Differential analysis of high-throughput quantitative genetic interaction data

Abstract

Synthetic genetic arrays have been very effective at measuring genetic interactions in yeast in a high-throughput manner and recently have been expanded to measure quantitative changes in interaction, termed 'differential interactions', across multiple conditions. Here, we present a strategy that leverages statistical information from the experimental design to produce a novel, quantitative differential interaction score, which performs favorably compared to previous differential scores. We also discuss the added utility of differential genetic-similarity in differential network analysis. Our approach is preferred for differential network analysis, and our implementation, written in MATLAB, can be found at http://chianti.ucsd.edu/~gbean/compute_differential_scores.m.

Figure 1

The paired experimental pipeline. (a) The pipeline for generating differential genetic interactions is the same as for static genetic interactions except for a split onto treated and untreated plates in the last step. (b) Normalized colony size profiles for the same experimental replicate across the two conditions (blue) have the greatest Pearson correlation, as compared to the profiles of two experimental replicates of the same condition (green) or the profiles of different queries (red). EMAP, Epistasis MAPping.

Figure 2

Theoretical and observed differential variances. Bar plot of the observed static, expected differential (assuming independence), and observed differential variances of normalized colony size residuals. The median values across all double mutants are shown. Bandyopadhyay et al. [6]; Guénolé et al. [7].

Figure 3

Differential profile similarity between SWI4 and HIR. (a) Bar plot showing the Pearson correlation of HIR1/2/3 profiles with SWI4 for untreated (UT), MMS, and differential (dS) scores. (b) Heatmaps of the untreated, MMS, and differential interaction profiles of SWI4 and HIR1; the bottom panel illustrates the interactions with greatest similarity between SWI4 and HIR1.

Figure 4

False discovery rate and reproducibility of the dS score. (a) Plot of the false discovery rate of the dS score as a function of score magnitude. (b,c) Scatter of differential scores calculated on independent replicate subsets using (b) the B scores and (c) the dS score; the points shown in either panel are only those scored by both analyses. (d) Plot comparing the Pearson correlation of significant interactions for the B and dS scores (blue and green, respectively) over a full range of significance thresholds - that is, the correlation of the top n percent of the interactions for n = 0.1% (left side) to n = 100% (right side); error bars (non-bolded lines) indicate the 95% confidence intervals of the correlation coefficient.

Figure 5

Performance of dS score and differential profile similarity. (a-d) Precision-recall plots comparing the biological enrichment of B and dS scores and their corresponding profile similarity scores for DDR interactions (a,c) and co-complex interactions (b,d) using the data from Bandyopadhyay et al. [6] (a,b) and Guénolé et al. [7] (c,d).