experiment_prediction

Predicting Novel Binding Pairs with E3FP

The goal of this analysis is to make novel compound-target binding predictions using SEA with E3FP that cannot be made using SEA with the 2D fingerprint ECFP.

Relevant Data Files

• chembl20_zinc_in-stock_mw800_unique_proto.smi.bz2: SMILES file for all "in-stock" compounds in the ZINC database, as of 2015-09-24, mapped to ChEMBL20
• available_targets_pdsp_chembl20_map.tab.bz2: Table of human, mouse, and rat analogs to protein targets for which NIMH PDSP has a binding assay, mapped to ChEMBL20 target IDs.
• early_params.cfg: Parameters chosen early into the optimization process that were originally used for this analysis.

Generating Fingerprints

See README instructions within drug_selection and target_selection for ECFP4 and E3FP fingerprint generation.

Searching Molecules against Targets

See target_drug_search.

Computing All Pairwise Tanimoto Coefficients (TCs)

To compute all pairwise TCs between target-associated molecules and drug molecules for either E3FP or ECFP4, run

python ../fingerprint_comparison/get_pairwise_tcs.py <drug_molecules> <target_molecules> --memmap_file <mmap_file>

Then, to count the number of TCs pairs that map to specific values at a specified precision

python ../fingerprint_comparison/count_pairwise_tcs.py <ecfp4_mmap_file> <e3fp_mmap_file> --names ECFP4 E3FP

which produces an output file ecfp4_e3fp_tcs.csv.gz.

Computing Pairwise TCs Between Specific Target/Molecule Pairs

To compute pairwise max TCs for a specific drug/target pair, run:

python get_mol_vs_target_tcs.py <mol_name> <target_id> <drug_molecules> <target_molecules> <target_targets> --affinity <affinity> --out_file <out_file>