Motivation
QCRI deep learning models for crystallization propensity prediction, DeepCrystal and B, BCrystal is ready to compute.
How to get started?
Perform the following steps to signup:
1. Navigate to the Sign Up tab in the top navigation bar.
2. Fill in your details and press register, a registration confirmation mail will be sent, you will need to click the confirmation link before you will be able to log in.
3. Navigate to the sign in page and sign in using the username and password you selected.
4. If you’re not sure how to proceed from here then check out the Help page from the top navigation bar. In short, when signed in there will be a tab in the top bar labeled “Run”, select either BCrystal or DeepCrystal or DeepSol or DeepRepurpose (compound-viral protein activity predictor) from this dropdown. BCrystal and DeepCrystal only require you to input a fasta file of the sequences you want to predict for, less than 1001 sequences per job for DeepCrystal and less than 6 per job for BCrystal. BCrystal is better, but slower as it utilizes many features derived from the sequence in the model, homology based features and these take time to generate. DeepSol also requires you to input a fasta file and provide a no 1,2 or 3 for DeepSol1, DeepSol2 or DeepSol3. DeepSol1 is based only on sequence and is fast whereas DeepSol2 and DeepSol3 extract additional features based on homology and are usually slower. DeepRepurpose is our latest consensus framework of in-silico sequence-based compound-viral protein activity predictor. It requires 2 input files, a first file containing Uniprot_Id and viral protein sequence in a comma separated file with header (UniProtId,Sequence). The second file is the list of compounds for which the activity scores are to be predicted containing Inchikey and canonical smiles representation of compounds in a comma separated file with header (Inchikey,Canonical_SMILES).