Software Development

PICO

From research to practice: improving a bioinformatics tool

Context

A team from NUTRISS developed an analysis tool called PICO (Pipeline of Interpretable Classification for Omics). This software makes it possible to apply machine learning algorithms to untargeted metabolomics datasets acquired by liquid chromatography coupled with mass spectrometry. The goal is to extract the most important features, as they represent potential novel biomarkers. The interface is designed to be easy to use and intuitive, even for those with little or no experience in programming and artificial intelligence.

Our contribution: making PICO more accessible

We were called upon to optimize the code for this project. Following productive exchanges with the NUTRISS team, we identified several key areas for improving the tool and preparing it to be shared with a broader community.

Achievements

Distribution Optimization

We created a Python package on PyPI (pico-ml) and a Docker image (elinaff/pico), simplifying the installation of PICO to a single command. 

				
					# With Python
pip install pico-ml

# With Docker
docker run elinaff/pico:latest

Cross-platform compatibility

We improved the installation process on Windows, Linux, and macOS, making it easier for more researchers to use regardless of their operating system.

Performance Improvements

By optimizing the code, we significantly increased the execution speed of the analysis pipeline, allowing users to obtain results more quickly.

Stability and maintenance 


The addition of a logger enabled better tracking of operations and improved debugging. We also fixed several bugs and implemented improvements to make PICO more stable and reliable.

Improved 
user experience 

Mindful of the importance of user experience, we enhanced it through three improvements:

  1. Improved smoothness when the user navigates between the different tabs of the application.
  2. Immediate parameter validation to alert the user if the requested experiment is invalid, avoiding errors after lengthy processing.
  3. Automatic saving of parameters and results in unique timestamped files, with the file location displayed to the user.

The result

Toward a Scientific Publication

Our work helped make PICO not only a more powerful tool, but also easier to install and use. These improvements were essential for the student leading the project, allowing her to focus on writing a scientific article and preparing her tool for publication to the research community.

This project demonstrates how collaboration between software development and scientific research can help improve and share technological tools.

Collaborators