SARS-CoV-2 Prediction of recurrent mutations

This repository contains supplementary material to the publication Prediction of Recurrent Mutations in SARS-CoV-2 Using Artificial Neural Networks .

Contents:

• Models for prediction of recurrent mutations and positions.
• Predicting mutations using just the nucleotide sequence.
• Dataset.
• Citation.
• Other information.

Models usage for prediction/inference

• Dowload the models from: 7zfile
• Unzip the pos_mut_models.7z file with

7z e pos_mut_models.7z

You should be able to see the following directories after extraction:

tree -d pos_mut_models/
pos_mut_models/
├── mut_pred_5_10_15
│   ├── th10
│   ├── th15
│   └── th5
└── pos_pred_1_5_10_15
    ├── th1
    ├── th10
    ├── th15
    └── th5

Create a new conda environment with the requirements of the file spec-file.txt . You can replace the new_env_name with your own new environment name

conda create --name new_env_name --file spec-file.txt

Further information to manage conda environments can be seen in link

The input file for prediction is expected to be a csv file with the variable columns present in the files:

input_position_example.csv, for RM position prediction, 611 columns

input_mutation_example.csv, for RM mutation prediction 803 columns

Activate the conda environment with:

conda activate new_env_name

Run the prediction script, example mutation prediction

python3 pos_mut_predict.py -data_file="input_mutation_example.csv" \
-models_dir="./pos_mut_models/" \
-output_dir="./output_dir/" \
-batch_size=32 \
-th="15" \
-prediction_type="mutation" \
-file_separator="," \
-first_column_is_index=1 

position prediction example

python3 pos_mut_predict.py -data_file="input_position_example.csv" \
-models_dir="./pos_mut_models/" \
-output_dir="./output_dir/" \
-batch_size=32 \
-th="5" \
-prediction_type="position" \
-file_separator="," \
-first_column_is_index=0 

• data_file: input file with the features used to predict.
• models_dir: directory where the models are stored.
• output_dir: directory where to save the predictions.
• batch_size: Amount of rows from the data_file to process at the same time.
• th: Recurrent Mutation threshold. 1, 5, 10 and 15 available for position and 5, 10, 15 for mutation.
• prediction_type: Type of prediction, options available are position and mutation.
• file separator: character that separates the columns in the data_file.
• first_columns_is_index: set as 0 if the first columns in data_file is a feature and not an index.

The description of each feature column is in the file variable_names_and_description.csv.

Predict mutation using only nucleotide sequence

In addition to work presented, 3 additional models were introduced using only the nucleotide sequence with the format: TACAAACCTTTTC>G. The C in the middle is the position that mutates to >G.
Since the nucleotide sequence is used only, there is a loss of performance of 3%-5% as is present in seq_only_mut_pred_scores.csv. These models can be downloaded from link

python3 rm_predict_from_seq.py -sequence="TACAAACCTTTTC>G" \
-models_dir="./sequence_predict_models/" \
-output_dir="./output_dir/" \
-batch_size=32 \
-th="15" \
-prediction_type="mutation"

Or by using a file with a sequence per file like seq_file_txt. Each line with the format: TACAAACCTTTTC>G

python3 rm_predict_from_seq.py -sequence_file="seq_file.txt" \
-models_dir="./sequence_predict_models/" \
-output_dir="./output_dir/" \
-batch_size=32 \
-th="15" \
-prediction_type="mutation"

Dataset

The datasets for prediction of positions and mutations can be downloaded from link_position and link_mutation.

Citation

If you are using any of this material please cite the work as:

Saldivar-Espinoza, B.; Macip, G.; Garcia-Segura, P.; Mestres-Truyol, J.; Puigbò, P.; Cereto-Massagué, A.; Pujadas, G.; Garcia-Vallve, S. Prediction of Recurrent Mutations in SARS-CoV-2 Using Artificial Neural Networks. Int. J. Mol. Sci. 2022, 23, 14683. https://doi.org/10.3390/ijms232314683

BibLaTeX:

@article{saldivar-espinoza_prediction_2022,
	title = {Prediction of Recurrent Mutations in {SARS}-{CoV}-2 Using Artificial Neural Networks},
	volume = {23},
	rights = {http://creativecommons.org/licenses/by/3.0/},
	issn = {1422-0067},
	url = {https://www.mdpi.com/1422-0067/23/23/14683},
	doi = {10.3390/ijms232314683},
	pages = {14683},
	number = {23},
	journaltitle = {International Journal of Molecular Sciences},
	author = {Saldivar-Espinoza, Bryan and Macip, Guillem and Garcia-Segura, Pol and Mestres-Truyol, Júlia and Puigbò, Pere and Cereto-Massagué, Adrià and Pujadas, Gerard and Garcia-Vallve, Santiago},
	urldate = {2022-11-24},
	date = {2022-01},
	langid = {english},
	note = {Number: 23
Publisher: Multidisciplinary Digital Publishing Institute},
	keywords = {{COVID}-19, machine learning, mutations, {SARS}-{CoV}-2}
}

Other formats can be obtained form the citation files for your bibliography management software.

Other information

Additional technical aspects not described in the manuscript and source codes will be provided under request.