Results Developing the network For building gene-gene relationship networks, we considered the genes (nodes) from the Cancer Gene Census [23] only, since our aim was to find those gene-gene relationships which could be potential cross-talks among cancer signaling pathways. In order to identify such gene-pairs, we applied thresholds on their absolute Pearson Correlation Coefficient (PCC) values. These thresholds were 0.545 and 0.54 for parental and resistant conditions, respectively, which we selected from the corresponding distributions of all-pair absolute PCC values with the purpose of considering approximately the top 20% gene-pairs as pairwise relationships only. Applying these thresholds to the relationship values, 27,865 and 26,865 pair-wise relationships were identified in parental and resistant data matrices, respectively. Bayesian analysis For the two gene-gene relationship networks YkR and YkP, Bayesian inference of the parameters of the p1-model for an undirected network was applied. We used WinBUGS for scripting this inference and our scripts were inspired by Adams et al. [30]. We used 6000 MCMC iterations for parameter estimation with the first 5000 as ‘burn-in’. All the parameters in the p1-model appeared to converge rapidly during the burn-in iterations (data not shown). With the above settings, we estimated the posterior probabilities of each edge (gene-gene relationship) Pr(Yij1=1) in the two networks YkR and YkP. For each edge, the proportion of the 1000 sampled networks containing the edge was considered as the posterior probability of that edge being present in the network. Next, for each edge we calculated the odds ratio of their posterior probabilities as defined above. The rationale behind this calculation was that the edges (gene-pairs) found with high probabilities in resistant conditions but lower probabilities in parental conditions are more likely to be due to acquired resistance in cell lines. Therefore, we chose only gene-pairs with high odds ratio (≥ 10.0) and high posterior probabilities (≥ 0.5) of occurring in resistant conditions. We found 11,515 such gene-pairs (Additional file 2: Table S1) among all 68,265 [=(370×369)÷2] possibilities. We then observed whether the above gene-pairs overlap with the list of potential cross-talks between EGFR/ErbB signaling and other signaling pathways. Here, we collected 24 signaling pathways from Reactome [32], 35 signaling pathways from KEGG [33,34], and 63 signaling pathways from WikiPathway [35] databases, and respectively identified 1,083 (841 distinct), 2,179 (1,050 distinct) and 3,084 (876 distinct) gene-pairs (Additional file 3: Table S2, Additionalfile 4: Table S3 and Additional file 5: Table S4) between EGFR/ErbB and other signaling pathways (seeMaterials and method). Of the 11,515 gene-pairs identified above, we found 104 (97 distinct), 188 (99 distinct) and 299 (96 distinct) gene-pairs overlap with the potential EGFR cross-talks identified using Reactome, KEGG and WikiPathway, respectively. Note the number of potential cross-talks and the number of distinct gene-pairs are different because the same gene-pair can form cross-talks between multiple pathway-pairs (pathways are overlapping). We consider these overlapping gene-pairs as putative drug-resistant cross-talks between EGFR/ErbB and other signaling pathways. In these 104, 188 and 299 cross-talks, we found candidate EGFR/ErbB cross-talks with 13, 26 and 51 other signaling pathways, respectively. Moreover, among all 104, 188 and 299 cross-talks from Reactome, KEGG and WikiPathway, respectively, we found 32 distinct gene-pairs in at least two of these sets. Primary findings and detailed descriptions of all these putative cross-talks from the analyses of all three pathway sources are listed in Table 1, and Additional file 6: Table S5, Additional file 7: