Computer Vision News - May 2018

26 Friday ISBI DAILY The biggest product of my lab is definitely CellProfiler. It’s used by thousands of biologists, but I wouldn’t call it a breakthrough. There’s nothing technologically novel in CellProfiler as I originally wrote it, but it was important in this cultural shift of helping biologists to quantify things. The most important innovation from my lab has probably been the development of the Cell Painting assay, which is a protocol for biologists to prepare cells. It’s a certain combination of dyes that they add to the cells which allow them to see a lot of different structures within a cell in order to make these fingerprints that I mentioned earlier. You can characterise, for example, what is the state of the cell if you treat it with this drug or that drug? What’s the state of this patient’s cells versus that patient’s cells? That technology is now being adopted by pharmaceutical industries for all kinds of different applications. One is determining how a potential new drug works - to try and figure out its exact mechanism, which is an important step towards a drug getting on the market. Another is predicting the activity of drugs in a computational way, as opposed to having to do a lot of experiments. There’s even a pharma company, Recursion, that was launched on the basic idea of using this assay to identify the cellular impact of various disease genes. That company is well underway. You are also giving me super interesting answers, I should say! [ laughs ] Do you think there is an area in which our profession is not doing enough, that might give a pointer to students to focus on? Right now it seems like there’s enough momentum that neural networks are going to do a good job at a lot of tasks that we care about such as image segmentation and this image-based profiling that I’ve been talking about. I’m confident that’s going to develop over time, so it’s great if people want to work on that. The part that to me seems more of a blocker at present is coming up with clever ways to help us to understand what a given network is using to make its decisions. We find for many biological experiments, it’s not enough to come up with a perfect classifier that can distinguish cells of type A from cells of type B, we need to understand why the two types of cells are different from each other in Anne Carpenter “… a protocol for biologists to prepare cells. ” With 4-months-old daughter Elle BEST OF