What is the IMPROVE project?

IMPROVE aims to establish methodology to systematically and rigorously compare supervised learning models. To this end, IMPROVE has four components:

1. IMPROVE Python Package (improvelib): Python tools and protocols for standardizing model code structure, facilitating modular code allowing to design complex workflows

2. IMPROVE Workflows Evaluation Schemes: distributed and modular workflows for large-scale model evaluation and downstream model comparison

3. Benchmark Datasets: standardized datasets (for each scientific application), enabling consistent and rigorous model benchmarking

4. Community Curated Models: A diverse collection of models from the scientific community (a collocation per application), curated within the IMPROVE framework and organized by application, providing a baseline for comparison

IMPROVE can be applied to a variety of supervised learning models. We have begun with the Drug Response Prediction problem, standardizing pan-cancer, pan-drug community models that predict monotherapy drug response with a standardized benchmark dataset.

For more information

• Background
• Future Directions
• Acknowledgments

Access the code base https://github.com/JDACS4C-IMPROVE/

References

1. A. Partin et al. “Deep learning methods for drug response prediction in cancer: Predominant and emerging trends”, Frontiers in Medicine, Section Prediction Oncology, 2023

1. JC. Overbeek and A. Partin et al. “Assessing reusability of deep learning-based monotherapy drug response prediction models trained with omics data”, arXiv, 2024