Saved in:
Bibliographic Details
Main Authors: Weerasekara, Sachini, Darras, Natasha, Kamarthi, Sagar, Price, Colles, Isaacs, Jacqueline
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.07938
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915994042105856
author Weerasekara, Sachini
Darras, Natasha
Kamarthi, Sagar
Price, Colles
Isaacs, Jacqueline
author_facet Weerasekara, Sachini
Darras, Natasha
Kamarthi, Sagar
Price, Colles
Isaacs, Jacqueline
contents Single-cell representation learning (SCRL) from gene expression data offers a way to uncover the complex regulatory logic underlying cellular function. Inspired by large language models in natural language modeling, several single-cell pretrained models have recently been proposed that treat genes as tokens and cells as sentences. However, these models are fundamentally limited by the long-tailed nature of cell-type distributions and struggle to generalize under covariate shifts in gene expression data. While fine-tuning is often used to mitigate these issues, we observe that performance remains bounded. To address this challenge, we introduce CellRefine, a post-pretraining method that operates between the pretraining and fine-tuning stages of a single-cell foundation model. CellRefine uses a multi-faceted objective that incorporates marker-gene sets as structural priors to guide post-pretraining and refine the latent embedding manifold of cells. Across multiple computational biology tasks, empirical results show that CellRefine consistently improves downstream performance, yielding gains up to 15%.
format Preprint
id arxiv_https___arxiv_org_abs_2605_07938
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Prototype Guided Post-pretraining for Single-Cell Representation Learning
Weerasekara, Sachini
Darras, Natasha
Kamarthi, Sagar
Price, Colles
Isaacs, Jacqueline
Machine Learning
Single-cell representation learning (SCRL) from gene expression data offers a way to uncover the complex regulatory logic underlying cellular function. Inspired by large language models in natural language modeling, several single-cell pretrained models have recently been proposed that treat genes as tokens and cells as sentences. However, these models are fundamentally limited by the long-tailed nature of cell-type distributions and struggle to generalize under covariate shifts in gene expression data. While fine-tuning is often used to mitigate these issues, we observe that performance remains bounded. To address this challenge, we introduce CellRefine, a post-pretraining method that operates between the pretraining and fine-tuning stages of a single-cell foundation model. CellRefine uses a multi-faceted objective that incorporates marker-gene sets as structural priors to guide post-pretraining and refine the latent embedding manifold of cells. Across multiple computational biology tasks, empirical results show that CellRefine consistently improves downstream performance, yielding gains up to 15%.
title Prototype Guided Post-pretraining for Single-Cell Representation Learning
topic Machine Learning
url https://arxiv.org/abs/2605.07938