Exploiting time to phenotype cell morphology

<p dir="ltr">• The morphology and appearance of cells in time-lapse microscopy can be diagnostic for biological state.<br><br>• Classical measures of morphology offer an incomplete and poorly-aligned read-out for these presumed states.<br><br>• In computer vision, deep neural networks can produce rich representations of natural images that align with language models, but it is unclear how to align these representations to the underlying biology.<br><br>• Many biological states of interest vary only slowly in time, so we hypothesised that neural networks trained for time-consistency could better align with such states by learning to ignore short-term change like reorientation of organelles or drift in the focal plane.<br><br>Poster presented as part of the Crick BioImage Analysis Symposium 2025.<br>Permission has been given by authors to upload to Crick Figshare.<br><br>Copyright remains with the original authors.</p>