View Your Team's Submissions
Enter your token to view all your submissions with scores and private notes.
However, submissions can still be made. The limit on the number of submissions was removed. Questions regarding the challange can be sent to discord channel
DYRK1B (dual-specificity tyrosine-phosphorylation-regulated kinase 1B, UniProt Q9Y463) is a serine/threonine kinase involved in the regulation of cell cycle progression, cellular quiescence, metabolism, and stress responses. It has emerged as a therapeutically relevant target in multiple diseases.
In cancer, including pancreatic, ovarian, and colorectal malignancies, DYRK1B supports tumor cell survival and therapy resistance, making its inhibition a promising strategy to enhance chemosensitivity. DYRK1B has also been implicated in metabolic disorders such as insulin resistance and obesity, as well as in fibrotic and inflammatory pathways.
Participants may work individually or form teams of up to three members. Each team must register to obtain a submission token. Please use valid email addresses, as participants may be contacted during post-challenge evaluation.
A library of 3,365 structures is provided, including several validated DYRK1B inhibitors with IC50 values below 1 µM.
Your task is to select exactly 100 structures that are most likely to be active DYRK1B inhibitors. The primary goal is to maximize the number of true active hits.
Participants may use any computational strategy, including predictive modeling and virtual screening, to prioritize candidates.
A single X-ray crystal structure of DYRK1B in complex with the inhibitor AZ191 (PDB ID: 8C2Z) is currently available. The protein was prepared in a docking-ready format compatible with Vina, Smina, and Gnina (e.g. available in EasyDock) and in the source format, containing all hydrogens and protonation states (may be used in other programs or to calculate protein-ligand fingerprints).
The docking grid was centered on the reference ligand AZ191, with an additional 6 Å margin. The grid size may require adjustment for larger ligands.
Compounds were extracted from the latest ChEMBL release (CHEMBL5543). Salts were removed and canonical SMILES were generated to identify duplicates and aggregate activity values.
Stereochemistry was not enforced, so stereoisomers and undefined stereocenters may coexist in the dataset.
The archive contains SMILES and SDF files, grouped by activity type (IC50, Ki, Kd, inhibition). Datasets may overlap.
| Field | Description |
|---|---|
| Molecule ChEMBL ID | Structure identifier |
| Standard Type | Activity type within the dataset |
| activity | Average activity value for identical structures |
| Standard Units | Units of activity measurement |
| activity_sd | Standard deviation (NA if single measurement) |
| p_activity | Log-scaled activity value for QSAR modeling |
Files chembl_all_class.sdf / chembl_all_class.smi:
Compounds were labeled active if IC50/Ki/Kd ≤ 1 µM or inhibition ≥ 90%. Compounds with conflicting activity annotations were removed. This dataset is suitable for classification modeling.
| Field | Description |
|---|---|
| Molecule ChEMBL ID | Structure identifier |
| activity_class | Binary label (0 = inactive, 1 = active) |
This file contains precomputed LigandScout databases for actives and inactives as they were defined above. They were generated by the method icon-fast (up to 25 conformers for each compound). For compounds with missing stereoconfigurations LigandScout generated some random configurations.
The blind dataset contains 3,365 structures in multiple formats. Some molecules may have undefined stereochemistry, which should be considered when applying 3D-based methods.
Participants may use any publicly available data sources and software tools, including proprietary software, to support model development and screening.
Each submission must contain exactly 100 unique compound IDs. IDs may be uploaded as a text file or pasted directly into the submission field, with one ID per line. If fewer than 100 compounds are selected, remaining lines may be filled with random IDs.
Rankings are based on the proportion of correctly identified active compounds (recall). Each team may submit up to 10 submissions; only the best result will be considered.
The leaderboard is updated only after submissions. Use the Refresh button to view the latest results.
Teams are ranked by score, with submission time used as a tie-breaker. On-site and online teams are evaluated separately.
The top three teams in each category will receive RowanSci credits and will be invited to give short presentations describing their solutions.
Register your team members (1-3 people) to receive a team assignment and token.
Enter your token to view all your submissions with scores and private notes.