- Notifications
You must be signed in to change notification settings - Fork0
bzhanglab/DeepVEP
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
DeepVEP: mutation impact prediction on post-translational modifications using deep learning
$ git clone https://github.com/bzhanglab/DeepVEP
DeepVEP is a python3 package. TensorFlow (>=2.6) is supported. Its dependencies can be installed via
$ pip install -r requirements.txt
DeepVEP has been tested on both Linux and Windows systems. It supports training and prediction on both CPU and GPU.
The pretrained model files are available atDeepVEP model repository. After the model files are downloaded, decompress the *.tar.gz file and move all the files to themodels folder as shown below:
├── README.md├── deepvep.py├── lib│ ├── DataIO.py│ ├── Metrics.py│ ├── ModelView.py│ ├── MutationUtils.py│ ├── PTModels.py│ ├── PeptideEncode.py│ ├── RegCallback.py│ ├── Utils.py│ └── __init__.py├── models└── requirements.txtRun the following command line to show all command line options:
python deepmp.py predict -h -h or --help show this help message and exit -i or --input Input data for prediction -d or --db Protein database -o or --out_dir Output directory -w or --window_size Window size for mutation impact prediction. In default, 7 amino acids in both side of a mutation. -m or --model Trained model path -t or --task Prediction type: 1=Mutation impact prediction, 2=PTM site prediction -e or --ensemble Ensemble method, 1: average, 2: meta_lr, default is 1. -s or --explain_model Perform model interpretability analysis -b or --bg_data Data used as background data in model interpretabilityBelow is an example for mutation impact prediction. The input for-i is aTSV format file which contains mutation information. The input for-d is a protein database file in FASTA format which contains the protein sequences for the wild type of proteins.
python deepvep.py predict -m models/ -i example/mutation_input.tsv -d example/Q5S007.fasta -t 1 -o mutation_output_folderThe required columns for input of-i includeProtein,AA_Ref,AA_Pos andAA_Var. An example ("example/mutation_input.tsv") is shown below:
Protein AA_Ref AA_Pos AA_VarQ5S007 R 1441 CQ5S007 R 1441 HBelow please find the description of each column in the output file "example/mutation_input.tsv":
| Column name | Description |
|---|---|
| Protein | the protein name in the input fasta file |
| AA_Ref | the wild type amino acid |
| AA_Pos | the mutation position on the protein (1-based: the position of the first amino acid on the protein is 1) |
| AA_Var | the mutation amino acid |
For the above example input (-i), the wild type protein sequence of proteinQ5S007 should be present in the input protein database for-d. The first 10 lines of the file "example/Q5S007.fasta" are shown below:
>Q5S007MASGSCQGCEEDEETLKKLIVRLNNVQEGKQIETLVQILEDLLVFTYSERASKLFQGKNIHVPLLIVLDSYMRVASVQQVGWSLLCKLIEVCPGTMQSLMGPQDVGNDWEVLGVHQLILKMLTVHNASVNLSVIGLKTLDLLLTSGKITLLILDEESDIFMLIFDAMHSFPANDEVQKLGCKALHVLFERVSEEQLTEFVENKDYMILLSALTNFKDEEEIVLHVLHCLHSLAIPCNNVEVLMSGNVRCYNIVVEAMKAFPMSERIQEVSCCLLHRLTLGNFFNILVLNEVHEFVVKAVQQYPENAALQISALSCLALLTETIFLNQDLEEKNENQENDDEGEEDKLFWLEACYKALTWHRKNKHVQEAACWALNNLLMYQNSLHEKIGDEDGHFPAHREVMLSMLMHSSSKEVFQASANALSTLLEQNVNFRKILLSKGIHLNVLELMQKHIHSPEVAESGCKMLNHLFEGSNTSLDIMAAVVPKILTVMKRHETSLPVQLEALRAILHFIVPGMPEESREDTEFHHKLNMVKKQCFKNThe output folder ("mutation_output_folder") of the example command line looks like below:
mutation_output_folder├── deepvep-mutation_impact.tsv├── acetylation_k├── glycosylation_n├── methylation_k├── methylation_r├── phosphorylation_st├── phosphorylation_y├── sumoylation_k└── ubiquitination_kThe output file "mutation_output_folder/deepvep-mutation_impact.tsv" contains the predicted mutation impact on all the PTM sites supported by DeepVEP. This is the only file that users need to use for downstream analysis. The other folders contain intermediate prediction files.
