2026-09-19T20:27:37Zhttps://dspace.mit.edu/server/oai/request

oai:dspace.mit.edu:1721.1/1566262024-09-04T03:07:23Zcom_1721.1_7582com_1721.1_7581col_1721.1_131023

Uhler, Caroline Tan, Zipei Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science 2024-09-03T21:12:50Z 2024-09-03T21:12:50Z 2024-05 2024-07-11T14:36:09.597Z https://hdl.handle.net/1721.1/156626 https://orcid.org/0009-0002-2864-4935 Polygenic risk scores (PRS) estimate an individual’s risk of developing a certain disease, suggesting that differences between cells of individuals with high versus low PRS could give us insight into the cellular disease mechanisms. To study metabolic diseases, we analyze the distribution of cell states of lipocytes of individuals with different PRS for metabolic diseases, thereby associating individual-level genotypes with cell-level features. To accomplish this, we make use of a recent large-scale lipocyte microscopy imaging dataset. By learning a representation of multi-channel lipocyte microscopy images using a convolutional autoencoder, we perform unsupervised clustering on the learnt representations to identify different cell states. We analyze the distribution of these cell states in different individuals and associate their PRS to the observed cell state distributions. Finally, we show that it is possible to generate counterfactual lipocyte images and understand the effect of increased or reduced PRS on cell states through transforming the learnt representations. M.Eng. Massachusetts Institute of Technology Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) Copyright retained by author(s) https://creativecommons.org/licenses/by-nc-sa/4.0/ Representation Learning Associates Patients’ Risks for Metabolic Diseases with Features of Their Lipocytes Thesis application/pdf Master Master of Engineering in Electrical Engineering and Computer Science Publication unknown unknown <Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="d71e271a-f3ab-4bd2-9c1c-bc78992d0a29"> <Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type> <Title>Representation Learning Associates Patients’ Risks for Metabolic Diseases with Features of Their Lipocytes</Title> <PublishedIn> <Publication> </Publication> </PublishedIn> <PublicationDate>2024-05</PublicationDate> <Authors> <Author> <DisplayName>Tan, Zipei</DisplayName> <Affiliation> <OrgUnit> </OrgUnit> </Affiliation> </Author> </Authors> <Editors> </Editors> <Publishers> <Publisher> <DisplayName>Massachusetts Institute of Technology</DisplayName> <OrgUnit /> </Publisher> </Publishers> <License>https://creativecommons.org/licenses/by-nc-sa/4.0/</License> <Abstract>Polygenic risk scores (PRS) estimate an individual’s risk of developing a certain disease, suggesting that differences between cells of individuals with high versus low PRS could give us insight into the cellular disease mechanisms. To study metabolic diseases, we analyze the distribution of cell states of lipocytes of individuals with different PRS for metabolic diseases, thereby associating individual-level genotypes with cell-level features. To accomplish this, we make use of a recent large-scale lipocyte microscopy imaging dataset. By learning a representation of multi-channel lipocyte microscopy images using a convolutional autoencoder, we perform unsupervised clustering on the learnt representations to identify different cell states. We analyze the distribution of these cell states in different individuals and associate their PRS to the observed cell state distributions. Finally, we show that it is possible to generate counterfactual lipocyte images and understand the effect of increased or reduced PRS on cell states through transforming the learnt representations.</Abstract> <Access xmlns="http://purl.org/coar/access_right" > </Access> </Publication>