Table S6 and Additional file 8: Table S7, respectively. The network views of all these cross-talk sets from the analyses of individual pathway sources are shown in Figure 2. Netwalker analyses We conducted further analyses using Netwalker, a network analysis suite for functional genomics [36]. In this analysis, we observed the changes in GE values for each gene in the identified list of potential cross-talks. This was to verify our expectation that, since lapatinib is an EGFR/ErbB inhibitor, both genes involved in drug-resistant cross-talks should be up-regulated in resistant conditions compared to parental conditions, which may imply that the activation of other compensatory signaling pathways in resistant conditions can play a role in acquired resistance to inhibitors by activating the targeted pathway(s) [1,17]. Therefore, for all 67 genes involved in the above sets of 104, 188 and 299 drug-resistant cross-talks from Reactome, KEGG and WikiPathway, respectively, we made a heatmap image of GE values from both conditions (parental and resistant) (Figure 3A). For both resistant and parental conditions, we first averaged the gene expression values from the three samples corresponding to the three treatment conditions. Then these averaged gene expression values were transformed into z-scores (zero mean, unit standard deviation) and each z-score was normalized with the maximum of the absolute values of the z-scores across that particular gene. We observed that in 28 of these 67 genes (involved in 168 cross-talks), gene expression in one or more resistant conditions (0, 0.1 μM and 1 μM of lapatinib) was up-regulated relative to all the parental conditions (0, 0.1 μM and 1 μM of lapatinib) (Figure 3B) which may signify the insensitivity of these genes to inhibitors under resistant conditions. Note for Figure 3B only those genes are depicted for which both genes in some identified cross-talk had average GE values at resistant conditions greater than the average GE values at parental conditions. Figure 2 Network view of (A) 104, (B) 188, and (C) 299 putative drug-resistant cross-talks between pathways using Reactome, KEGG, and WikiPathway pathway databases in Breast Cancer Cell-line: SKBR3 (GSE38376). Nodes are genes, and the edges are the cross-talks. Note, all the cross-talks here possess posterior probabilities of appearing in resistant network ≥ 0.5 and Odds Ratio ≥ 10.0, which means the posterior probabilities of that cross-talk for appearing in parental network is ≤ 0.05. Figure 3 Heatmap of genes in putative drug-resistant cross-talks in breast cancer cell-line: SKBR3 (GSE38376). Heatmap image of comparative gene expression changes of parental and resistant conditions in (A) all 67 genes in all 104, 188 and 299 putative drug-resistant cross-talks using signaling pathways from Reactome, KEGG and WikiPathway database, respectively, and (B) 28 selected genes based on their differential regulation. Here, for each gene, the expression value at each of the 6 conditions (3 parental conditions, and 3 resistant conditions) is the average value of 3 sample patients [17]. For each gene, these 6 expression values (each of them is the average of 3 samples) were transformed into z-scores (zero mean, unit standard deviation) and each z-score was normalized with the maximum absolute value of the z-scores across that particular gene. Note, (B) includes only those genes which belonged to gene-pairs for which the average of GE values at resistant conditions was greater than the average of GE values at parental conditions. For both (A) and (B), red and green bars indicate up-regulation and down-regulation, respectively. Table 1 Primary findings from the analyses using signaling pathways from Reactome, KEGG and WikiPathway in breast cancer cell-line: SKBR3 (GSE38376) Pathway # of signaling Pathway of