Protein AA_Ref AA_Pos AA_Var pos diff_pos w_pep m_pep w_prob m_prob delta_prob ptmQ5S007 R 1441 C 1443 -2 FNIKARASSSPVILV FNIKACASSSPVILV 0.7837325 0.2552052 -0.5285273 phosphorylation_stQ5S007 R 1441 C 1444 -3 NIKARASSSPVILVG NIKACASSSPVILVG 0.88949174 0.43282443 -0.45666731 phosphorylation_stQ5S007 R 1441 C 1445 -4 IKARASSSPVILVGT IKACASSSPVILVGT 0.8771168 0.49486965 -0.38224715 phosphorylation_stQ5S007 R 1441 H 1443 -2 FNIKARASSSPVILV FNIKAHASSSPVILV 0.7837325 0.26106784 -0.52266466 phosphorylation_stQ5S007 R 1441 H 1444 -3 NIKARASSSPVILVG NIKAHASSSPVILVG 0.88949174 0.44303632 -0.44645542 phosphorylation_stQ5S007 R 1441 H 1445 -4 IKARASSSPVILVGT IKAHASSSPVILVGT 0.8771168 0.48524436 -0.39187244 phosphorylation_stBelow please find the description of each column in the output file "output_folder/site_prediction.tsv":
| Column name | Description |
|---|---|
| Protein | the protein name in the input fasta file |
| AA_Ref | the wild type amino acid |
| AA_Pos | the mutation position on the protein (1-based: the position of the first amino acid on the protein is 1) |
| AA_Var | the mutation amino acid |
| pos | the PTM site on the protein (1-based: the position of the first amino acid on the protein is 1) |
| diff_pos | the distance between the PTM site and the mutation site |
| w_pep | the wild type peptide sequence in which the center is PTM site |
| m_pep | the mutant peptide sequence in which the center is PTM site |
| w_prob | predicted PTM site probability for wild type sequence |
| m_prob | predicted PTM site probability for mutant sequence |
| delta_prob | mutation impact on the PTM site: m_prob - w_prob |
| ptm | PTM name |
The prediction took less than 3 minutes using CPU on a Linux server (64G RAM and 16 CPUs).
Below is an example for PTM site prediction. The input (-d) is a protein database file in FASTA format which contains the protein sequences to predict. The following command line is used to predict all PTM sites supported by DeepVEP.
python deepvep.py predict -m models/ -d example/Q5S007.fasta -t 2 -o output_folderThe output folder ("output_folder") of the example command line looks like below:
output_folder/├── site_prediction.tsv├── acetylation_k├── glycosylation_n├── methylation_k├── methylation_r├── phosphorylation_st├── phosphorylation_y├── sumoylation_k└── ubiquitination_kThe output file "output_folder/site_prediction.tsv" contains all the predicted PTM sites. This is the only file that users need to use for downstream analysis. The other folders contain intermediate prediction files.
protein aa pos x y_pred fpr ptmQ5S007 K 17 ASGSCQGCEEDEETLKKLIVRLNNVQEGKQI 0.99253386 0.0027506112469437 acetylation_kQ5S007 K 18 SGSCQGCEEDEETLKKLIVRLNNVQEGKQIE 0.54781103 0.1937652811735941 acetylation_kQ5S007 K 30 TLKKLIVRLNNVQEGKQIETLVQILEDLLVF 0.0013190061 0.9529339853300732 acetylation_kQ5S007 K 53 ILEDLLVFTYSERASKLFQGKNIHVPLLIVL 0.70806825 0.1253056234718826 acetylation_kQ5S007 K 58 LVFTYSERASKLFQGKNIHVPLLIVLDSYMR 0.0148314405 0.7331907090464548 acetylation_kQ5S007 K 87 MRVASVQQVGWSLLCKLIEVCPGTMQSLMGP 0.0005026354 0.9819682151589242 acetylation_kQ5S007 K 120 VGNDWEVLGVHQLILKMLTVHNASVNLSVIG 0.00056051335 0.9801344743276283 acetylation_kQ5S007 K 137 LTVHNASVNLSVIGLKTLDLLLTSGKITLLI 0.007324457 0.82059902200489 acetylation_kQ5S007 K 147 SVIGLKTLDLLLTSGKITLLILDEESDIFML 0.0003533643 0.9865525672371638 acetylation_kBelow please find the description of each column in the output file "output_folder/site_prediction.tsv":
| Column name | Description |
|---|---|
| protein | the protein name from the input fasta file |
| aa | the amino acid PTM site to predict |
| pos | the PTM site on the protein (1-based: the position of the first amino acid on the protein is 1) |
| x | a peptide sequence of 31 amino acids in which the center is the predicted PTM site |
| y_pred | predicted probability |
| fpr | false positive rate using the predicted probability as the threshold to define positive PTM site |
| ptm | PTM name |
The prediction took less than 3 minutes using CPU on a Linux server (64G RAM and 16 CPUs).
There is no a manuscript to cite yet.