Winter Update: POS, Squid Game, and Reading

I wanted to share some updated thoughts/ learnings about the applications of pooled-optical-screening that I wrote about this summer. And, a vignette.

I recently attended the SBI2 conference in Seaport, where I had a chance to learn about the origination and expansion of pooled optical screening tech from Paul Blainey himself. A quick takeaway from his presentation (specifically his response to what I asked him in Q&A) is that my questions about population-level heterogeneity are not as much of a concern when the experiment can survey hundreds of representative cells for a given knockout (not sure whether I am satisfied by that answer). Also, I learned during Ci Chu’s presentation that Insitro’s POSH method was put in place by David Feldman of Paul Blainey’s lab. No wonder the methods are so similar :)

I also wanted to touch on some learnings regarding the proposed utility of this screening platform. For those who are interested, there is currently more application on the target discovery side of the equation. While my questions have to do with connecting morphologies from a small molecule screen to the morphologies (‘morphological spectrum’) shown in the POSH screen, it seems as though uncovering new targets associated with known mechanisms is the current focus. For Insitro in particular, given they lean on DEL screening, this is a good way to start a target-based optimization campaign. Perhaps there will be some interest in the future in developing an arrayed method to make morphological comparisons between knock outs and various therapeutic modalities.

And finally, I recently learned that the POSH method is being adapted into a kit-based protocol, under the name of Bifrost Biosystems (cofounded by Paul Blainey, Johan Paulsson, and George Church). So, it is interesting to see that Insitro’s POSH method is an example of an industry application of the soon-to-be kit assay. For investors who use platform data as a differentiator, how will they look at POSH’s target discovery platform once the kit is commercially available? Perhaps all labs (academic and private) will eventually have access to this data as kit purchases scale (like scRNAseq), renewing academia’s reputation as a hotbed for target discovery.

It will be interesting to see how labs and companies build on top of this setup. I think that the main creative choices labs will have are in the cells (ie. patient avatars) and analysis methods they use. Maybe they can get creative and run the combinatorial experiments I proposed… The thing that I’m thinking about however, is how to link genetic perturbations to small molecule screens (I think this is pretty obvious from what I have written so far). Last, but certainly not least, is how labs will adapt the live imaging setup described in the original paper (2019)… I think different companies/ labs are thinking about this in different ways.

I also wanted to share a vignette of a larger essays to come. I love drug discovery, and find analogies to be really helpful ways to think about the challenges/ opportunities companies face.

The Squid Game Analogy (preview):

In Squid Game, we see challenges serve as filters. If the person can meet a threshold of performance (finish the given challenge(s)), then they can move on to the next filter. If we expanded the starting population to Earth, then we would think that this ‘flow scheme’ would filter for a pool of similar individuals who all possess the same qualities. One important realization is that the order of the filters is actually some kind of bias, and when the hope is that there is to discover a ‘best function’ associated with the end result, then you have actually missed a lot of ‘folks’ who could be on the spectrum of ‘best’. In drug discovery, if you have ‘order bias’ then that means that you are excluding potential molecular starting points that might actually be more easily refined to meet your end criteria.