All Distinct All putative Distinct # of other source pathways interest Cross-talks gene-pairs § drug-resistant gene-pairs ¶ signaling of interest cross-talks pathways REACTOME 23 EGFR 1,083 841 104 97 13 KEGG 35 ErbB 2,179 1,050 188 99 26 WikiPathway 63 ErbB 3,084 876 299 96 51 ¶Number of distinct gene-pairs involved in all EGFR/ErbB cross-talks with all other signaling pathways; §Number of distinct gene-pairs commonly involved in all EGFR/ErbB cross-talks and drug resistance. For these 28 selected genes (168 cross-talks), we observed the relative changes in GE values (parental vs resistant conditions) in their candidate signaling pathways. First we analyzed EGFR signaling pathway from Reactome and found that many of the constituent genes were up-regulated in one (or more) resistant conditions whereas in all of their corresponding parental conditions they were down-regulated (Additional file 1: Figure S1). These 168 selected cross-talks associated EGFR (or ErbB) signaling pathways with 6 other signaling pathways that were found in at least two different pathway analyses (i.e. Reactome and KEGG, or KEGG and WikiPathway, or Reactome and WikiPathway). In those 6 other signaling pathways, we also observed a similar phenomenon as above (Additional file 1: Figure S1). These 6 signaling pathways are Notch signaling (in Reactome, KEGG and WikiPathway), Wnt signaling (in Reactome, KEGG and WikiPathway), insulin receptor/IGF1R signaling (in Reactome and WikiPathway), GPCR signaling (in Reactome and WikiPathway), hedgehog (in KEGG and WikiPathway), and TGF- β receptor signaling (in Reactome and WikiPathway). Again, for many of the constituent genes of these 6 signaling pathways, expression was up-regulated in at least one of the resistant conditions whereas in all the corresponding parental conditions they were down-regulated. Primary findings regarding these 168 selected drug-resistant cross-talks are listed in Additional file 9: Table S8, and the top 50 of those 168 cross-talks (based on sorted Odds ratio) are shown in Table 2. Table 2 Description of top 50 (based on sorted Odds ratio) cross-talks among all 168 potential drug-resistant cross-talks between EGFR/ErbB signaling and other pathways from all the analyses using Reactome, KEGG and WikiPathway databases in GSE38376 gene i ::gene j EGFR/ErbB :: PrYijR=1 PrYijP=1 Odds ratio Avg GEiP : Avg GEiR : Signaling pathway j Avg GEjP Avg GEjR AKT2::MAML2 §,¶ Notch signaling 0.5 0.03 16.67 87.71::76.59 96.84::78.6 MDM2::APC §,$ Wnt signaling 0.5 0.03 16.67 76.33::82.43 77.9::86.76 KIT::CDC73 § Wnt signaling 0.5 0.03 16.67 82.14::104.01 82.68::110.88 MDM2::CDC73 § Wnt signaling 0.5 0.03 16.67 76.33::104.01 77.9::110.88 KIT::GNAQ § GPCR signaling 0.5 0.03 16.67 82.14::130 82.68::139.33 MDM2::GNAQ §,$ GPCR signaling 0.5 0.03 16.67 76.33::130 77.9::139.33 KIT::TSHR § GPCR signaling 0.5 0.03 16.67 82.14::71.32 82.68::71.66 MDM2::TSHR § GPCR signaling 0.5 0.03 16.67 76.33::71.32 77.9::71.66 AKT2::APC ¶ Wnt signaling 0.5 0.03 16.67 87.71::82.43 96.84::86.76 AKT2::APC ¶ Hippo signaling 0.5 0.03 16.67 87.71::82.43 96.84::86.76 AKT2::CDH1 ¶ Hippo signaling 0.5 0.03 16.67 87.71::74.2 96.84::79.8 AKT2::GNAQ ¶ Gnrh signaling 0.5 0.03 16.67 87.71::130 96.84::139.33 AKT2::GNAQ ¶ Calcium signaling 0.5 0.03 16.67 87.71::130 96.84::139.33 AKT2::MDM2 ¶ p53 signaling 0.5 0.03 16.67 87.71::76.33 96.84::77.9 MDM2::AKT2 $ Regulation of toll-like 0.5 0.03 16.67 76.33::87.71 77.9::96.84 receptor signaling MDM2::AKT2 $ insulin signaling 0.5 0.03 16.67 76.33::87.71 77.9::96.84 MDM2::AKT2 $ RANKL/RANK signaling 0.5 0.03 16.67 76.33::87.71 77.9::96.84 MDM2::AKT2 $ AMPK signaling 0.5 0.03 16.67 76.33::87.71 77.9::96.84 MDM2::AKT2 $ MAPK signaling 0.5 0.03 16.67 76.33::87.71 77.9::96.84 MDM2::AKT2 $ Tweak signaling 0.5 0.03 16.67 76.33::87.71 77.9::96.84 MDM2::AKT2 $ Toll-like 