Perfecting PDD in 2023: Your Team’s New Process Philosophy
This essay introduces the 'heterogenous reaction environment' and ideas for team-oriented standardization thereof. In the end, it's all about the patients.
If you want upstream phenotypic data to have the most predictive ability/ value, then the reaction environment must be as reproducible and heterogenous as possible. For phenotypic assays, the reaction environment is a well of seeded cells, and heterogeneity is required so you can measure nuanced biological differences between closely resemblant perturbagens. Perhaps the heterogeneity should resemble human disease most closely (via co culture etc.). But maybe conventionally defined “therapeutic window” is misleading, and maybe mutational background is also misleading.
Maybe you should simply pick a handful of different cell types, that represent different parts of the body, and just focus on which trends in your primary screen help you predict DFAs. Reproducibility is required, so that these nuanced differences in your heterogeneous environment are believed to be the truth. If you run your experiment multiple times in the same plate, in various plates within the same run, or in different runs, the differences you measure in your cells should connect to whatever standard statistical robustness metrics are agreed upon.
Today I will talk more about reproducibility, and in a few days I will have some postulates about the ‘heterogenous reaction environment’ and analogies that exist in the tech world (ie. Instagram and Tik Tok). Before writing specifically about opportunities for workflow optimization (as I said I would in Perfecting PDD in 2023: Motivation), I want to pose a question: What if the water levels in your incubators were the reasons why your sophisticated AI/ ML models failed to find correlates between images and DFAs? To dig a bit deeper into this question, perhaps the reason why you couldn’t find the correlates is because there was so much variation from batch to batch, due to humidity differences run to run, which influenced cell growth kinetics pre and post compound treatment, that you had to perform “whitening” data transformations prior to performing subsequent analysis across all collected data.
I’m asking this kind of question because if the answer is YES, then what would you do? Would you monitor your water levels and try to normalize data according to these measurements, or would you try to physically tweak your incubator setups from run to run by topping off the water to ‘max’ volume before a new run. A more radical approach may be to drain and sterilize your incubators at the end of every run and fill them to the ‘max’ before every subsequent new run. In the end, maybe it’s not just your incubators that need resetting, but your process philosophy as a whole! The approach of fully reseting segments of workflows is one principle I hope AI/ ML-enabled PDD companies will follow, given the challenges of maintaining phenotypic reproducibility in assays.
The next principle I have in mind has to the do with the following question: Is it better to have small variations from batch to batch (many batches with small variation), or fewer batches with greater variation between them? Is the ultimate goal then to have something like fewer batches (largest batches possible) with the smallest batch-to-batch variation? Perhaps that is what your teams should be striving for. Maybe the largest batch possible, and incubator maintenance, requires team mates to follow 3 shifts and work 6 days in a row. What if the benefits are that you will have reproducibly captured the nuanced phenotypic richness of your cells, and sped up your discovery process by months or even years? What would it take for your team to believe these stringent parameters were worth following?
These kinds of process-related sacrifices are only possible with teamwork and total buy in regarding process standards. If even one team mate does not believe that the standards are worth it, what they mean is that saving the patients is not worth it to them. You should assess how much the patients matter to your team, before beginning such an operation to standardize the heterogeneous reaction environment, because it might literally take your blood, sweat, and tears to do so! But again, buy in means that the whole team has agreed that the rules in place are worth following from beginning to end, so flashy proof of concept may be required and/ or agreement to compromise if it is believed that particular rules are unnecessary.
If it’s unclear what the primary goals of phenotypic drug discovery are, beyond doing anything and everything to save the patients, please read From Plate to Patient. And finally, on a more serious note, wouldn’t it be true that it would be easiest to make these process-related decisions if you had as few people on the team as possible? What if the 3 shifts and no compromising on workflow rules required no conversation at all? Enter, High Res Biosolutions, your source for end-to-end assay automation.
Stay tuned for my next essay talking more about their offerings and my hopes for their future.