Project description
Connecting research and collaborative filtering for drug screening
Drug development pipelines incur high expenses, possess short lifespans and exhibit elevated failure rates. As such, drug repurposing, involving the systematic screening of existing compounds to uncover novel therapeutic uses, is a solution. Collaborative filtering leverages established drug-disease associations to formulate fresh recommendations. However, prior investigations lack emphasis on human supervision and exhibit resistance to biological data incorporation. With this in mind, the MSCA-funded RECeSS project aims to bridge the gap between drug research and collaborative filtering through the use of a classifier. This classifier addresses the issue of class imbalance within drug-disease pairings and the presence of incomplete drug/disease features using a semi-supervised learning approach. It establishes connections between projected matches and perturbed biological pathways via enrichment analyses.
Objective
In 2021, drug development pipelines last 10 years in average, and cost around $2 billion, while facing high failure rates, as only around 10% of Phase 0 drug candidates reach the commercialization stage. These issues can be mitigated through drug repurposing, where existent compounds are systematically screened for new therapeutic indications. Collaborative filtering is a semi-supervised learning framework that leverages known drug-disease matchings to make novel recommendations. However, prior works cannot be leveraged because of their lack of focus on human oversight and robustness to biological data.
This project aims at bridging the gap between drug research and collaborative filtering by implementing a RECeSS classifier, that is
(1) Robust: deals with class imbalance in drug-disease matchings, and missing drug/disease features, by semi-supervised learning;
(2) Explainable: connects predicted matchings to perturbed biological pathways through enrichment analyses, based on the learnt importance of features in the model;
(3) Controllable: guarantees a bound on the false positive rate using an adaptive learning scheme;
(4) Standard: algorithms are trained and tested by a standardized open-source pipeline.
Predicted matchings will be independently validated by structure-based methods. This innovative interdisciplinary project relies on a solid basis of newly curated data (up to 1,386 drugs, 1,599 diseases, 12 feature types). It is primarily supervised by Pr. Olaf Wolkenhauer, at SBI Rostock, whose team has an expertise in drug repurposing, in systems biology and data imbalance in machine learning. This project will help the fellow develop new skills, and enhance her professional maturity in academia.
In the short term, this would yield the first method that fully integrates biological interpretation and risk assessment to collaborative filtering-based repurposing. Long-term outcomes might help define sustainable and transparent drug development for rare diseases.
Fields of science
Programme(s)
- HORIZON.1.2 - Marie Skłodowska-Curie Actions (MSCA) Main Programme
Funding Scheme
HORIZON-TMA-MSCA-PF-EF - HORIZON TMA MSCA Postdoctoral Fellowships - European FellowshipsCoordinator
18055 Rostock
Germany