0.5 0.03 16.67 76.33::87.71 77.9::96.84 receptor signaling MDM2::APC $ BDNF signaling 0.5 0.03 16.67 76.33::82.43 77.9::86.76 MDM2::APC $ Wnt signaling Netpath 0.5 0.03 16.67 76.33::82.43 77.9::86.76 MDM2::APC $ Wnt signaling 0.5 0.03 16.67 76.33::82.43 77.9::86.76 and Pluripotency MDM2::COL1A1 $ Nanoparticle-mediated 0.5 0.03 16.67 76.33::91.44 77.9::102.54 activation of receptor signaling MDM2::COL1A1 $ Osteoblast signaling 0.5 0.03 16.67 76.33::91.44 77.9::102.54 MDM2::GNAQ $ TSH signaling 0.5 0.03 16.67 76.33::130 77.9::139.33 MDM2::GNAQ $ Serotonin Receptor 2 0.5 0.03 16.67 76.33::130 77.9::139.33 and STAT3 signaling MDM2::GNAQ $ Serotonin Receptor 2 0.5 0.03 16.67 76.33::130 77.9::139.33 and ELK-SRF/GATA4 signaling MDM2::ITK $ T-Cell Receptor and 0.5 0.03 16.67 76.33::89.86 77.9::93.27 Co-stimulatory signaling MDM2::ITK $ Tcr signaling 0.5 0.03 16.67 76.33::89.86 77.9::93.27 MDM2::KIT $ Kit receptor signaling 0.5 0.03 16.67 76.33::82.14 77.9::82.68 MDM2::PAX5 $ ID signaling 0.5 0.03 16.67 76.33::68.91 77.9::71.02 MDM2::TSHR $ TSH signaling 0.5 0.03 16.67 76.33::71.32 77.9::71.66 AKT2::TP53 § Notch signaling 0.5 0.04 12.5 87.71::128.73 96.84::155.09 KIT::APC § Wnt signaling 0.5 0.04 12.5 82.14::82.43 82.68::86.76 KIT::MAML2 § Notch signaling 0.5 0.04 12.5 82.14::76.59 82.68::78.6 KIT::STK11 § IGF1R signaling 0.5 0.04 12.5 82.14::71.97 82.68::74.95 KIT::STK11 § insulin receptor signaling 0.5 0.04 12.5 82.14::71.97 82.68::74.95 KIT::TP53 § Notch signaling 0.5 0.04 12.5 82.14::128.73 82.68::155.09 MDM2::MAML2 §,$ Notch signaling 0.5 0.04 12.5 76.33::76.59 77.9::78.6 MDM2::STK11 § IGF1R signaling 0.5 0.04 12.5 76.33::71.97 77.9::74.95 MDM2::STK11 § insulin receptor signaling 0.5 0.04 12.5 76.33::71.97 77.9::74.95 MDM2::TP53 § Notch signaling 0.5 0.04 12.5 76.33::128.73 77.9::155.09 AKT2::GNAS ¶ Gnrh signaling 0.5 0.04 12.5 87.71::5465.46 96.84::6212.43 AKT2::GNAS ¶ Calcium signaling 0.5 0.04 12.5 87.71::5465.46 96.84::6212.43 AKT2::NF2 ¶ Hippo signaling 0.5 0.04 12.5 87.71::85.75 96.84::87.36 AKT2::TP53 ¶ P53 signaling 0.5 0.04 12.5 87.71::128.73 96.84::155.09 AKT2::TP53 ¶ Wnt signaling 0.5 0.04 12.5 87.71::128.73 96.84::155.09 CBL::CDH1 ¶ RAP1 signaling 0.5 0.04 12.5 194.46::74.2 208.45::79.8 Cross-talks found using signaling pathways from §Reactome, ¶KEGG, and $xx−xxWikiPathway Databases; Pathway j is the pathway containing gene j; PrYijR=1 and PrYijP=1 are the posterior probabilities of gene i:gene j in Resistant and Parental networks, respectively; AvgGEiP is the average GE value of all Parental conditions (each of which is an average of 3 samples) for gene i, AvgGEiR is similar but with Resistant conditions, and others are likewise similar. Signaling cross-talk between EGFR/ErbB and other signaling pathways Cross-talk between EGFR/ErbB and Notch signaling We investigated literature evidence regarding the putative cross-talks between EGFR/ErbB signaling and other signaling pathways. We found AKT2:MAML2 (in Reactome and KEGG), AKT2:TP53 (in Reactome), AKT2:MYC (in Reactome), KIT:MAML2 (in Reactome), KIT:TP53 (in Reactome), MDM2:MAML2 (in Reactome and WikiPathway), MDM2:TP53 (in Reactome), and TP53:MAML2 (in WikiPathway) gene-pairs as putative cross-talks between EGFR/ErbB signaling and Notch signaling pathways. Up-regulation of the Notch signaling pathway inhibits apoptosis and thus contributes to breast carcinogenesis [37]. The Notch signaling pathway cross-talks with EGFR/ErbB signaling at the mediator level [1], e.g. when activated, Notch1 contributes to cell growth and survival via Akt-activation in melanoma [38]. The Notch1 co-activator complex binds to the HES1 promoter [39] which encodes a transcription repressor that represses the expression of PTEN, a PI3K/Akt pathway inhibitor [40] contributing to tyrosine kinase inhibitor (TKI) resistance. Furthermore, Notch1 