datasets from large-scale patient cohorts makes integrative analyses and data interpretation highly challenging.
In a new study published in Nature Biotechnology, a team of European scientists explored deep-learning models to solve this challenge for discovering drug-omics associations in type 2 diabetes. Their deep-learning-based framework, multi-omics variational autoencoders (MOVE), was applied to a cohort of 789 individuals newly diagnosed type 2 diabetes, from the DIRECT consortium. Using in silico perturbations, they identified drug-omics associations across the patient cohort for the 20 most prevalent drugs used to treat type 2 diabetes. Compared to univariate statistics, this method enabled identification of drug-omics interactions with significant higher sensitivity.
Result highlights included identification of novel associations between metformin and the gut microbiota and opposite molecular responses for the two statins simvastatin and atorvastatin.
The cohort size was limited, but with larger cohorts MOVE has the potential to identify molecular associations and treatment outcomes for individual patients, paving the way towards personalized medicine.
Congratulations to all authors, including our colleagues Helle Krogh Pedersen, Josef Korbinian Vogt and Joachim Johansen, for being part of this groundbreaking publication!