
Ities across AT. The basic interest of MFA as…
Ities across AT. The basic interest of MFA as a data integration procedure [22] was to ensure that each tissue influence was equally weighed in a manageable number of comprised factors, which can then be related to external phenotypic characteristics. Figure 2 shows the diagnostic MFA plot. The first dimension (Dim1) of the MFA accounted for Gefapixant 31 of the variation, and PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/28495172 correlated significantly to variations in AT weights (r = 0.78 with PRAT, and r = 0.76 with SCAT, p < 0.001), so PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/28495082 that pigs fed the HF diet and pigs fed the LF diet were oppositely represented along Dim1. The second dimension of the MFA (Dim2) explained 8.2 of the variation and did not clearly separate the data according to diet or adiposity phenotypic traits. As illustrated (Fig. 2), the two first comprised variables synthesizing the molecular information of the originallydistinct data tables (dim_1_PRAT and dim_1_SCAT, respectively) were projected together along Dim1 in the MFA plot. This means that large similarities betweenGondret et al. BMC Genomics (2016) 17:Page 4 ofFig 2 (See legend on next page.)Gondret et al. BMC Genomics (2016) 17:Page 5 of(See figure on previous page.) Fig 2 Multi-way datasets analysis: consensus in microarray data relative to dietary effect across two adipose tissues. The first two synthetic variables obtained for the perirenal (Dim_1_PRAT) and subcutaneous adipose tissue (Dim_1_SCAT) transcriptomes were projected in the correlation circle of the multiple factor analysis (MFA), an integrated statistical method used to reveal communalities across separate datasets. Large similarities across adipose tissues can be deduced from molecular variables contributing to the first dimension (Dim1) of MFA (Fig. 1a). Relative weights of perirenal fat ( PRAT) and subcutaneous fat ( SCAT) were superimposed on the plot, showing strong correlation between the molecular probes contributing to Dim1 and adiposity variations. Pigs were represented on the scatter MFA plot (Fig. 1b) and colored following the diet they received (HF: high-fat high-fiber; LF: low-fat high-starch). This shows a well-defined partition of pigs between diets along DimATs could be inferred by focusing on the DEP that contributed mainly to Dim1. A total of 1,128 DEP showing the highest correlation with Dim1 (r > |0.70|; p < 0.001) were thus commonly regulated by diet across the two ATs. The identity of these DEP together with their fold-changes between HF and LF diets are listed in Additional file 1: Table S1 for PRAT and for SCAT, respectively; as indicated in this additional file, all these DEP were altered by diet (BH adjusted p-value < 0.05 in PRAT and BH adjusted p-value < 0.08 in SCAT). They corresponded to 436 unique differentially-expressed genes (DEG). This statistical approach was thus useful in reducing the number of DEG commonly regulated by diet across the two ATs to those having the strongest correlation with variations in AT weights. A functional analysis was performed to understand the biological meaning behind this subset of 436 DEG, by using automatic functional annotation tools to identify biological gene ontology (GO) terms and clustering redundant annotation terms in enriched biological pathways. Enrichment (E) score 1.3 and p-value < 0.5 were used to select the top-enriched clusters among these DEG. As indicated in Table 2, the modificationdependent protein catabolic process, intracellular protein transport, coenzyme metabolic process, cellular response to stress, p.