Because these mutants were generated via random mutagenesis, it's difficult to interpret phenotypic variability of each mutant's lipid metabolism in the context of their varied genetic backgrounds. It may be worth showing lipid-productivity data in mutants that have similar levels of starch content to wild-type to show a clear baseline for interpretation.

It would help to show more of the screening process that led to the five chosen lines. For example, reporting how many colonies were screened, the growth conditions used during screening (e.g., BBM vs BBM−N, duration, light), how starch phenotypes were called from iodine staining, and how many candidates fell into each class (low starch / high starch). A semi-quantitative summary of iodine phenotypes would let readers estimate the probability of generating starch phenotypes via this mutagenesis protocol (an informative result on its own).

Here biomass productivity is calculated as u (growth rate) x B_end, but the more common way to calculate BP is B_end - B_start / time. u x B could overstate productivity and/or make comparisons phase-sensitive. Sometimes those early timepoints may be harder to collect/quantify, but is it known that the cultures are in exponential phase for the duration of the experiment?

How many mutants did you have to screen to find five starch high/low phenotypes? Did you also consider mutants that had spatially abnormal starch distribution?

Quite a remarkable result, also supported by the signal-to-noise ratios shown in Figure S7! Was there ever a consideration to combine spectra from all four experimental conditions as a means to find the upper-bounds of classification? Are there technical limitations/statistical legalities that prevent this? From a practical standpoint, collecting multiple spectra is a small price to pay if it leads to phenotypically distinguishable samples/strains/species.

Curious to know if these effects are concentration dependent, particularly in Chlorella Vulgaris which seem to be phenotypically similar with/without .06M NaCL

Measuring gene similarity within each Chromosome (the traditional method of detecting relatedness) would be a very strong supplemental figure to establish a baseline of comparison to GeneCompare. How does the previously biased approach compare to this new unbiased approach?

Because of the high number of match queries, it unlikely that the percent ratios would change much with subsequent runs of GenomeCompare. However to alleviate concerns of algorithm stability, it may be worth running GeneCompare at these three granularities multiple times to add confidence intervals to these ratios.

As mentioned, changing the granularity of chromosome comparisons does not perfectly preserve the rank order of relatedness (eg chromosome 16 going from third lowest at 32bp to lowest at 200bp, while chromosomes 1,6,11 remain ranked third, first and second respectively). However, this isn't necessarily "more" indicative of relatedness. Applying CompareGenome to more species (especially with varied evolutionary histories, genetic architectures, mutation rates, etc), seems like the next logical step (as suggested in the discussion section) towards providing evidence for this claim.