Comparing Multiple Clones to Clone A and E

[gwaybio]

Previously, @mekelley and @ayberman sequenced Clones A and E and observed PSMB5 mutations (see #49 (comment)).

One goal is to determine if any of the four resistant clones (BZ001, BZ002, BZ003, and BZ004) also harbors a PSMB5 mutation based on a signature of morphology. The initial approach was to see where clone A and clone E clustered in comparison to each of the four "groups" we previously observed in #25. However, those groups appear to be artifactual - see #51.

An additional wrinkle in comparing the six clones, is that there appears to be batch effects between the batches analyzed. We do not observe substantial batch effect between BZ001, BZ002, BZ003, and BZ004 clones which were measured across 3 different batches (which totaled 5 different plates) (see #48). We also do not observe substantial batch effects between the two different batches (2 total plates) that measured clone A and clone E (batch 1 and 2). See below, which was added in #52

However, we do observe batch effects when combining the two groups of data (also added in #52):

Note that we can visually inspect for batch effects in this reduced dimension space (UMAP). It is not a great way to detect batch effects, but usually when we see batches cluster separately, especially when they both contain wild-type parental profiles, it hints towards a batch effect.

The current normalization procedure calculates z-scores for all samples. We can explore if alternative normalization procedures overcomes this batch effect.