A Gaussian process-based definition reveals new and bona fide genetic interactions compared to a multiplicative model in the Gram-negative Escherichia coli

Abstract

Motivation: A digenic genetic interaction (GI) is observed when mutations in two genes within the same organism yield a phenotype that is different from the expected, given each mutation's individual effects. While multiplicative scoring is widely applied to define GIs, revealing underlying gene functions, it remains unclear if it is the most suitable choice for scoring GIs in Escherichia coli. Here, we assess many different definitions, including the multiplicative model, for mapping functional links between genes and pathways in E.coli.

Results: Using our published E.coli GI datasets, we show computationally that a machine learning Gaussian process (GP)-based definition better identifies functional associations among genes than a multiplicative model, which we have experimentally confirmed on a set of gene pairs. Overall, the GP definition improves the detection of GIs, biological reasoning of epistatic connectivity, as well as the quality of GI maps in E.coli, and, potentially, other microbes.

Availability and implementation: The source code and parameters used to generate the machine learning models in WEKA software were provided in the Supplementary information.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Performance measurements of GP over other GI definitions. (A) Error values shown for the indicated definitions using training and test pairs compiled from the genome integrity study (Kumar et al., 2016). (B, C) GI (ε) score distributions of training (B-i), test (B-ii) and for all (C) gene pairs plotted against various definitions with normal distribution centered around zero (dotted line), and with tails signifying aggravating (-ve) and alleviating (+ve) interactions. The zoom in of the tails is shown as an inset in the top-right corner. The overlap of aggravating or alleviating GIs between GP and the multiplicative definition is shown as a Venn diagram (C) in the inset. (D) Box plots showing the distribution of GI (ε) scores for gene pairs classified differently by the GP and the multiplicative definitions (grouped under various categories, see text), and their respective normalized single (Wx: donor; Wy: recipient) and double (Wxy)-mutant colony sizes that exhibit slow (< 1.0), normal (1.0) and fast (>1.0) growth

Fig. 2.

Assessment and validation of GI pairs from GP definition. (A) Accuracy and error rate of the GI definitions estimated on 800 test pairs compiled from the genome integrity study (Kumar et al., 2016). (B) Percentage gene pairs defined as aggravating (-ve), alleviating (+ve) and noninteracting under GP definition enriched (FDR-corrected P-value ≤ 0.05; enriched GO term in the ranked gene set was assessed by Kolmogorov–Smirnov test) for a shared Gene Ontology (GO) term function. (C) Percentage gene pairs with epistatic relationships between orphan (gene of unknown function)–orphan, annotated–annotated or orphan–annotated genes. (D) Confirmation of gene pair classifications by the GP model that were either in agreement or in disagreement with the multiplicative model. For this comparison, the top 25% of the ranked GI pairs were selected. GI (ε) score ≤ −0.152 for GP and ≤ −0.221 for the multiplicative model are considered as aggravating (-ve sign), and ≥0.153 for GP and ≥0.203 for multiplicative as alleviating (+ve). Scores that did not meet the chosen aggravating or alleviating threshold were considered as noninteracting (with no sign indicated). The fitness of the double-mutant colonies and their respective GI scores from the indicated model should dictate their predictive capability of identifying reliable GI type. Hfr Cavalli donor (nonessential or *essential hypomorphic) single mutant marked with chloramphenicol (Cm; i) and F⁻-recipient single-mutant (nonessential or *essential hypomorphic) strains marked with kanamycin (Kan; ii) were crossed and selected on plates with Cm and Kan to generate double mutants (iii), which are shown along with their respective scores from GP and multiplicative definitions below the colony images. Multiple colonies represent replicates. Functionally unrelated gene JW5028 was used as a control. (E) Sensitivity or resistance (i) of the indicated deletion mutant strains to DNA-damaging agents that induce double-strand breaks, as reported in a chemical-genetic screen (Nichols et al., 2011). Growth sensitivity (ii) is shown for mutants and wild-type (WT) cells grown on bleomycin (BLM) and MMS. Phenotypic complementation shows the overexpression of yicR in trans. (F) Cell morphology micrographs (i) and cell lengths (ii) of WT and mutant strains before and after DNA damage with MMS (t = 2 h) treatment. Cell lengths of the WT or mutant strain is an average of ∼100 different cells; * indicate significant (Student’s t-test) difference between double mutants and their single mutants. Scale bar equals 10 μm