stimulates MYC transcription [41] and this stimulation can lead to the down-regulation of MYC via the Akt-pathway [42,43]. This putative gene-pair, AKT2:MYC was also found in our results as a potential drug-resistant cross-talk between the EGFR/ErbB and TGF- β receptor signaling pathways. Again, in HER2/neu-mediated resistance to DNA-damaging agents, the Akt pathway becomes activated which eventually suppresses p53 functions via enhancing MDM2-mediated ubiquitination [44]. Protein-protein interaction between MDM2 and p53 is evident as contributing to various cancer related activities [45,46]. Cross-talk between EGFR/ErbB and Wnt signaling We found MDM2:APC (in Reactome and WikiPathway), KIT:CDC73 (in Reactome), MDM2:CDC73 (in Reactome), CBL:APC (in Reactome and KEGG), PDGFRA:APC (in Reactome), and CBL:CDC73 (in Reactome), AKT2:APC (in KEGG), AKT2:TP53 (in KEGG), and TP53:APC (in WikiPathway) as putative drug-resistant cross-talks between EGFR/ErbB and Wnt signaling pathways. Deregulation of the Wnt/ β-catenin signaling pathway plays a critical role in various cancers including breast, colorectal, pancreatic and colon cancer [47,48], and its association with drug-resistance has been studied by several research groups [47-50]. Recently, it has been reported that resistant cell lines exhibited increased Wnt signaling in both breast and colon cancer [49,50]. Loh et al. showed that genes in the Wnt signaling pathway, in both the β-catenin dependent (AXIN2, MYC, CSNK1A1) and the independent arms (ROR2, JUN), were up-regulated in cell lines resistant to tamoxifen compared to the parental MCF7 cell line [49]. Furthermore, ROR1, a constituent gene of Wnt signaling pathway, plays a sustainer role in EGFR-mediated prosurvival signaling in lung adenocarcinoma via signaling cross-talk and was therefore reported to be a potential therapeutic target [51]. APC and MDM2 in the MDM2:APC cross-talk are both tumor suppressors; they co-regulate DNA polymerase- β [52,53] which is reported to be hyper-activated in a cis-diamminedichloroplatinum(II) resistant P388 murine leukemia cell line [54]. Again, β-catenin whose stability is negatively regulated by APC [55], confers resistance to PI3K/Akt inhibitors in colon cancer [56]. Cross-talk between EGFR/ErbB and GPCR signaling Between EGFR/ErbB and GPCR signaling pathways, we found KIT:GNAQ (in Reactome), MDM2:GNAQ (in Reactome and WikiPathway), CBL:GNAQ (in Reactome), FGFR2:GNAQ (in Reactome), PDGFRA:GNAQ (in Reactome), KIT:TSHR (in Reactome), MDM2:TSHR (in Reactome), CBL:TSHR (in Reactome), PDGFRA:TSHR (in Reactome), KIT:GNAS (in Reactome), MDM2:GNAS (in Reactome and WikiPathway), KIT:SMO (in Reactome), MDM2:SMO (in Reactome), TP53:GNAQ (in WikiPathway), and MYC:GNAQ (in WikiPathway). GPCR-like signaling contributes to acquired drug resistance after being mediated by Smoothened (SMO) via activating Gli, a canonical hedgehog (Hh) transcription factor [57]. GPCR and EGFR/ErbB over-expression often contributes to cancer growth. Cross-talk between the two at the receptor level contributes to HNSCC (head and neck squamous cell carcinoma) via triggering EGFR/ErbB signaling by a GPCR ligand [58]. For the MDM2:SMO cross-talk, found between the EGFR/ErbB and GPCR signaling pathways, a SMO-mutant from Hh signal transducer activates PI3K/Akt/Gli pathway that eventually increases MDM2 phosphorylation [59]. This in turn increases MDM2-mediated p53 degradation and thus reduces p53-induced apoptosis [59]. Furthermore, recently it has been reported that SMO (Hh signal transducer) functions like a G-protein coupled receptor due to its structural resemblance to GPCRs [60,61] which may be further evidence for a drug-resistant cross-talk between hedgehog signaling and EGFR/ErbB signaling [1]. Cross-talk between EGFR/ErbB and IR (insulin receptor)/IGF1R signaling Several studies