#Graphical Models for Identifying Pore-forming Proteins This repository provides codes for paper Nan Xu, Theodore W. Kahn, Theju Jacob, Yan Liu, Graphical Models for Identifying Pore-forming Protein.

We put the extracted proteins in ./ExtractedProteins. You can reproduce our experiments with the following codes, with the assumption that we are at location: ./project_package.

Requirements

This code package was tested with python 3.7.3 on Linux. For other library dependencies, you can

1. First install Anaconda
2. Refer to environment.yaml and install in terminal with Anaconda:

   conda env create -f environment.yaml
   

3. Download the UniProt20 database from the following link (around 12G, takes about 30m to download):

   http://wwwuser.gwdg.de/~compbiol/data/hhsuite/databases/hhsuite_dbs/old-releases/uniprot20_2016_02.tgz
   

   1. uncompress zip file
   2. rename the folder to uniprot20_2016_02, ensure all files in this folder except md5sum have prefix uniprot20_2016_02
   3. move the folder to packages/TGT_Package/databases

Run Experiment

1. Activate the environment:

   conda activate SCRFs
   

2. Change parameters in config.py

   The table below provides an overview of the parameters:

   Name	Type	Description	Default	Suggestion
   blastp_cmd_path	string	Path of Blastp package; Compute sequence identity; Details: https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE_TYPE=BlastDocs&DOC_TYPE=Download	'./packages/ncbi-blast-2.10.1+/bin/blastp'	DO NOT Change
   TGT_pkg_path	string	Path of TGT package; Run HHblits to generate A3M; Details: https://github.com/realbigws/TGT_Package	'./packages/TGT_Package'	DO NOT Change; Ensure database UniProt20 downloaded
   Predict_Property_path	string	Path of Predict Property package; Predict 3- and 8-digit secondary structures with TGT format protein profile; Details: https://github.com/realbigws/Predict_Property	'./packages/Predict_Property'	DO NOT Change
   records_dir	string	Storage path of results	'./records'	Either relative (like the default) or absolute path should be fine.
   sub_records_dir	string	Subfolder path under records_dir	''	DO NOT Change unless for debug use
   task_type	list of strings	Definition of tasks: selection for searching proteins with similar structure to group 1/2/3; filtering for combining results from different fasta files for group 1/2/3	['selection', 'filtering']	DO NOT Change
   group_list	list of char	Definition of group index: we studied three groups of training PFT proteins with similar structures in this project	['1', '2', '3']	DO NOT Change
   cur_task	list of strings	Tasks to perform according to task_type definition	['selection', 'filtering']	Normally use the default; Or ['selection']; Or ['filtering']
   selection_parameters	dictionary	Parameters for task selection	-	-
   		group_index_list: list of group index to consider	['1', '2', '3']	Normally use the default; Or ['1'], ['2'], ['3']; Or combination of any two group indices
   		source_folder: source folder of fasta files	'records/records_20201219'	Use your own folder path (either relative or absolute); NOTE: only proteins in files ending with '.fasta' will be considered
   		target_folder: target folder of results	'selection'	Normally use the default or specify your relative path; NOTE: absolute path is not supported here
   		seqlen_limit: list of integer, first is the allowed minimum sequence length, and second is the allowed maximum length	[50, 2000]	Use your own sequence limitation values
   		pos_seq_identity_threshold: float, testing proteins should be with sequence identity to training PFT proteins in each group smaller this value	0.3	Use your own sequence identity threshold
   		PFTs_prob_threshold: float, probability of being pore-forming toxins from ProtCNN model for testing proteins should be higher than this value	0.5	Use your own probability threshold (higher than 0.5)
   		mutual_seq_identity_threshold: float, remove selected proteins with mutual sequence identity higher than this value to avoid repetition	1	Use your own probability threshold (positive value not beyond 1)
   		batch_size: integer, batch size for ProtCNN model	100	Use your own batch size; NOTE: smaller than 1000 if gpu is used and the memory is around 12G
   		num_threads: integer, threads that run parallel to speed up computing	10	Use your own threads number; NOTE: observe your cpu and memory utilization, decrease the threads number if the server is overloaded
   		device: string, device to run ProtCNN model	'cpu'	Use your own device; NOTE: determine the device index if gpu is used, e.g., 'cuda:0' or 'cuda:1'
   filtering_parameters	dictionary	parameters for task filtering	-	-
   		source_folder: source folder of results (files ends with '_summary.csv') computed from selection task	'selection'	Normally use the target_folder in selection_parameters
   		target_folder: target folder of results	'filtering'	Normally use the default or specify your relative path; NOTE: absolute path is not supported here; intra-group results stored in 'group_x_summary.csv' and inter-group results stored in 'dataset_summary.csv'
   		mutual_seq_identity_threshold: float, remove selected proteins with mutual sequence identity higher than this value to avoid repetition	1	Use your own probability threshold (positive value not beyond 1)
   PosTest_threshold	dictionary	threshold for structure prediction scores: keep the testing protein if its SCRFs score is higher than threshold	2.72 for group_1, 424.75 for group_2, 145.38 for group_3	DO NOT Change

