EGFR activity model
EGFR is a receptor tyrosine kinase that relays cell growth signals. Activating mutations drive non-small cell lung cancer, and drugs such as erlotinib, gefitinib and osimertinib bind the ATP site of its kinase domain.
A model trained on measured activity data for EGFR (Oncology). Give it compounds and it predicts their IC50 against EGFR, so you can rank candidates by it.
Predict with this model
- UniProt
- P00533
- Reference structure
- PDB 1M17, ligand AQ4
- Training data
- TSAP (Boltz-2): 58 compounds, IC50TSAP (AI-QSAR): 66 compounds, IC50
- Validation
- Predicted and measured values are compared on compounds held out of training.
- Cost
- From 100 cr, quoted before you submit
- Model version
- TSAP (Boltz-2): TSAP pipeline v2.2TSAP (AI-QSAR): Ridge regression on docking features
Validation results
Predicted and measured values compared on compounds held out of training. RMSE is in log units (pIC50 and similar).
TSAP (Boltz-2)
Pearson r
0.90
R²
0.69
RMSE (log units)
0.71
Holdout compounds
17
TSAP (AI-QSAR)
Pearson r
0.90
Square root of RSQ
RSQ
0.80
Holdout compounds
17
How to use it
Under Predict with this model, press TSAP (Boltz-2) or TSAP (AI-QSAR). The wizard of the workflow you pressed opens with the EGFR model already selected.
Upload the SMILES of the compounds to predict as a .smi, .csv or .txt file.
Check the estimated credits and time, then submit.
The result is a table of compounds ranked by predicted IC50.
Limitations
- Predictions can be less accurate for compounds whose structures differ a lot from the training data.
- Validation used only about 17 compounds, so performance on new compounds can differ from these metrics.
- Predictions do not replace experiments. Use them to decide what to synthesize and measure first.