Causal analysis approaches in Ingenuity Pathway Analysis

Abstract

Motivation: Prior biological knowledge greatly facilitates the meaningful interpretation of gene-expression data. Causal networks constructed from individual relationships curated from the literature are particularly suited for this task, since they create mechanistic hypotheses that explain the expression changes observed in datasets.

Results: We present and discuss a suite of algorithms and tools for inferring and scoring regulator networks upstream of gene-expression data based on a large-scale causal network derived from the Ingenuity Knowledge Base. We extend the method to predict downstream effects on biological functions and diseases and demonstrate the validity of our approach by applying it to example datasets.

Availability: The causal analytics tools 'Upstream Regulator Analysis', 'Mechanistic Networks', 'Causal Network Analysis' and 'Downstream Effects Analysis' are implemented and available within Ingenuity Pathway Analysis (IPA, http://www.ingenuity.com).

Supplementary information: Supplementary material is available at Bioinformatics online.

Fig. 1.

Overlap P-value (A) and activation Z-score (B) calculation (see text). In (B) the pointed arrowheads represent activating relationships, and the blunt arrowheads represent inhibitory relationships

Fig. 2.

Enrichment of ‘causal transitive triangles’ (A) indicates causal dependency of upstream regulators A and B [compare (B) versus (C); see text]

Fig. 3.

Replacing multi-step paths from root regulators to target genes (A) by ‘virtual’ edges with the same net effect (B). The pointed arrowheads represent activating (+1) relationships, and the blunt arrowheads represent inhibitory relationships (–1). The dashed lines indicate virtual relationships composed of the net effect of the paths between the root regulator and the target genes

Fig. 4.

URA result table for example (1) in Section 5.1 (beta-estradiol-treated MCF-7 cells)

Fig. 5.

Mechanistic network for beta-estradiol [example (1) in Section 5.1]. In this network, beta-estradiol is postulated to activate ESR1 (the estrogen receptor), NCOA3 (a key estrogen receptor co-regulator) and to affect a number of other regulators to explain the gene-expression changes in the dataset. The set of regulators in total connect to 320 dataset genes (not shown), with beta-estradiol connecting directly to 183 of them

Fig. 6.

DEA results for example (2) in Section 5.2 (TNF-stimulated HUVEC cells). The visualization is a TreeMap (hierarchical heatmap) where the major boxes represent a family (or category) of related functions. Each individual colored rectangle is a particular biological function or disease and the color orange indicates its predicted state: increasing (orange), or decreasing (blue). Darker colors indicate higher absolute Z-scores. In this default view, the size of the rectangles is correlated with increasing overlap significance (using FET P-value). The image has been cropped for better readability

Fig. 7.

CNA result for SOST (see Section 5.3). SOST is the master regulator (or ‘root’ regulator) of a small causal network containing five intermediate regulators that may explain the up and down regulation of the 26 dataset molecules shown in the bottom row (red indicates up-regulation and green down-regulation). The regulators are colored by their predicted activation state: activated (orange) or inhibited (blue). Darker colors indicate higher absolute Z-scores. The edges connecting the nodes are colored orange when leading to activation of the downstream node, blue when leading to its inhibition, and yellow if the findings underlying the relationship are inconsistent with the state of the downstream node. Pointed arrowheads indicate that the downstream node is expected to be activated if the upstream node connected to it is activated, while blunt arrowheads indicate that the downstream node is expected to be inhibited if the upstream node that connects to it is activated