
Welcome to our compound-kinase activity prediction platform! This application allows you to predict the probability of inhibition for small molecules, represented by SMILES strings, against a broad range of protein kinases.
The predictions are powered by two multitask deep neural network (MTDNN) models trained using large-scale bioactivity data with over 1 million human kinase bioactivity annotations and more than 400,000 unique small molecules from databases ChEMBL and the Kinase Knowledge Base (KKB):
- Kinase-Cutoff6: This model covers 406 kinases and classifies molecules as active or inactive based on a pActivity cutoff value of 6. It provides wide kinase coverage, offering broad insights across many targets.
- Kinase-Cutoff7: This model covers 328 kinases and classifies molecules using a stricter pActivity cutoff value of 7. While the kinase coverage is smaller, it provides higher accuracy and more reliable predictions. It is ideal for users seeking precise inhibition probabilities for a refined set of kinases.