########################### #### DATA PREPARATION ##### ########################### 1. Create input for protein binding sites Input: PDB file and [co-crystalized ligand or list of key residues] Method I. Given PDB file and a co-crystalized ligand - Generate DSSP file using the DSSP program - Use PocketFEATURE package to generate the ptf and ff files a.) Download PocketFEATURE at https://simtk.org/projects/pocketfeature b.) Use script GenerateCavityPoint_Vectorize.pl in the PocketFEATURE package to generate the ptf and ff files Example: ./GenerateCavityPoint_Vectorize.pl list1.txt 6 (See README.txt in PocketFEATURE for more information) Method II. Given PDB file and a list of key residues - Generate DSSP file using the DSSP program - Use generate_ptf_from_key_res.py to generate the ptf file Example: python generate_ptf_from_key_res.py PDB_ID key_res_list - Run FEATURE program to generate the ff file featurize -P 1xxx_XXX.ptf > 1xxx_XXX.ff 2. Create input for small molecule Input - SMILES string for the ligand ############################ ######## TRAIN MODELS ###### ############################ 1. DUDE cross validation THEANO_FLAGS=floatX=float32 python graph_cnn_DUDE_CV_neg_pockets.py train fold 2. DUDE full model THEANO_FLAGS=floatX=float32 python graph_cnn_DUDE_all_folds_neg_assay.py train ############################ ######## TEST MODELS ####### ############################ 1. DUDE cross validation - Evaluate ROC on DUDE THEANO_FLAGS=floatX=float32 python graph_cnn_DUDE_CV_neg_pockets.py ROC fold 2. DUDE full model - Evaluate ROC on MUV THEANO_FLAGS=floatX=float32 python graph_cnn_DUDE_all_folds_neg_assay.py ROC - User defined tests: Evaluate input pocket against predefined ligand database THEANO_FLAGS=floatX=float32 python graph_cnn_DUDE_all_folds_neg_assay.py user 1xxx_XXX.ff # where 1xxx_XXX.ff is the user defined pocket name and locates at ./data/User/ff/ ############################## ######## OBTAIN POC FP ####### ############################## 1. Autoencoder pretrained POC FP THEANO_FLAGS=floatX=float32 python poc_autoencoder_II.py eval # Place user defined pockets in ./data/User/ff # see the in-line comments for user defined pockets 2. Binding classifier fine-tuned POC FP THEANO_FLAGS=floatX=float32 python graph_cnn_DUDE_all_folds_eval_poc_fp.py user # Place user defined pockets in ./data/User/ff # see the in-line comments for user defined pockets