The network generated in this research presented significant differences from the interaction networks previously reported for B. taurus. Differences were observed in the sources of information, the methods applied to construct the networks and their coverage, and the number of established interactions. For example, in 2011, Lim et al.  employed a literature mining tool to predict genes specifically associated with marbling in cattle and derived two networks primarily associated with the characteristic of interest based on the orthologous relationship between B. taurus and Homo sapiens (interologous method). The first network demonstrates high reliability and consists of 52 genes. Among these genes, 61 interactions were established. The second network is a widespread network composed of 1090 genes and 1517 interactions. After a topological analysis, 20 genes (with a node degree ≥ 25) were selected as candidate genes related to bovine marbling. Five of these genes were associated with bovine marbling when the expression profile of each gene was evaluated.
Similarly, Hulsegge et al.  prioritized candidate genes for reproductive characteristics in cattle based on PPIs reported for existing orthologous genes between B. taurus and H. sapiens in the STRING database. The genes were prioritized using the average of two calculated scores. The first score was based on the expression profiles of each gene. The second score was based on a literature search. An enrichment analysis was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID), and represented biological processes were observed. In this work, 59, 89, 53, 23 and 71 candidate genes were identified with associations with reproductive traits in the amygdala, dorsal hypothalamus, hippocampus, anterior pituitary and ventral hypothalamus, respectively.
Moreover, the coverage values established in BosNet (16,065 genes and 1,747,160 interactions, equivalent to 73 % coverage) were higher than the values estimated by Lim and Hulsegge (4.9 and 27 %, respectively). Thus, BosNet relies on the concept of functional interaction networks and the integration of a wide variety of heterogeneous biological data (orthology relationships with different organisms, interactions reported in various databases, correlations between expression levels, similarities between nucleotide sequences, and shared functional domains), whereas the above-mentioned networks were based on data extracted from only a few sources of information.
In BosNet, each integrated experiment, whether genetic or computational, added evidence for gene associations; thus, a greater number of genes and biological processes could be represented, which improved the coverage and precision of the network . This improvement is evident in the results plotted in the ROC curves, which assess the predictive power of each of the networks derived for B. taurus. The networks derived from a single source of information exhibit a low level of predictive power, low coverage and a reduced number of interactions relative to the networks generated through the integration of diverse biological data. The coverage (27 %) obtained by Hulssege et al.  is noteworthy because the coverage was greater than that achieved in previously reported networks and exhibited greater predictive power than STRING (AUC 0.51) in this study, which was similar to the performance obtained in the integrated network BosNet (AUC 0.64). These results were expected because the interactions in STRING were generated using an integrative method that is conceptually similar to the methodology applied in the present study . Another important point is that the predictive power (i.e., ROC curve) of the networks reported for B. taurus that indicates the ability of each of these networks to correctly identify genes involved in a particular characteristic have not been assessed.
The coverage and number of interactions established in BosNet are similar to the results of functional interaction networks reported for other organisms of major economic and scientific importance, such as Oryza sativa, Arabidopsis thaliana, Saccharomyces cerevisiae, Caenorhabditis elegans, Mus musculus and H. sapiens, whose coverages range from 50 to 95 % of the genes reported for each of the organisms, with the number of established interactions ranging from 100,000 to 1.7 million [11, 13–17].
Currently, the availability of different types of biological data, such as functional annotations for B. taurus genes, is limited compared with the information available for more thoroughly studied organisms, such as H. sapiens .
Recently, systems biology approaches have revealed that genes associated with the same or related phenotypes tend to participate in common functional modules (such as protein complexes and metabolic pathways). Moreover, the analysis of protein interaction networks and the neighborhood of a given protein within the network have been used to functionally characterize proteins (guilt-by-association approach).
The guilt-by-association strategy has been widely applied. For example, Lee et al., in 2008 , 2010  and 2011 , identified genes directly associated with different phenotypes in C. elegans, A. thaliana and O. sativa, respectively, through an analysis of functional interaction networks.
Due to high genetic variation in the genome, SNPs have become the most useful type of marker for gene mapping and association studies. In bovines, different strategies have been used to discover SNPs and assess SNP associations with ERTs. Lee et al.  reported a pipeline to analyze non-synonymous SNPs in B. taurus after screening the SNPs, which were reported as coding SNPs (cSNPs). They detected 15,353 candidate cSNPs and established a panel of 41 SNPs to evaluate associations with puberty age, facial eczema resistance and meat yield. Three SNPs were nominally associated with facial eczema resistance (P < 0.01).
Commercial arrays in genome-wide association studies (GWAS) have been widely used to understand the genetic basis of complex traits in B. taurus; however, the genetic variation underpinning these traits cannot be exclusively explained by this approach. High-throughput sequencing technology could serve as an alternative, but sequencing large numbers of individual genomes remains prohibitively expensive.