have reported extensive cross-talk between IR (insulin receptor)/IGF1R (insulin-like growth factor-1 receptor) and EGFR/ErbB signaling pathways contributing to acquired drug resistance in various cancers [62-64]. Loduvini et al. reported significant correlation between worse disease-free survival and high co-expression of both EGFR/ErbB and IGF1R in NSCLC (non-small-cell lung cancer) patients [65]. EGFR/ErbB can physically interact with other non-ErbB family receptors at the cell surface and can form heterodimers with receptors like IGF1R, PDGFR etc. [62]. Moreover, the EGFR/ErbB and IGF1R pathways can also cross-talk indirectly via physical interactions between their downstream shared-components [62]. It has been reported recently that gefitinib (an EGFR TKI) inhibits the phosphorylation of IRS1 by IR, but also triggers the association between IRS1 and IGF1R which in turn induces drug-resistance [66]. Knowlden et al. showed the cross-talk between IGF1R and EGFR signaling pathways occurred in tamoxifen-resistant MCF7 and T47D breast cancer cell-lines but not in non-resistant cells [18]. Our findings suggest KIT:STK11 (in Reactome), MDM2:STK11 (in Reactome), MDM2:AKT2 (in WikiPathway), MYC:AKT2 (in WikiPathway), TP53:AKT2 (in WikiPathway), MDM2:CBL (in WikiPathway), MDM2:SOCS1 (in WikiPathway), and TP53:SOCS1 (in WikiPathway) as putative drug-resistant cross-talks between the IGF1R/IR and EGFR/ErbB signaling pathways. For the MDM2 and STK11 (also known as LKB1) genes, which we identified as a putative cross-talk between EGFR and IGF1R signaling, we did not find any direct supporting evidence in the literature. However, this association is plausible in the resistant conditions given that Yamaguchi et al. suggested EGFR signaling may cross-talk with the AMPK/LKB signaling pathway [1]. Moreover, Levine et al. reported interconnections between p53 and IGF1R/AKT/mTOR pathways where both LKB1 and MDM2 participate in a series of pathway cross-talks [67]. Validation of the framework using BT474 cell-line (GSE16179) To further illustrate our method, we analysed a second dataset (GSE16179) containing gene expression profiles of breast cancer cell-line BT474 under two conditions (parental and lapatinib resistant) [16]. The reason for choosing this dataset was that it was obtained using a similar experimental design to the primary dataset GSE38376, but with an additional treatment condition using foretinib (GSK1363089) only and with combined drug use (lapatinib + foretinib). There were three samples per treatment condition. However, to adapt simply and be coherent with the previous experiment, we only considered expression values of parental conditions (3 samples with basal condition: GSM799168, GSM799169 and GSM799170; 3 samples with 1 μM of lapatinib treatment: GSM79917, GSM799172 and GSM799173), and the same conditions with lapatinib resistant cells (3 samples with basal condition: GSM799174, GSM799175 and GSM799176; 3 samples with 1 μM of lapatinib treatment: GSM799180, GSM799181 and GSM799182). Among the 375 cancer genes from Cancer Gene Census [23], there were 357 genes which had gene expression values. We identified 27,358 and 26,292 pair-wise gene-gene relationships (undirected edges) in resistant and parental networks by applying the thresholds 0.71 and 0.81, respectively. Bayesian inference of the p1-model parameters for an undirected network was applied to these two gene-gene relationship networks as before. Thereafter, among all 63,546 [= (357 ×356) ÷2] possibilities, we found 10,811 gene-pairs (Additional file 10: Table S9) with the same thresholds of odds ratio (≥10.0) as previously, but smaller posterior probability (≥0.15) of occurring in the resistant network. With this set of putative drug-resistant gene-pairs, we also observed the overlap of potential cross-talks of EGFR/ErbB