Franziska Jessica Schöb

• Schöb, Franziska Jessica; Binder, Alexander; Zamarian, Valentina; Sordi, Valeria; Scholz, Hanne & Malthe-Sørenssen, Anders [Vis alle 7 forfattere av denne artikkelen] (2026). Deep learning for predicting stem cell efficiency for use in beta cell differentiation. Scientific Reports. 16(1). doi: 10.1038/s41598-026-42830-3. Fulltekst i vitenarkiv Vis sammendrag

  Recent clinical trial data show curative potential of cell therapy for diabetes, however the cells required are a bottleneck. Cell differentiation exhibits substantial variability, even among clones of stem cells generated from the same patient. Human experts struggle to see the difference between highly- and lowly-efficient cell clones early. We therefore propose an image-based deep learning model to guide the selection of the most efficient clones. We apply different deep learning models to learn the morphological differences between good and bad stem cell clones and classify them based on phase-contrast imaging. To gain insight into the learned features, we use layer-wise relevance propagation, and Fourier-based frequency analysis. Using an EfficientNet-V2-S model, we obtain a novel early prediction for the outcome of the differentiation process from patient-derived stem cells to β -cells using imaging. Clone level accuracy is 96.7 % at 53 hours after start of differentiation. The analysis of learned features shows that the structure of the cell population is an important predictive feature. This study is a proof-of-concept that deep learning combined with label-free imaging can be highly predictive and guide selection of stem cell clones, thereby reducing cost of β -cell production.

Se alle arbeider i NVA

• Schöb, Franziska Jessica; Dysthe, Dag Kristian; Scholz, Hanne; Binder, Alexander & Malthe-Sørenssen, Anders (2026). From Pixels to Phenotypes: Deep Learning on Label-Free Microscopy Data for Cell Fate Prediction How Algorithms Help Us See More. Universitetet i Oslo. Fulltekst i vitenarkiv Vis sammendrag

  In this thesis, we used supervised deep learning on label-free microscopy data to predict cell fate and state from purely visual information, without relying on fluorescent labeling or manual segmentation. Across three studies, we combined phase-contrast and holotomographic imaging with convolutional neural networks to tackle problems that are central to stem cell biology and cancer research: predicting the differentiation efficiency of induced pluripotent stem cell clones, identifying germ-layer lineages in differentiating cultures, and detecting macrophage-induced cancer cell death. By building segmentation-free pipelines that operated directly on image patches, we reduced the need for explicit cell masks and made our methods easier to deploy in realistic, noisy experimental settings. The overarching results showed that biological label-free images contained sufficient morphological information for deep networks to recover biologically meaningful patterns, but also revealed clear limits to what static morphology could predict. In stem cell differentiation, our models already distinguished high- from lowefficiency clones early in the protocol, and learned image cues that matched intuitive notions of colony coverage and uniformity. In germlayer prediction, endpoint-trained models decoded lineage identity from holotomographic morphology and reconstructed a plausible timeline of lineage divergence when applied retrospectively, while struggling more in mixed cultures with overlapping lineages. In the cancer cell death study, deep learning distinguished dying from alive cells at the time of death, yet early prediction beyond a short pre-death window remained difficult, indicating that death morphology emerged abruptly rather than gradually over many hours. Together, these findings support the view that deep learning on label-free microscopy could provide non-destructive, scalable readouts of cell fate, while emphasizing the need for time-aware models, richer training labels, and improved domain generalization to fully exploit biological label-free imaging in biomedical workflows.

Se alle arbeider i NVA