Here, we used BosNet to prioritize novel and reported genetic variation in six candidate genes based on SNPs and performed an association study for growth traits.
Because IGF1R is established in the bovine somatotropic axis, the IGF1R gene is one of the only BosNet-prioritized candidate genes that was previously associated with bovine growth traits. The IGF1R gene is the primary receptor for insulin-like growth factors (IGFs), which perform the metabolic signal transduction responsible for cell proliferation, bone growth and protein synthesis in the GH-IGF pathway.
The IGF1R/Taq I polymorphism in one of the introns of this gene, which was identified by Moody et al. , has been analyzed in several studies but has not been associated with growth traits. Researchers have concluded that this lack of association is caused by the absence of one of its alleles in B. taurus; its low frequency in B. indicus; and its location on chromosome 21, which is one of the least favorable chromosomes for finding loci associated with growth and carcass composition [21–23]. Here, we identified novel polymorphic markers in IGF1R both in Charolais and Brahman cattle. Of these markers, rs210778604 and rs208140993, located in the IGF1R coding regions, were significantly associated with BW/FS and WW, respectively. However, validation of these results with a higher number of animals is required.
The RXRA gene produces a protein that belongs to a family of transcription factors and plays an important role in fat storage and movement. In knockout mice, this transcription factor demonstrated resistance to obesity induced by chemicals that can be found in diets. Adipogenesis and lipolysis were also affected . This gene demonstrated high genetic variation in the studied populations. We confirmed at least 20 SNPs. SNP g106,0040,449 demonstrated a significant association with WW and YW in the Charolais population. BW is correlated with calving ease and survival, and WW is a reliable index of adult weight performance and productive efficiency . Therefore, confirmation of the association is important to include this marker as a tool for marker-assisted selection based on these traits.
Finally, EGFR, which is located on the cell surface, is a mediator of cellular proliferation and differentiation. The binding of its ligand activates a tyrosine kinase that phosphorylates various substrates, thus activating pathways promoting cell growth and DNA synthesis . Here, we found that animals with the AA genotype for the rs385131275 marker from the EGFR gene exhibited WWs that were 40 and 30 kg higher than those of animals with heterozygous (GA) and homozygous (GG) genotypes, respectively.
Insulin is a polypeptide hormone produced and secreted by the beta cells of the islets of Langerhans in the pancreas. Insulin improves the absorption of glucose in cells. Qui et al.  proposed insulin gene as a candidate gene for the genetic analysis of complex traits, such as growth rate, body composition and fat deposition, in chickens. They analyzed the associations of four polymorphisms located in non-coding regions with 13 different characteristics of growth and body composition. Their findings indicated that one of the polymorphisms and a combination of haplotypes were significantly associated with BW adjusted to 28 days.
Here, we confirm polymorphisms of novel and previously reported SNPs located in the bovine INS gene. However, no association with the analyzed growth traits was observed.
The participation of the remaining candidate genes (i.e., USF1 and TCF15) in bovine growth could be deduced based on the function established for each of the genes (no association results for this trait were identified in this study, and none have been identified in cattle to date). In mice, the TCF15 gene revealed that this transcription factor is an important regulator of a subset of myogenic cells of the dorsolateral dermomyotome associated with the formation of non-migratory hypaxial muscles (abdominal and intercostal) . Moreover, USF1 is a transcription factor that has been suggested to act as a negative regulator of cell proliferation because it competes for DNA binding sites with transcription factors, such as Myc, which is involved in transformation, cellular proliferation and apoptosis [29, 30].
From a panel of 79 SNPs, we determined that markers rs210778604 and rs208140993 (located in the IGF1R coding regions) were associated with BW/FS and WW, respectively (Table 2). In addition, markers rs385131275 and g.106,004,449 (located on the EGFR and RXRA genes, respectively) were significantly associated with WW and YW in Charolais cattle.
The number of nominally significant associations and the strength of these associations with growth traits were compared to the results obtained from studies that applied the GWAS approach to identify markers associated with growth traits . Thus, BosNet can be used as a prioritization tool to direct the search for novel SNPs that are potentially associated with ERTs.
Updating BosNet is a dynamic process that adds new genes and increases the robustness of each represented biological process. Thus, novel interactions appear that may change the prioritization weighting of each interaction net. Because of this effect, BosNet users must consider that after an update, genes prioritized with a previous version of BosNet may no longer receive prioritization, even if they are still part of the interaction. Here, we use data from the 2012 version of BosNet, as it was at that time that we initially prioritized all the candidate genes that were genotyped and associated with growth traits. According to our records, the prioritization weightings for these genes did not change significantly from those obtained using the BosNet version updated in December 2014; however, in the current version of BosNet (March 2015), none of the previously prioritized genes reached the confidence threshold. We are currently working to improve the network topology analysis. Meanwhile, BosNet users must consider the uniformity of the selected candidate genes and favor those genes that increase the number of strong interactions.