Back

mcBERT: Patient-Level Single-cell Transcriptomics Data Representation

Querfurth, B. v.; Lohmoeller, J.; Pennekamp, J.; Bleckwehl, T.; Kramann, R.; Wehrle, K.; Hayat, S.

2024-11-06 bioinformatics
10.1101/2024.11.04.621897 bioRxiv
Show abstract

Single-cell RNA sequencing (scRNA-seq) transcriptomics improves our understanding of cellular heterogeneity in healthy and pathological states. However, most scRNA-seq analyses remain confined to single cells or distinct cell populations, limiting their clinical applicability. Addressing the need to translate single-cell insights into a patient-level disease understanding, we introduce mcBERT, a new method that leverages scRNA-seq data and a transformer-based model to generate integrative patient representations using a self-supervised learning phase followed by contrastive learning to refine these representations. Our evaluations of mcBERT across 7 million cells from 1223 individuals encompassing diverse disease states in heart, kidney, blood cell, and lung tissues show that learned representations facilitate a robust identification of disease cohorts and enable comparisons of patient similarity. Moreover, our findings indicate that mcBERT can accurately classify disease phenotypes, also in previously unseen biospecimens and patients. Independent of the specific tissue, mcBERT extends the utility of scRNA-seq data from cellular analysis to potentially actionable patient-centric applications.

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.