Wondering about applicability of OpenPhenom-S/16 to non-homogenous mixtures of cells

[mat10d]

Hello, I am working with data where the cells in a field of view/tile are not all of the same perturbation. Could I apply OpenPhenom-S/16 to get single-cell measurements of these cells? Have you tried this in the past? Essentially, we are talking about a tile based image (1200x1200) with 4 channels and with cellular segmentation masks. Would love to chat more to see if we can integrate this into our pipeline. Thank you!

Matteo

[kian-kd]

Hi @mat10d! Cool usecase! I think a cell-segmentation mask would work well! For example, you could leverage perturbation-specific masks that include at inference time so that the model only looks at the cells that are relevant to a specific perturbation. This should work well, the model is great at processing subsets of unmasked pixels (that's what it was trained for)!