Data splitting to avoid information leakage with DataSAIL

Abstract Information leakage is an increasingly important topic in machine learning research for biomedical applications. When information leakage happens during a model’s training, it risks memorizing the training data instead of learning generalizable properties. This can lead to inflated performa...

Full description

Saved in:
Bibliographic Details
Main Authors: Roman Joeres, David B. Blumenthal, Olga V. Kalinina
Format: Article
Language:English
Published: Nature Portfolio 2025-04-01
Series:Nature Communications
Online Access:https://doi.org/10.1038/s41467-025-58606-8
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850145381249187840
author Roman Joeres
David B. Blumenthal
Olga V. Kalinina
author_facet Roman Joeres
David B. Blumenthal
Olga V. Kalinina
author_sort Roman Joeres
collection DOAJ
description Abstract Information leakage is an increasingly important topic in machine learning research for biomedical applications. When information leakage happens during a model’s training, it risks memorizing the training data instead of learning generalizable properties. This can lead to inflated performance metrics that do not reflect the actual performance at inference time. We present DataSAIL, a versatile Python package to facilitate leakage-reduced data splitting to enable realistic evaluation of machine learning models for biological data that are intended to be applied in out-of-distribution scenarios. DataSAIL is based on formulating the problem to find leakage-reduced data splits as a combinatorial optimization problem. We prove that this problem is NP-hard and provide a scalable heuristic based on clustering and integer linear programming. Finally, we empirically demonstrate DataSAIL’s impact on evaluating biomedical machine learning models.
format Article
id doaj-art-a8603ca4033f49539d554f91b21eb5fe
institution OA Journals
issn 2041-1723
language English
publishDate 2025-04-01
publisher Nature Portfolio
record_format Article
series Nature Communications
spelling doaj-art-a8603ca4033f49539d554f91b21eb5fe2025-08-20T02:28:07ZengNature PortfolioNature Communications2041-17232025-04-0116111110.1038/s41467-025-58606-8Data splitting to avoid information leakage with DataSAILRoman Joeres0David B. Blumenthal1Olga V. Kalinina2Helmholtz Institute for Pharmaceutical Research Saarland (HIPS), Helmholtz Centre for Infection Research (HZI)Department Artificial Intelligence in Biomedical Engineering, Friedrich-Alexander-Universität Erlangen-NürnbergHelmholtz Institute for Pharmaceutical Research Saarland (HIPS), Helmholtz Centre for Infection Research (HZI)Abstract Information leakage is an increasingly important topic in machine learning research for biomedical applications. When information leakage happens during a model’s training, it risks memorizing the training data instead of learning generalizable properties. This can lead to inflated performance metrics that do not reflect the actual performance at inference time. We present DataSAIL, a versatile Python package to facilitate leakage-reduced data splitting to enable realistic evaluation of machine learning models for biological data that are intended to be applied in out-of-distribution scenarios. DataSAIL is based on formulating the problem to find leakage-reduced data splits as a combinatorial optimization problem. We prove that this problem is NP-hard and provide a scalable heuristic based on clustering and integer linear programming. Finally, we empirically demonstrate DataSAIL’s impact on evaluating biomedical machine learning models.https://doi.org/10.1038/s41467-025-58606-8
spellingShingle Roman Joeres
David B. Blumenthal
Olga V. Kalinina
Data splitting to avoid information leakage with DataSAIL
Nature Communications
title Data splitting to avoid information leakage with DataSAIL
title_full Data splitting to avoid information leakage with DataSAIL
title_fullStr Data splitting to avoid information leakage with DataSAIL
title_full_unstemmed Data splitting to avoid information leakage with DataSAIL
title_short Data splitting to avoid information leakage with DataSAIL
title_sort data splitting to avoid information leakage with datasail
url https://doi.org/10.1038/s41467-025-58606-8
work_keys_str_mv AT romanjoeres datasplittingtoavoidinformationleakagewithdatasail
AT davidbblumenthal datasplittingtoavoidinformationleakagewithdatasail
AT olgavkalinina datasplittingtoavoidinformationleakagewithdatasail