VTechWorks staff will be away for the winter holidays starting Tuesday, December 24, 2024, through Wednesday, January 1, 2025, and will not be replying to requests during this time. Thank you for your patience, and happy holidays!
 

Systematic auditing is essential to debiasing machine learning in biology

Simple item page
dc.contributor.authorEid, Fatma-Elzahraaen
dc.contributor.authorElmarakeby, Haitham A.en
dc.contributor.authorChan, Yujia Alinaen
dc.contributor.authorFornelos, Nadineen
dc.contributor.authorElHefnawi, Mahmouden
dc.contributor.authorVan Allen, Eliezer M.en
dc.contributor.authorHeath, Lenwood S.en
dc.contributor.authorLage, Kasperen
dc.contributor.departmentComputer Scienceen
dc.date.accessioned2021-05-19T13:02:58Zen
dc.date.available2021-05-19T13:02:58Zen
dc.date.issued2021-02-10en
dc.description.abstractBiases in data used to train machine learning (ML) models can inflate their prediction performance and confound our understanding of how and what they learn. Although biases are common in biological data, systematic auditing of ML models to identify and eliminate these biases is not a common practice when applying ML in the life sciences. Here we devise a systematic, principled, and general approach to audit ML models in the life sciences. We use this auditing framework to examine biases in three ML applications of therapeutic interest and identify unrecognized biases that hinder the ML process and result in substantially reduced model performance on new datasets. Ultimately, we show that ML models tend to learn primarily from data biases when there is insufficient signal in the data to learn from. We provide detailed protocols, guidelines, and examples of code to enable tailoring of the auditing framework to other biomedical applications. Fatma-Elzahraa Eid et al. illustrate a principled approach for identifying biases that can inflate the performance of biological machine learning models. When applied to three biomedical prediction problems, they identify previously unrecognized biases and ultimately show that models are likely to learn primarily from data biases when there is insufficient learnable signal in the data.en
dc.description.notesWe thank Yu Xia (McGill University), Paul A. Clemons (Broad Institute of MIT and Harvard), and Lucas Janson (Harvard University) for helpful discussions and Shuyu Wang (UCSF) for help in dataset preparation. This work was supported by grants from The Stanley Center for Psychiatric Research, the National Institute of Mental Health (R01 MH109903), the Simons Foundation Autism Research Initiative (award 515064), the Lundbeck Foundation (R223-2016-721), a Broad Next10 grant, and a Broad Shark Tank grant. Y.A.C. was funded by a Human Frontier Science Program Postdoctoral Fellowship [LT000168/2015-L].en
dc.description.sponsorshipStanley Center for Psychiatric Research; National Institute of Mental HealthUnited States Department of Health & Human ServicesNational Institutes of Health (NIH) - USANIH National Institute of Mental Health (NIMH) [R01 MH109903]; Simons Foundation Autism Research Initiative [515064]; Lundbeck FoundationLundbeckfonden [R223-2016-721]; Broad Next10 grant; Broad Shark Tank grant; Human Frontier Science Program Postdoctoral FellowshipHuman Frontier Science Program [LT000168/2015-L]en
dc.format.mimetypeapplication/pdfen
dc.identifier.doihttps://doi.org/10.1038/s42003-021-01674-5en
dc.identifier.eissn2399-3642en
dc.identifier.issue1en
dc.identifier.other183en
dc.identifier.pmid33568741en
dc.identifier.urihttp://hdl.handle.net/10919/103378en
dc.identifier.volume4en
dc.language.isoenen
dc.rightsCreative Commons Attribution 4.0 Internationalen
dc.rights.urihttp://creativecommons.org/licenses/by/4.0/en
dc.titleSystematic auditing is essential to debiasing machine learning in biologyen
dc.title.serialCommunications Biologyen
dc.typeArticle - Refereeden
dc.type.dcmitypeTexten
dc.type.dcmitypeStillImageen

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
s42003-021-01674-5.pdf
Size:
826.74 KB
Format:
Adobe Portable Document Format
Description:
Published version
Download

Collections

Scholarly Works, Computer Science