The repository contains the evaluation workflow for the first DREAM Target 2035 Drug Discovery Challenge. Target 2035 is an open-science global movement consisting of international scientists and researchers, focusing on the creation of chemical and biological tools to study human proteins and inform drug discovery.
The success of Target 2035 relies on future breakthroughs in machine learning (ML) for drug discovery. To that end, the SGC and its industry partners are populating the AIRCHECK platform with training data and (soon) best performing ML models.
The challenge is split into 3 "steps":
- Step 1: participants submit a 4-column CSV file
- Step 2: participants submit a 4-column CSV file
- Step 3: top 5 teams from Step 1 and top 20-25 teams from Step 2 are invited to participate in Step 3 (more details coming soon)
Metrics returned and used for ranking are:
- Number of clusters per selection label
- Number of hits per selection label
- Cluster PR-AUC per selection label
Additional metrics returned but not used for ranking:
- AUROC
- AUC-PR
Scripts for validation and scoring are vailable under ./evaluation