Home • AutoAI-Pandemics • Installing • How To Use • Citation
Published paper: in progress
BioPrediction is part of a bigger project that proposes to democratize Machine Learning for the analysis, study, and control of epidemics and pandemics. Take a look!!!
BioPrediction - Selected to participate in Prototypes for Humanity 2023, during COP28-Dubai, chosen from 3000 entries, from more than 100 countries, standing out among the 100 best, Prototypes for Humanity - COP28-Dubai. Take a look!!!
BioPrediction - Awarded as the best undergraduate project in the world in computer science by the Global Undergraduate Awards 2024, marking the first time this award in the field has been given to Latin America. Take a look!!!
Proteins play crucial roles in various biological processes, primarily through interactions with other molecules, such as other proteins. These interactions are essential for cellular pathways and the maintenance of life. Predicting Protein-Protein Interactions (PPIs) is challenging but vital to understanding cellular functions and disease mechanisms. To address this challenge, this paper proposes BioPrediction-PPI, an end-to-end Machine Learning (ML) framework for PPI prediction, aiming to simplify the process by automating a complete ML pipeline, eliminating the need for manual intervention. BioPrediction-PPI automates the entire ML process, from feature extraction to interpretability, ensuring a user-friendly and accessible experience for researchers without specialized expertise. A review of related work shows that BioPrediction-PPI is one of the few state-of-the-art end-to-end studies, the only one that does not rely on deep learning. It stands out for its ability to provide interpretability graphs, allowing the creation of minimally auditable models, and for having been extensively validated with the largest number of studies and datasets during its evaluation. The performance of BioPrediction-PPI was evaluated through comparative experiments with other studies mentioned in the literature, using various datasets. In five distinct experiments, our framework was compared with more than 31 tools developed by experts, covering more than 15 datasets. These experiments included benchmarks such as the yeast and H. pylori datasets, which have been extensively tested by numerous tools, as well as experiments involving human protein interactions and human-viral protein interactions, highlighting their importance in the study and development of therapies. We also investigated the framework’s ability to handle cross-species cases, where it was trained on interactions and proteins from one species and tested on others. In addition, we illustrate how the performance and interpretability graphs generated by BioPrediction-PPI can be used for model evaluation and experimental design, highlighting their practical benefits for informed decision-making. In summary, BioPrediction-PPI demonstrates competitive performance in nearly all evaluated studies on at least one dataset, even without employing deep learning. Its additional strengths as a white-box model with auxiliary graphs and extensive validation further position BioPrediction-PPI as a promising tool for democratizing AI model development, making it accessible and beneficial for biologists and non-experts in ML. This advancement can significantly accelerate research and development in biology and related fields.
-
To the best of our knowledge, this is the first study to propose a white-box end-to-end pipeline for feature engineering and model training to classify interactions between biological sequences, competitive with models developed by experts.
-
The pipeline was mainly tested on datasets regarding Protein-Protein Interactions.
-
BioPrediction does not require specialist human assistance.
-
BioPrediction can accelerate new studies, democratizing the use of ML techniques by non-experts.
-
Robson Parmezan Bonidia and Bruno Rafael Florentino
-
Correspondence: rpbonidia@gmail.com or bonidia@usp.br or brunorf1204@usp.br
You can download this repository and upload it to your Google Drive. Then, use the execution_form.ipynb file. With this approach, you don't need to install Python to use BioPrediction-PPI, as this file will create the environment in Google Colab, and you can fill out a form with the paths to your datasets for execution.
IMPORTANT: do not upload this GitHub repository into another folder in your Drive, as this will alter the paths, and the environment will not be created correctly. Also, upload your data inside the BioPrediction-PPI folder, as the environment will search for files within this folder. Lastly, in the reproducibility_file.ipynb, you can find some examples, as it contains the code to generate most of the results from the article.
Installing BioPrediction using Miniconda to manage its dependencies, e.g.:
$ git clone https://github.com/0nurB/BioPredictionPPI.git BioPredictionPPI
$ cd BioPredictionPPI
$ git submodule init
$ git submodule update1 - Install Miniconda:
See documentation: https://docs.conda.io/en/latest/miniconda.html
$ wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
$ chmod +x Miniconda3-latest-Linux-x86_64.sh
$ ./Miniconda3-latest-Linux-x86_64.sh
$ export PATH=~/miniconda3/bin:$PATH
2 - Create environment:
conda create --name bioprediction-ppi python=3.11.5
3 - Activate environment:
conda activate bioprediction-ppi
4 - Install packages:
pip install -r requeriments.txt
Execute the BioPrediction pipeline with the following command:
...
To run the code (Example): $ python BioPrediction.py -h
where:
-input_interactions_train: CSV format file with the interaction table (firts and second columns the proteins name and third the label with 1 for interaction and 0 for non interaction), e.g., all-data/data_human_virus/Sars/interaction.csv
-input_interactions_candidates: CSV format file with the interaction candidates to the prediction (also three columns, but the third put 2 for unlabeled candidates), e.g., all-data/data_human_virus/Sars/interaction.csv
-sequences_dictionary: fasta format file with all the sequences, e.g., all-data/data_human_virus/Sars/dictionary.fasta
Those dictionaries must contain all sequences in train, test, and candidates.
-topological_features: uses topology features to characterization of the sequences, e.g., yes or no, default=yes)
-output: output path, e.g., sars_experiment
Use a string formant to execute BioPrediction, like this complete exemple:
!python BioPrediction.py -input_interactions_train all-data/data_human_virus/Sars/interaction.csv -sequences_dictionary all-data/data_human_virus/Sars/dictionary.fasta -output sars_test -input_interactions_candidates all-data/data_human_virus/Sars/interaction.csv
...