Within each batch we tested the hypothesis that the means of the fitness distributions for different classes were different. We found
significant differences in fitness betweenIRA1andGPB1,IRA1andIRA2,GPB1andGPB2, and found that mutation type made a
significant difference in the fitness ofIRA1mutants (test was not significant for mutation type inIRA2). For the diploids, we found
that a third copy of chromosome 11 gave significant fitness benefit, but a third copy of chromosome 12 did not. Additional coding
mutations did not significantly change the fitness of diploids.
We also tested the null hypothesis that the distribution ofIRA1mutants was the same as the distributions ofPDE2andGPB2mu-
tants. We used the non-parametric Kolmogorov-Smirnov (KS) test to test for any difference between distributions. The KS test com-
pares the CDFs of two empirical distributions, and compares the largest gap between them (which is distributed in a way independent
of distribution). The fact that the data pass the KS tests as well give us confidence that our results are not due to noise-modeling
assumptions.
The results are also robust to changes in the fitness inference algorithm. If we instead use a weighted log-linear regression,
choosing^sby
bs=arg min
sXiðlogðri+ 1 =Ri+ 1 Þlogðri=RiÞ+misÞ^2
ki=ri+ 1(18)the fitness estimates change by1% per generation at most. The differences in distributions and relative orderings of fitnesses
persist.
DATA AND SOFTWARE AVAILABILITY
Data Resources
All Illumina sequencing data (for both the whole-genome sequencing and the fitness measurement assays) can be found under NIH
BioProject: PRJNA310010.
Software
The software repository for the barcode counting code can be found athttps://github.com/sunthedeep/BarcodeCounter.
Cell 167 , 1585–1596.e1–e15, September 8, 2016 e15