Cell-DINO: Self-supervised image-based embeddings for cell fluorescent microscopy

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 DINOv2, 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 apply DINOv2 to cell phenotyping problems, and compare the performance of resulting models, called Cell-DINO models, on a wide variety of tasks across two publicly available imaging datasets of diverse specifications and biological focus. Compared to supervised and other self-supervised baselines, Cell-DINO models demonstrate improved performance, especially in low annotation regimes. For instance, to classify protein localization using only 1% of annotations on a challenging single-cell dataset, Cell-DINO performs 70% better than a supervised strategy, and 24% better than another self-supervised alternative. The results show that Cell-DINO can support the study of unknown biological variation, including single-cell heterogeneity and relationships between experimental conditions, making it an excellent tool for image-based biological discovery.Author summary: Cellular imaging is widely used in biological studies, and often enables quantitative analysis of phenotypic variation through the use of machine learning algorithms. In this work, we introduce Cell-DINO , an adaptation of the DINOv2, self-supervised representation learning strategy, and demonstrate its ability to produce better features for cell phenotyping problems. On the Human Protein Atlas (HPA) dataset, our experiments show the versatility of the approach, outperforming a supervised vision transformer strategy by 20% on average and by 70% when using only 1% of the labels. Compared to other self-supervised strategies, Cell-DINO outperforms MAE and SimCLR on the most challenging tasks, for instance protein localization classification on the HPA single-cell dataset, with 34% and 20% improvement in performance, respectively. Cell-DINO can also be useful to quantify the effects of pharmacological compounds at high-throughput, where the phenotypes of interest cannot be annotated by hand. Cell-DINO improves the ability to predict the mechanism of action of compounds using Cell Painting images, outperforming SimCLR, MAE, and supervised CNN baselines, as well as CellProfiler.