Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective

Abstract Visualizing high-dimensional data is essential for understanding biomedical data and deep learning models. Neighbor embedding methods, such as t-SNE and UMAP, are widely used but can introduce misleading visual artifacts. We find that the manifold learning interpretations from many prior wo...

Full description

Saved in:
Bibliographic Details
Main Authors: Zhexuan Liu, Rong Ma, Yiqiao Zhong
Format: Article
Language:English
Published: Nature Portfolio 2025-05-01
Series:Nature Communications
Online Access:https://doi.org/10.1038/s41467-025-60434-9
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850231495267975168
author Zhexuan Liu
Rong Ma
Yiqiao Zhong
author_facet Zhexuan Liu
Rong Ma
Yiqiao Zhong
author_sort Zhexuan Liu
collection DOAJ
description Abstract Visualizing high-dimensional data is essential for understanding biomedical data and deep learning models. Neighbor embedding methods, such as t-SNE and UMAP, are widely used but can introduce misleading visual artifacts. We find that the manifold learning interpretations from many prior works are inaccurate and that the misuse stems from a lack of data-independent notions of embedding maps, which project high-dimensional data into a lower-dimensional space. Leveraging the leave-one-out principle, we introduce LOO-map, a framework that extends embedding maps beyond discrete points to the entire input space. We identify two forms of map discontinuity that distort visualizations: one exaggerates cluster separation and the other creates spurious local structures. As a remedy, we develop two types of point-wise diagnostic scores to detect unreliable embedding points and improve hyperparameter selection, which are validated on datasets from computer vision and single-cell omics.
format Article
id doaj-art-802b8bc55aff4658b324ec231c3f9a8e
institution OA Journals
issn 2041-1723
language English
publishDate 2025-05-01
publisher Nature Portfolio
record_format Article
series Nature Communications
spelling doaj-art-802b8bc55aff4658b324ec231c3f9a8e2025-08-20T02:03:31ZengNature PortfolioNature Communications2041-17232025-05-0116111610.1038/s41467-025-60434-9Assessing and improving reliability of neighbor embedding methods: a map-continuity perspectiveZhexuan Liu0Rong Ma1Yiqiao Zhong2Department of Statistics, University of Wisconsin-MadisonDepartment of Biostatistics, T.H. Chan School of Public Health, Harvard UniversityDepartment of Statistics, University of Wisconsin-MadisonAbstract Visualizing high-dimensional data is essential for understanding biomedical data and deep learning models. Neighbor embedding methods, such as t-SNE and UMAP, are widely used but can introduce misleading visual artifacts. We find that the manifold learning interpretations from many prior works are inaccurate and that the misuse stems from a lack of data-independent notions of embedding maps, which project high-dimensional data into a lower-dimensional space. Leveraging the leave-one-out principle, we introduce LOO-map, a framework that extends embedding maps beyond discrete points to the entire input space. We identify two forms of map discontinuity that distort visualizations: one exaggerates cluster separation and the other creates spurious local structures. As a remedy, we develop two types of point-wise diagnostic scores to detect unreliable embedding points and improve hyperparameter selection, which are validated on datasets from computer vision and single-cell omics.https://doi.org/10.1038/s41467-025-60434-9
spellingShingle Zhexuan Liu
Rong Ma
Yiqiao Zhong
Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective
Nature Communications
title Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective
title_full Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective
title_fullStr Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective
title_full_unstemmed Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective
title_short Assessing and improving reliability of neighbor embedding methods: a map-continuity perspective
title_sort assessing and improving reliability of neighbor embedding methods a map continuity perspective
url https://doi.org/10.1038/s41467-025-60434-9
work_keys_str_mv AT zhexuanliu assessingandimprovingreliabilityofneighborembeddingmethodsamapcontinuityperspective
AT rongma assessingandimprovingreliabilityofneighborembeddingmethodsamapcontinuityperspective
AT yiqiaozhong assessingandimprovingreliabilityofneighborembeddingmethodsamapcontinuityperspective