3. run main.py and log stdout to log.txt

   python main.py >> logThe below commands assume we are at location: `./project_package`.
   

Requirements

This code package was tested with python 3.7.3 on Linux. For other library dependencies, you can

1. First install Anaconda
2. Refer to environment.yaml and install in terminal with Anaconda:

   conda env create -f environment.yaml
   

3. Download the UniProt20 database from the following link (around 12G, takes about 30m to download):

   http://wwwuser.gwdg.de/~compbiol/data/hhsuite/databases/hhsuite_dbs/old-releases/uniprot20_2016_02.tgz
   

   1. uncompress zip file
   2. rename the folder to uniprot20_2016_02, ensure all files in this folder except md5sum have prefix uniprot20_2016_02
   3. move the folder to packages/TGT_Package/databases

Run Experiment

1. Activate the environment:

   conda activate SCRFs
   

2. Change parameters in config.py

   The table below provides an overview of the parameters:

   Name	Type	Description	Default	Suggestion
   blastp_cmd_path	string	Path of Blastp package; Compute sequence identity; Details: https://blast.ncbi.nlm.nih.gov/Blast.cgi?PAGE_TYPE=BlastDocs&DOC_TYPE=Download	'./packages/ncbi-blast-2.10.1+/bin/blastp'	DO NOT Change
   TGT_pkg_path	string	Path of TGT package; Run HHblits to generate A3M; Details: https://github.com/realbigws/TGT_Package	'./packages/TGT_Package'	DO NOT Change; Ensure database UniProt20 downloaded
   Predict_Property_path	string	Path of Predict Property package; Predict 3- and 8-digit secondary structures with TGT format protein profile; Details: https://github.com/realbigws/Predict_Property	'./packages/Predict_Property'	DO NOT Change
   records_dir	string	Storage path of results	'./records'	Either relative (like the default) or absolute path should be fine.
   sub_records_dir	string	Subfolder path under records_dir	''	DO NOT Change unless for debug use
   task_type	list of strings	Definition of tasks: selection for searching proteins with similar structure to group1/2/3; filtering for combining results from different fasta files for group1/2/3	['selection', 'filtering']	DO NOT Change
   group_list	list of char	Definition of group index: we have three groups of proteins with similar structures in this project	['1', '2', '3']	DO NOT Change
   cur_task	list of strings	Tasks to perform according to task_type definition	['selection', 'filtering']	Normally use the default; Or ['selection']; Or ['filtering']
   selection_parameters	dictionary	Parameters for task selection	-	-
   		group_index_list: list of group index to consider	['1', '2', '3']	Normally use the default; Or ['1'], ['2'], ['3']; Or combination of any two group indices
   		source_folder: source folder of fasta files	'records/records_20201219'	Use your own folder path (either relative or absolute); NOTE: only proteins in files ending with '.fasta' will be considered
   		target_folder: target folder of results	'selection'	Normally use the default or specify your relative path; NOTE: absolute path is not supported here
   		seqlen_limit: list of integer, first is the allowed minimum sequence length, and second is the allowed maximum length	[50, 2000]	Use your own sequence limitation values
   		pos_seq_identity_threshold: float, testing proteins should be with sequence identity to positive proteins in each group smaller this value	0.3	Use your own sequence identity threshold
   		PFTs_prob_threshold: float, probability of being Pore-forming toxins from ProtCNN model for testing proteins should be higher than this value	0.5	Use your own probability threshold (higher than 0.5)
   		mutual_seq_identity_threshold: float, remove selected proteins with mutual sequence identity higher than this value to avoid repetition	1	Use your own probability threshold (positive value not beyond 1)
   		batch_size: integer, batch size for ProtCNN model	100	Use your own batch size; NOTE: smaller than 1000 if gpu is used and the memory is around 12G
   		num_threads: integer, threads that run parallel to speed up computing	10	Use your own threads number; NOTE: observe your cpu and memory utilization, decrease the threads number if the server is overloaded
   		device: string, device to run ProtCNN model	'cpu'	Use your own device; NOTE: determine the device index if gpu is used, e.g., 'cuda:0' or 'cuda:1'
   filtering_parameters	dictionary	parameters for task filtering	-	-
   		source_folder: source folder of results (files ends with '_summary.csv') computed from selection task	'selection'	Normally use the target_folder in selection_parameters
   		target_folder: target folder of results	'filtering'	Normally use the default or specify your relative path; NOTE: absolute path is not supported here; intra-group results stored in 'group_x_summary.csv' and inter-group results stored in 'dataset_summary.csv'
   		mutual_seq_identity_threshold: float, remove selected proteins with mutual sequence identity higher than this value to avoid repetition	1	Use your own probability threshold (positive value not beyond 1)
   PosTest_threshold	dictionary	threshold for structure prediction scores: keep the testing protein if its SCRFs score is higher than threshold	2.72 for group_1, 424.75 for group_2, 145.38 for group_3	DO NOT Change

