henhaobio
Sign in

bioRxivBioinformaticsPreprint

Unbiased single-cell morphology with self-supervised vision transformers

Michael DoroniD, Théo MoutakanniiD, Zitong CheniD, Nikita MoshkoviD, Mathilde CaroniD, Hugo TouvroniD, Piotr BojanowskiiD, Wolfgang M. PerniceiD, Juan C. CaicedoiD

PDFDOI

Abstract summary

Claude Opus 5.5 · xhigh effort · 18 Oct 2026 · from the abstract · CC BY-ND 4.0

Self-supervised DINO features, learned without labels, describe cell morphology from subcellular detail to experimental groups and expose a hierarchy of biological and technical variation in three public datasets.

Measuring cell morphology is described as an active research area that has produced many computer vision algorithms, and quantifying it accurately at scale would strengthen existing single-cell approaches. The claim here is that DINO, a self-supervised vision transformer, learns rich morphology representations with no manual annotation or any other supervision.

Key results

  1. ClaimedDINO embeddings capture morphology at several levels at once: subcellular structure, individual cells, groups of cells, and whole aggregated experimental conditions.

    Reported from evaluation on many tasks spanning three public imaging datasets with different specifications and biological focus; the abstract names no metric, score or competing feature extractor. · Abstract

  2. ClaimedThe learned representation organises imaging datasets into a hierarchy that mixes biological factors with technical ones.

    Stated as a finding across the three datasets, with no quantification of how the factors separate and no comparison against supervised or hand-crafted descriptors. · Abstract

  3. ClaimedThe authors position the embeddings as a way to study biological variation that is not known in advance, including single-cell heterogeneity and relationships between samples.

    Framed as a conclusion from the evaluations; no discovery result, validation experiment or downstream biological finding is reported in the abstract. · Abstract

Methods, data and reproducibility

DINO, a self-supervised vision transformer trained without manual annotation, is used as a feature extractor for cell images, and its representations are evaluated on a range of tasks across several publicly available microscopy datasets.

  • Publicly available cell imaging datasets of differing specifications and biological focus · n = three datasets
Methods & tools
DINO, vision transformer
Reproducibility gaps
The imaging datasets are described as publicly available but are not named or linked, and no code repository or model weights are mentioned.
Limitations (2)
  • Abstract only: no quantitative results and no baseline comparison against supervised models or classical morphology features, so the size of the advantage is unknown. · noted in this summary
  • Evaluation is limited to three imaging datasets, and the datasets, imaging assays and organisms involved are not identified. · noted in this summary

Read this if you profile cells from imaging data and want to know whether self-supervised features learned without annotation capture morphology from subcellular to aggregated scales.

Citations
46

Abstract

Accurately quantifying cellular morphology at scale could substantially empower existing single-cell approaches. However, measuring cell morphology remains an active field of research, which has inspired multiple computer vision algorithms over the years. Here, we show that DINO, a vision-transformer based, self-supervised algorithm, has a remarkable ability for learning rich representations of cellular morphology without manual annotations or any other type of supervision. We evaluate DINO on a wide variety of tasks across three publicly available imaging datasets of diverse specifications and biological focus. We find that DINO encodes meaningful features of cellular morphology at multiple scales, from subcellular and single-cell resolution, to multi-cellular and aggregated experimental groups. Importantly, DINO successfully uncovers a hierarchy of biological and technical factors of variation in imaging datasets. The results show that DINO can support the study of unknown biological variation, including single-cell heterogeneity and relationships between samples, making it an excellent tool for image-based biological discovery.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISPfaendler '23Zhang '26Lu '19Murthy '25Wu '22Lavaee '26
66% semantic matchSystems Biology
Ramon Pfaendler, Jacob Hanimann, Sohyon Lee+1 more·bioRxiv·Jan 16, 2023· 26 citations