DeepFRI

Deep functional residue identification

Citing

@article {Gligorijevic2019,
	author = {Gligorijevic, Vladimir and Renfrew, P. Douglas and Kosciolek, Tomasz and Leman,
	Julia Koehler and Cho, Kyunghyun and Vatanen, Tommi and Berenberg, Daniel
	and Taylor, Bryn and Fisk, Ian M. and Xavier, Ramnik J. and Knight, Rob and Bonneau, Richard},
	title = {Structure-Based Function Prediction using Graph Convolutional Networks},
	year = {2019},
	doi = {10.1101/786236},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2019/10/04/786236},
	journal = {bioRxiv}
}

Dependencies

DeepFRI is tested to work under Python 3.6.

The required dependencies for deepFRI are Keras, TensorFlow, Numpy and scikit-learn. To install all dependencies run:

pip install .

Protein function prediction

To predict protein functions use predict.py script with the following options:

• seq str, Protein sequence as a string
• cmap str, Name of a file storing a protein contact map and sequence in *.npz file format (with the following numpy array variables: A_ca_10A, sequence, L)
• cmap_csv str, Filename of the catalogue (in *.csv file format) containg mapping between protein names and directory with *.npz files (see examples/catalogue_pdb_chains.csv)
• fasta_fn str, Fasta filename
• output_fn_prefixstr, Output filename for saving predictions and class activation maps.
• verbose bool, Whether or not to print function prediction results.
• saliency bool, Whether or not to compute class activaton maps.

Generated files:

• output_fn_prefix_MF_predictions.csv Predictions in the *.csv file format with columns: Protein, GO-term/EC-number, Score, GO-term/EC-number name.
• output_fn_prefix_MF_saliency_maps.pckl pickle file storing a dictionary of saliency maps for each predicted function of every protein.

DeepFRI offers 4 possible options for predicting functions. See examples below.

Option 1: predicting functions of a protein from its contact map

Example: predicting MF-GO terms for Parvalbumin alpha protein using its sequence and contact map (PDB: 1S3P):

>> python predict.py --cmap examples/1S3P-A.npz --verbose

Output:

Protein GO-term/EC-number Score GO-term/EC-number name
1S3P-A GO:0043167 0.95134 ion binding
1S3P-A GO:0046872 0.90832 metal ion binding
1S3P-A GO:0043169 0.90517 cation binding
1S3P-A GO:0005509 0.87179 calcium ion binding
1S3P-A GO:0043168 0.06332 anion binding
1S3P-A GO:0031072 0.00247 heat shock protein binding
1S3P-A GO:1901567 0.00099 fatty acid derivative binding
1S3P-A GO:0045159 0.00009 myosin II binding
1S3P-A GO:0032027 0.00001 myosin light chain binding

Option 2: predicting functions of a protein from its sequence

Example: predicting MF-GO terms for Parvalbumin alpha protein using its sequence (PDB: 1S3P):

>> python predict.py --seq 'SMTDLLSAEDIKKAIGAFTAADSFDHKKFFQMVGLKKKSADDVKKVFHILDKDKDGFIDEDELGSILKGFSSDARDLSAKETKTLMAAGDKDGDGKIGVEEFSTLVAES' --verbose

Output:

Protein GO-term/EC-number Score GO-term/EC-number name
query_prot GO:0005509 0.99434 calcium ion binding
query_prot GO:0046872 0.98379 metal ion binding
query_prot GO:0043169 0.98373 cation binding
query_prot GO:0043167 0.97652 ion binding
query_prot GO:0017022 0.00591 myosin binding
query_prot GO:0043236 0.00045 laminin binding
query_prot GO:0045159 0.00015 myosin II binding
query_prot GO:0019237 0.00003 centromeric DNA binding
query_prot GO:0017151 0.00002 DEAD/H-box RNA helicase binding
query_prot GO:0032027 0.00001 myosin light chain binding
query_prot GO:0000171 0.00001 ribonuclease MRP activity

Option 3: predicting functions of proteins from a fasta file

>> python predict.py --fast_fn examples/pdb_chains.fasta

Option 4: predicting functions of proteins from contact map catalogue

>> python predict.py --cmap_csv examples/catalogue_pdb_chains.csv -o DeeFRI_pdb_chain --saliency

Output:

see files:

examples/outputs/DeepFRI_pdb_chains_MF_saliency_maps.pckl`, `examples/outputs/DeepFRI_pdb_chains_MF_predictions.csv`

Training DeepFRI

To train deepFRI run the following command from the project directory:

>> python train_DeepFRI.py --model_name model_name_prefix

Output

Generated files:

• model_name_prefix_model.hdf5 trained model with architecture and weights saved in HDF5 format
• model_name_prefix_pred_scores.pckl pickle file with predicted GO term/EC number scores for test proteins
• model_name_prefix_metadata.pckl pickle file with metadata (GO terms/names, EC numbers/names, optimal thresholds for each GO term/EC number)

See examples of pre-trained models in: trained_models/.

Flags

A number of FLAGS is available to specify the behavior of deepFRI, both for prediction and training:

• model_name str, name of the model. Default: GCN-LM_model
• gcn_dims list (int), dimensions of GCN layers. Default: [128, 256, 512]
• hidden_dims list (int), dimensions of Dense layers. Default: [512]
• dropout float, dropout rate for Dense layer. Default: 0.30
• l2_reg float, l2 regularization coefficient for GCN layers. Default: 1e-4
• epochs int, number of epochs to train the model. Default: 100
• batch_size int, Batch size. Default: 64
• pad_len int, maximum padding length for sequences and contact maps. Default: 1000
• results_dir str, directory with exported models and results. Default: ./results/
• ont str, GO or EC ontology. Default: molecular function
• cmap_type str, type of contact maps (A_nbr, A_ca or A_all). Default: A_ca
• lm_model_name str, keras pre-trained LSTM Language Model name. Default: ./trained_models/lstm_lm.h5
• split_fn str, pickle file with train/test/valid PDB IDs and their annotatin matrix. Default: train_test_split_seqsim_30.pckl (see Data section below)
• catalogue str, csv file mapping PDB IDs to numpy files storing individual contact maps. Default: catalogue.csv
• train_tfrecord_fn str, train tfrecords file name. Default: train.tfrecords (see Data section below)
• valid_tfrecord_fn str, validaiton tfrecords file name. Default: valid.tfrecords (see Data section below)

Functional residue identification

To visualize class activation (saliency) maps use viz_saliency.py script with the following options:

• saliency_fn str, pickle filename with saliency maps generated by predict.py script (see Option 4 above)
• list_all bool, list all proteins and their predicted GO terms with corresponding class activation (saliency) maps
• protein_id str, protein (PDB chain), saliency maps of which are to be visualized for each predicted function
• go_id str, GO term, saliency maps of which are to be visualized
• go_name str, GO name, saliency maps of which are to be visualized

Generated files:

• saliency_fig_PDB-chain_GOname.png class activation (saliency) map profile over sequence (see fig below, right)
• pymol_viz.py pymol script for mapping salient residues onto 3D structure (pymol output is shown in fig below, left)

Example:

>>> python viz_gradCAM.py --saliency_fn examples/outputs/DeepFRI_pdb_chains_MF_saliency_maps.pckl --protein_id 1S3P-A --go_id GO:0005509

Output:

Data

Data (train.tfrecord, valid.tfrecord files) used for re-training the model can be downloaded from:

https://users.flatironinstitute.org/vgligorijevic/public_www/DeepFRI_data