3. run main.py and log stdout to log.txt

   python main.py > log.txt 2>&1
   

Toy Results

The fasta file records/records_20201219/Crickmore_CryProteins.txt is split into 10 files ending with .fasta. If we use the default parameters, i.e., the config.py is not modified, we are expecting to have results in records/2021-01-xx-xx-xx-xx (the folder with latest date)

1. group summaries: records/2021-01-07-14-21-31/filtering/group_1_summary.csv, proteins in each file sorted by SCRFs Score. No testing proteins are predicted to have similar structure as PFTs in group 2 and group 3. Take the group_1_summary.csv as an example:

   	ID	PFT Probability	SCRFs Score	Sequence
   0	Cry21Ga1	0.9998382329940796	3.79	MADLSN...
   1	Cry32Ga1	0.9999923706054688	2.73	MYQNYN...

2. dataset summaries: records/2021-01-07-14-21-31/filtering/dataset_summary.csv with proteins sorted by PFT Probability. The summary file looks as follows:

   	ID	PFT Probability	Sequence
   0	Cry32Ga1	0.9999923706054688	MYQNYN...
   1	Cry21Ga1	0.9998382329940796	MADLSN...

3. You can check log.txt to debug and remove it later together with target folder of task selection: records/2021-01-07-14-21-31/selection.txt 2>&1

Toy Results

The fasta file records/records_20201219/Crickmore_CryProteins.txt is split into 10 files ending with .fasta. If we use the default parameters, i.e., the config.py is not modified, we are expecting to have results in records/2021-01-xx-xx-xx-xx (the folder with latest date)

1. group summaries: records/2021-01-07-14-21-31/filtering/group_1_summary.csv, proteins in each file sorted by SCRFs Score. No testing proteins are predicted to have similar structure as PFTs in group 2 and group 3. Take the group_1_summary.csv as an example:

   	ID	PFT Probability	SCRFs Score	Sequence
   0	Cry21Ga1	0.9998382329940796	3.79	MADLSN...
   1	Cry32Ga1	0.9999923706054688	2.73	MYQNYN...

2. dataset summaries: records/2021-01-07-14-21-31/filtering/dataset_summary.csv with proteins sorted by PFT Probability. The summary file looks as follows:

   	ID	PFT Probability	Sequence
   0	Cry32Ga1	0.9999923706054688	MYQNYN...
   1	Cry21Ga1	0.9998382329940796	MADLSN...

3. You can check log.txt to debug and remove it later together with target folder of task selection: records/2021-01-07-14-21-31/selection