with other signaling pathways using Reactome, KEGG and WikiPathway databases. We found 83 (72 distinct), 133 (87 distinct) and 277 (81 distinct) cross-talks between EGFR/ErbB and other signaling pathways from Reactome, KEGG and WikiPathway (Additional file 11: Table S10, Additional file 12: Table S11 and Additional file 13: Table S12), respectively. The numbers of signaling pathways that were involved in those EGFR/ErbB cross-talks were 10, 18 and 54, respectively. Among the 83, 133 and 277 cross-talks, we found 50 distinct gene-pairs in at least two of these sets. Table 3 shows the comparative findings between our primary dataset (SKBR3 cell-line, GSE38379) and our secondary dataset (BT474 cell-line, GSE16179). In Table 3, we show that some important signaling pathways that were involved in the EGFR/ErbB cross-talks (i.e. Notch, WNT, GPCR, IR/IGF1R, TGF- β signaling pathways) in our primary dataset, have some overlap with our secondary dataset. Table 3 Comparative results between primary dataset (SKBR3 cell-line, GSE38376) and validation dataset (BT474 cell-line, GSE16179) Pathway name Found in Pathway source Found in Pathway source Common cross-talks in both Studies ¶ (GSE38376) (GSE16179) Notch Signaling Reactome, Reactome, MAP2K4::NOTCH1 KEGG, KEGG, WikiPathway WikiPathway GPCR signaling Reactome, Reactome, CBL::TSHR WikiPathway WikiPathway FGFR1::TSHR PDGFRA::GNAQ KIT::TSHR LCK::TSHR MDM2::TSHR PDGFRA::TSHR WNT Signaling Reactome, Reactome, AKT2::CCND2 KEGG, KEGG, MAP2K4::CCND2 WikiPathway WikiPathway MAP2K4::TP53 MDM2::MAP2K4 Insulin (IGF1R) Signaling Reactome, Reactome, MDM2::MAP2K4 WikiPathway WikiPathway TP53::MAP2K4 TGF- β Signaling Reactome, Reactome, MDM2::TFE3 WikiPathway KEGG, TP53::TFE3 WikiPathway MAPK signaling KEGG, KEGG, MDM2::MAP2K4 WikiPathway WikiPathway ¶These common cross-talks were found using the primary dataset (104, 188 and 299 cross-talks from Reactome, KEGG and WikiPathway databases, respectively) and validation datasets (83, 133 and 277 cross-talks from Reactome, KEGG and WikiPathway databases, respectively). Cross-talks mentioned with Bold face are those consistent with our hypothesis that both genes in the particular cross-talk are up-regulated in resistant conditions but down-regulated in parental conditions. There were 78 genes involved in these sets of 83, 133 and 277 putative cross-talks. We performed a similar Netwalker analyses with these 78 genes as we did for the dataset GSE38376, and found 37 genes (involved in 86 cross-talks (Additional file 14: Table S13)) consistent with our hypothesis that both genes in a particular cross-talk should be up-regulated in resistant conditions but down-regulated in parental conditions. In Figure 4, the selected genes from the secondary dataset exhibit an even clearer pattern of up-regulation in resistant conditions than the selected genes from our primary dataset. Figure 4 Heatmap of genes in putative drug-resistant cross-talks in breast cancer cell-line: BT474 (GSE16179). Heatmap image of comparative gene expression changes of parental and resistant conditions in (A) all 78 genes in all 83, 133 and 277 putative drug-resistant cross-talks using signaling pathways from Reactome, KEGG and WikiPathway database, respectively, and (B) 37 selected genes based on their differential regulation. Here, for each gene, the expression value at each of the 4 conditions (2 parental conditions, and 2 resistant conditions) is the average value of 3 sample patients [16]. For each gene, these 4 expression values (each of them is the average of 3 samples) were transformed into z-scores (zero mean, unit standard deviation) and each z-score was normalized with the maximum absolute value of the z-scores across that particular gene. For both (A) and (B), red and green bars indicate up-regulation and down-regulation, respectively.