Prediction of phenotype and gene expression for combinations of mutations

Abstract

Molecular interactions provide paths for information flows. Genetic interactions reveal active information flows and reflect their functional consequences. We integrated these complementary data types to model the transcription network controlling cell differentiation in yeast. Genetic interactions were inferred from linear decomposition of gene expression data and were used to direct the construction of a molecular interaction network mediating these genetic effects. This network included both known and novel regulatory influences, and predicted genetic interactions. For corresponding combinations of mutations, the network model predicted quantitative gene expression profiles and precise phenotypic effects. Multiple predictions were tested and verified.

Figure 1

Outline of modeling strategy. (A) Overall modeling strategy. (B) Genetic influences decomposition and corresponding networks for a simplified system of two seed genes, A and B (circles), influencing the expression of three target genes, X, Y, and Z (boxes). The data matrix is decomposed in terms of influence and activity variables, corresponding to all possible influences in the network. (C) Illustration with synthetic expression data of three genes in four strain backgrounds. Positive and negative numbers in the influence matrix with magnitude greater than the significance cutoff (0.1) map to green and red edges in the network, respectively. Black numbers correspond to expression data and activities fixed by genetic backgrounds (all wild-type activities are 1, the activity of the deleted gene A is g_A^A=0, etc), while the red numbers are the best-fit solution of the system. In this example, the genotype matrix element of gene B in the A-deletion strain is reduced (g_B^A=0.5), from which a positive influence from A to B is deduced and shown in the network.

Figure 2

Directed data integration: modeling how TEC1, CUP9, and SKN7 control the expression of gene DDR48. (A) Network of inferred influences that cause genetic interactions. Edges indicate the direction of positive (green) and negative (red) influences, with intensity indicating magnitude of influence. (B) Network of physical interactions connecting the four genes from high-throughput data sets. This network is too dense and disorganized to identify functional pathways. Interactions are protein–protein (blue), protein–DNA (orange), and protein phosphorylation (violet). (C) Integrated network constructed from the subset of molecular paths in (B) that are specific candidates for transmission of influences in (A). Influences from the remaining seed genes (SOK2 and SFL1) have been omitted for clarity.

Figure 3

SKN7-positive influence network. An influence network is shown that maps putative molecular paths of influence from SKN7 to target genes, which are shaded green in proportion to their influence coefficient. White nodes are proteins that fall on the shortest directionally consistent putative paths of influence from SKN7 to the enriched transcription factors shown in yellow. Interactions are colored as: protein–protein in blue and protein–DNA in orange.

Figure 4

Seed gene influence network, filamentation-specific molecular network, and topological motifs. (A) Genetic influences between the seed genes inferred from genetic interactions. Green (red) arrows represent positive (negative) influence on regulatory activity. Color intensity is proportional to influence magnitude. (B) Mode-2 molecular network. The green box represents the Mode-2 genes with positive expression influences from the three topmost seed genes (TEC1, CUP9, and SFL1). Seed genes influence each other as in (A). Yellow nodes are transcription factors with enriched binding targets among the influenced genes. White nodes are proteins that fall on the shortest directionally consistent putative paths of influence from each seed gene to yellow transcription factors. Interactions are colored as: protein–protein in blue, protein–DNA in orange, and protein phosphorylation in violet. Black arrows denote inferred influences for which no molecular path with fewer than five interactions was found. Genes TEC1, CUP9, and MGA1 are themselves members of the Mode-2 gene set. (C) Network topologies. The three network motifs and corresponding phenotype predictions for novel double-mutant pairs in (B). Labeled nodes denote deleted genes and inequalities represent phenotype predictions. Gray nodes represent genes that influence the Mode-2 gene set (green box) and white nodes represent candidates for transmission of the influences. Edges represent paths of influence involving any number of nodes and physical interactions.