U133 plus 2.0 array

Gene expression profiles have been widely used with protein interaction networks to identify protein complexes, predict protein functions, construct dynamic protein interaction networks, and discover disease-related genes [24 (link)-26 (link)]. In this research, the human body index-transcriptional profiling of tissue-specific gene expression data set was downloaded from the gene expression omnibus (GEO) for GSE 7307 series [27 ] to predict disease genes. The dataset consisted of a total of 677 samples, representing over 90 distinct tissue types. Normal and diseased human tissues were profiled for gene expression using the Affymetrix U133 plus 2.0 arrays. Based on the case studies which has used in this study, detailed gene-expressions of 7 tissues were selected.

Clinical characteristics and pre-processed log2 median-centered microarray intensity values from the “Ju CKD Glom” [30 (link)] and “Ju CKD TubInt” [31 (link)] datasets were accessed from Nephroseq (v4; nephroseq.org, May 2020, University of Michigan, Ann Arbor, MI). Specifically, these datasets include microarray expression data generated from microdissected kidney glomerular and tubulointerstitial biopsy samples in the ERCB (S1 Dataset). The ERCB biopsies were obtained from patients after informed consent and with approval of the local ethics committees. Clinical and gene expression information from patients are accessible in a non-identifiable manner. From the open source dataset, subjects with DN, FSGS, HN, IgAN, lupus nephritis, MGN, MCD, thin basement membrane disease, tumor nephrectomy, and vasculitis were examined. Kidneys of HLD served as the control group. As per published reports, mRNA levels in the ERCB were derived using Affymetrix GeneChip Human Genome U133A 2.0 and U133 Plus 2.0 arrays [32 (link)].

MCbiclust was applied to the CCLE dataset (48 (link)) composed of 969 samples with gene expression levels measured as mRNA using Affymetrix U133 plus 2.0 arrays and updated probe set definition files from Brainarray (49 (link)). Before analysis completed by Barretina et al. (48 (link)) the dataset was background corrected using RMA (47 (link)) and quantile normalization methods, with quality assessment to identify low performing microarrays. To study mitochondrial related biclusters, MCbiclust was run 1000 times on the 1098 MitoCarta (50 (link)) genes known to be related to mitochondria. MCbiclust was additionally run 1000 times on random gene sets containing 1000 genes to find biclusters affecting general pathways.

Organoid RNA was isolated with the RNeasy mini kit after snap freezing organoids on dry ice. Samples were hybridized on Affymetrix U133 plus 2.0 arrays. Raw microarray data were normalized using the robust multi-array average (RMA) method^{76 (link)} followed by quantile normalization as implemented in the “affy”^{77 (link)} R/Bioconductor package. In order to exclude the presence of batch effects in the data, principal component analysis and hierarchical clustering were applied. Consensus molecular subtypes were determined as described previously^{78 (link)} using the single sample CMS classification algorithm with default parameters as implemented in the R package “CMSclassifier”. In all cases, differential gene expression analyses were performed using a moderated t test as implemented in the R/Bioconductor package “limma”^{79 (link)}. Gene set enrichment analyses were performed using ConsensusPathDB^{80 (link)} for discrete gene sets or GSEA as implemented in the “fgsea”^{81 (link),82} R/Bioconductor package for ranked gene lists. Wikipathways^{83 (link)} or Reactome^{84 (link)} were used for pathway analysis. Gene expression analysis was done in R version 4.0.0. When possible, packages were installed via bioconductor.

We reanalyzed a recently published rosacea mRNA expression dataset (7 (link)). This dataset included samples derived from 29 participants which were grouped as follows: ETR—7 patients, PPR—6 patients, PhR—6 patients, healthy volunteers—10 participants. Patients were diagnosed on the basis of the classification system of the American Rosacea Society (7 (link)). Total RNA was extracted from biopsies taken from the participants’ nasolabial fold and hybridized in duplicate to Affymetrix U133 plus 2.0 arrays (7 (link)). Raw data files (CEL files) were downloaded from the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/; accession GSE65914) (7 (link)) and then preprocessed using the RMA (Robust Multi-chip Average) pipeline (51 (link)) in combination with the most current reannotated probe set definitions (52 (link)). To determine differential mRNA expression, a linear model was fit using an empirical Bayes methodology for more robust variance estimates (53 (link),54 ). The false discovery rate (FDR) was computed as an adjusted P value (55 ) to account for multiple testing and a cut-off of 10% FDR, as well as an absolute fold change of 1.5 or greater was used to define differential expression. Clustering analysis was performed as previously described (56 (link)).

We used data from the Agilent G4502A_07 platform for TCGA, with measurements of 17,814 genes. