Evaluating Bin2Cell for High-Resolution Morphological and Transcriptomic Integration*

Session Number

3

Advisor(s)

Dr. Wei Wu, University of California San Francisco

Location

A147

Discipline

Medical and Health Sciences

Start Date

15-4-2026 2:15 PM

End Date

15-4-2026 3:00 PM

Abstract

Accurately interpreting how cells interact within their respective tissue environments is integral for studying biological systems. Recent advances in spatial transcriptomics technologies have proved capable for this purpose, with platform VisiumHD capturing data at resolutions of roughly 2 micrometers. Despite this improvement, analyzing VisiumHD results has proved a challenge, as default analysis aggregates its extremely precise data into larger 8-micrometer squares, blurring true cell boundaries and leading to inaccurate cell location, identity, and gene expression profiles. Bin2Cell proves as a promising solution for this issue, using a deep learning model called StarDist to find cell nuclei within an image, then focusing nearby gene data into “reconstructed” cells by using the nuclei as anchor points. In this study, we analyze the effectivity of Bin2Cell as a computational method for reconstructing biologically meaningful cell-level representations from Visium HD spatial transcriptomic data using both benchmark datasets provided with the Bin2Cell pipeline and independent datasets processed through a similar workflow.

Share

COinS
 
Apr 15th, 2:15 PM Apr 15th, 3:00 PM

Evaluating Bin2Cell for High-Resolution Morphological and Transcriptomic Integration*

A147

Accurately interpreting how cells interact within their respective tissue environments is integral for studying biological systems. Recent advances in spatial transcriptomics technologies have proved capable for this purpose, with platform VisiumHD capturing data at resolutions of roughly 2 micrometers. Despite this improvement, analyzing VisiumHD results has proved a challenge, as default analysis aggregates its extremely precise data into larger 8-micrometer squares, blurring true cell boundaries and leading to inaccurate cell location, identity, and gene expression profiles. Bin2Cell proves as a promising solution for this issue, using a deep learning model called StarDist to find cell nuclei within an image, then focusing nearby gene data into “reconstructed” cells by using the nuclei as anchor points. In this study, we analyze the effectivity of Bin2Cell as a computational method for reconstructing biologically meaningful cell-level representations from Visium HD spatial transcriptomic data using both benchmark datasets provided with the Bin2Cell pipeline and independent datasets processed through a similar workflow.