h-BiP

h-BiP is a python package that predicts the human Binding Potential from the S protein sequences of alpha and beta coronaviruses as described in the following paper:

Gonzalez-Isunza, G., Jawaid, M.Z., Liu, P., Cox, D.L., Vazquez, M. and Arsuaga, J., 2023. Using machine learning to detect coronaviruses potentially infectious to humans. Scientific Reports, 13(1), p.9319.

Results presented at Gonzalez-Isunza et al. are available at:

• Alpha and beta coronaviruses:
  ./hbip_scores/alpha_beta_scores.csv
• Alpha and beta coronaviruses excluding SARS2:
  ./hbip_scores/alpha_beta_before_2020_scores.csv

Any publication that discloses findings arising from using this data or the source code should cite Gonzalez-Isunza et al.

NEW (30/09/24): Use our colab to make predictions on a fasta file and get the h-BiP score in seconds!!

Basic usage

Setup

• Clone or download ZIP the repository.
• Create and activate an environment using conda or a virtual environment with Pyhon=3.9.7.
• CD into the repository directory (h-BiP)
• Pip install the hbip package

pip install .

Reproducing results from Gonzalez-Isunza et al.

• Open the terminal, navigate to /h-BiP and type:

python3 hbip_reproduce.py

• To train any other dataset, create a config file using the template provided at the data folder and add the file path at the end.

python3 hbip_reproduce.py ./data/my_own_data_config.yml

Predicting h-BiP scores for spike amino acid sequences
By default, h-BiP will compute the score from the alpha_beta model (full dataset).

• If no fasta file is provided, it will compute de score for SARS-CoV-2. After navigating in the terminal to /h-BiP type:

python3 hbip_predict.py

• From a fasta file:

python3 hbip_predict.py path_to_fasta_file

• To use a different model for prediction, add the model name at the end.

python3 hbip_predict.py path_to_fasta_file model_name

Data

Available datasets

• Alpha and beta coronaviruses (full dataset):
  • 2,534 unique spike protein sequences annotated for binding condition to human receptor.
  • alpha_beta_hip_train.csv and alpha_beta_hip_test.csv
• Alpha and beta coronaviruses excluding SARS2 viruses and all viruses uploaded to NCBI after Dec. 31, 2019:
  • alpha_beta_before_2020_hip_train.csv and alpha_beta_before_2020_hip_test.csv

Dataset structure

Fields marked with asterix are used in the code (position is irrelevant).

• *Accession: string
  Identifier for the sequence.
• *Sequence: string
  Amino acid sequence for the spike protein (S).
• Species: string
  Virus' species.
• *Species_agg: string
  Simplified (aggregated) label for Species. Current labels are:
  • hCoV-OC43, hCoV-HKU1, Beta 1, MERS, MERS-related, other, SARS-CoV-2, SARS-CoV-1, Other Sarbecovirus, Beta other, PorcineEp, hCoV-NL63, hCoV-229E, Other Alpha.
  • The only label needed in the code is 'SARS-CoV-2'
• *Virus: string
  Virus name.
• Host: string
  Host's name.
• Host_agg: string
  Simplified (aggregated) label for Host.
• Country: string
  Virus' origin.
• Human: integer (0/1)
  Human host will have a value of 1 and 0 otherwise.
• Binds: integer (0/1)
  Experimental evidence of binding to human receptor.

Citing this work

If you use the code or data in this package, please cite:

@article {Gonzalez-Isunza2023using,
	author = {Gonzalez-Isunza, Georgina and Jawaid, M. Zaki and Liu, Pengyu and Cox, Daniel L. and Vazquez, Mariel and Arsuaga, Javier},
	title = {Using machine learning to detect coronaviruses potentially infectious to humans},
	journal = {Scientific reports},
	year = {2023},
	publisher = {Nature Publishing Group UK London}
}