Benchmark_GASS module - README

This module include two separate experiments: NOS and TRYPSIN_like, as described below:

====== NOS experiment =====

data/ --- PDB list, extracted positions of train and test sites, and processed files for training and evaluating the NOS models

	PDB list of train and test datasets
		test_neg_PDBs.txt
		train_neg_PDBs.txt
		train_pos_PDBs.txt

	ptf/ :	Extracted positions of train and test sites
		
	numpy/ and pytables/ :	numpy arrays and pytables of extracted local protein boxes in voxel representation
		
code/ --- 
	
	Extract all sites of defined amino acid type from PDB structures
		get_all_res_for_defined_type_from_pdb_NOS_test.py

	Extracting test sites boxes
		cut_box_site_NOS_detect.py

	Store extracted examples as pytables
		store_pytable_NOS.py

	Definition and training and of 3DCNN models
		FSCNN_sulfur_NOS.py
		layers.py

	Evaluate test site probabily scores using trained 3DCNN models
		post_train_eval_FSCNN_NOS_detect.py

	Record representative probabilty score for each site and generate summary results files for NOS site detection
		get_NOS_fold_pos_pdb.py

	Evaluate AUC scores for classifying positive NOS structures against random structures at residue site and structure levels, respectively

		AUC_ROC.py

results/ ---

	detect_prob/ : Predicted probability scores of test sites

	detect_results/ : Summary of detection results using probability scores from detect_prob

		Final summary of detection results are written in 
			detection_summary_true_sites.txt 
				Each row in the results file corresponding to a true site annotated in CSA, the final column record our detection result of the site -- "True" represents a successful detection and "False" represents an unsuccessful detection
			detection_summary_fp_sites_neg_pdb.txt
			detection_summary_fp_sites_pos_pdb.txt

	weights/ : Trained model weights 


====== TRYPSIN_like ====== 

data/ --- PDB files, extracted positions of test sites (ptf files) and processed files for evaluating the TRYPSIN_like models

	pdb/: PDB files of TRYPSIN_like enzymes
	ptf/: Extracted positions of test sites
	numpy/: numpy arrays and pytables of extracted local protein boxes in voxel representation

	SCOP PDB list and CSA annotation of TRYPSIN_like families:                         
		SCOP_PDB.txt
		csa_pdb_HIS_SER_annotation.txt 
		SCOP_PDB_csa_key_res.txt

code/
	
	Extracting test sites boxes
		cut_box_site_TRYPSIN.py

	Evaluate test site probabily scores using trained 3DCNN models
		layers.py
		eval_prob_score_numpy.py

	Record representative probabilty score for each site and generate summary results files for TRYPSIN_like site detection
		eval_TRYPSIN_detection.py


results/

	Summary files of performances detecting TRYPSIN_like sites in 1,447 enzymes. Final detection results are summarized in
		
		detected_TRYPSIN_CSA_site_summary.txt

			Each row in the results file corresponding to a true site annotated in CSA, the final column record our detection result of the site -- "True" represents a successful detection and "False" represents an unsuccessful detection

		detected_TRYPSIN_nonCSA_site_summary.txt

			Sites detected by our models but not annotated in CSA. Literature confirmed that these are true TRYPSIN_like sites.
		

	/post_prob: Predicted probability scores of test sites
	/weights: Trained model weights of TRYPSIN_SER and TRYPSIN_HIS models copied from the PROSITE module