We no longer observe four clear "groups" in batch 3 data

[gwaybio]

Previously, as described in #25, we observed four distinct groups in batch 3 data. Interestingly, these four groups seemed to have similar representation of resistant and wildtype clones.

In #25, I describe results from a subcluster analysis that was performed to determine if any features were consistently different between resistant and wild-type clones. As shown here #25 (comment), we did not observe many features with consistent differences. This suggests that core differences between wild-type and resistant clones is not consistent across these groups.

Additionally, it was determined that one driving force behind the groupings was because of wonky normalized profiles, primarily b/c of costes features. So, starting in #32, I began reprocessing all profiles using a unified pipeline based on pycytominer. In an upcoming pull request, I will add all software, data, and results, but I will show below that we no longer observe four clear groups. There is still definitely interesting structure in the data, but the groupings are not as striking, and an analysis that uses these groups is not likely to find consistent and reproducible patterns.

Batch 3 Data

Note that batch 3 profiles were all untreated. For more details on available data, see #40.

Conclusion

Recently, in #49, it was hypothesized that clones A and E (batch 1 and 2) data may harbor similar resistance features as other resistant clones (BZ001, BZ002, BZ003, and BZ004). We know that clones A and E have PSMB5 mutations, and these mutations confer bortzeomib resistance (Lu and Wang). If we can link morphology signatures of A and E clones to the other resistant clones, this may suggest that other resistant clones also have PSMB5 mutations.

One way to perform this analysis is to see where clones A and E fall within each "group". Since the grouping structure does not appear to be as consistent as before, I think an alternative strategy will be more effective. I will outline an alternative strategy in a separate issue!

[AnneCarpenter]

I'm honestly relieved... I didn't love the "group because they look similar though we don't know why and then look for differences within each group" approach!