The other two cases were only 3 stains, but this one used all 5 seen in CellPaint. If brightfield can really extract this much information using strictly ML methods it means it should be able to extract information about other molecular locations in the cell that are not normally stained, making it better than CellPaint for reasons other than fixation/spectrum limitations.
Awesome post, Abhish. This was a really fun story. You distilled the complicated portions masterfully. I love following Recursion's story and can't wait to see what they build with their painted, un-painted datasets.
This is your strongest post yet. Really interesting and well-explained about a subject that I'm not that familiar with.
Perfect timing for me! Thank you for that case study I hadn't seen! (https://doi.org/10.1371/journal.pcbi.1011323)
The other two cases were only 3 stains, but this one used all 5 seen in CellPaint. If brightfield can really extract this much information using strictly ML methods it means it should be able to extract information about other molecular locations in the cell that are not normally stained, making it better than CellPaint for reasons other than fixation/spectrum limitations.
Nice article. Just FYI, the brightfield image is actually several cells and the video you posted is not brightfield, it's holotomography.
My bad, didn't realize that, thank you for letting me know!
Awesome post, Abhish. This was a really fun story. You distilled the complicated portions masterfully. I love following Recursion's story and can't wait to see what they build with their painted, un-painted datasets.