Figure 1.
GOGANPA transforms a GO similarity matrix into a gene network. Gene weights are then evaluated for each pathway (represented by transparent coloured boxes), and the weights are integrated into the gene expression data to evaluate the test statistics and weighted pathway test statistics.
Figure 2.
Node sizes correspond to the gene weights evaluated by GOGANPA. The functional-centrality of the TP53 gene is highlighted by being assigned a high weight. The CDKN1A gene, a gene responsible for cell-cycle regulation and DNA-damage response, also receives a high weight due to its functional significance within the P53 Pathway.
Figure 3.
(A) The gene-of-interest (yellow node) is connected to certain single-protein-coding genes and two groups of MSP-coding genes (inside blue shades). The presence of MSP-coding genes inflates both and
for the gene-of-interest inside the pathway shaded in yellow. (B) Upon collapsing the MSP-coding gene groups into single units, both
and
are reduced at the protein level.
Table 1.
p53 Data – Results.
Figure 4.
Deeper colour represents stronger differential expression (i.e. higher ). Grey nodes represent genes with missing expression measurements. Node sizes correspond to the gene weights evaluated by GANPA (A) and GOGANPA
(B). Comparing to GANPA, while GOGANPA
has down-weighted the differentially expressed BCL2 gene and MAPKAPK2 gene, it has up-weighted the differentially expressed FAS, TNF, and IL1A genes, and has hence produced a higher pathway test statistic and a smaller
-value for the HSP27 Pathway.
Figure 5.
See caption of Figure 4 for descriptions. The highly differentially expressed BAX gene, considered less important by GANPA (A), has been strongly up-weighted by GOGANPA (B), allowing GOGANPA
to discover the ceramide pathway’s significance.
Table 2.
Breast Cancer Data – Results.
Figure 6.
Breast Cancer Data: CDC20 Pathway.
See caption of Figure 4 for descriptions. (A) In GANPA, a huge amount of co-expressing genes-pairs form a strongly connected network, and GANPA cannot distinguish the highly differential genes from the other less differentially expressed genes. (B) GOGANPA, on the other hand, only considers functional relationships, and hence provides a much sparser network that highlights the importance of the highly differentially expressed UBE2C and CDK1 genes.
Table 3.
Asthma Data – Results.
Figure 7.
See caption of Figure 4 for descriptions. (A) The GANPA network. (B) The GOGANPA network.
Figure 8.
Asthma Data: Basigin Interaction Pathway.
See caption of Figure 4 for descriptions. (A) The GANPA network. (B) The GOGANPA network.