A cell’s fate is often decided long before any stain or reporter lights up — and it turns out the decision is visible in the cell’s shape and motion, if you know how to look. Today’s digest is about deep learning that forecasts fate, cycle, and behavior from live, often label-free, microscopy. Van Valen et al.’s DeepCell first made it possible to ‘segment the cytoplasms of mammalian cells … without a fluorescent cytoplasmic marker.’ Buggenthin et al. then predicted blood-lineage choice ‘up to three generations before conventional molecular markers are observable.’ Waisman et al. spotted stem-cell differentiation ‘just 20 min after the onset.’ Kegeles et al. called retinal-organoid fate ‘before the onset of reporter gene expression,’ beating a human expert. Eulenberg et al. reconstructed the cell cycle ‘on-the-fly.’ And Zaritsky et al. built an interpretable model that reads metastatic potential from label-free images — surfacing ‘pseudopodial extensions and increased light scattering’ too subtle for a human eye. The cell tells you where it’s going.
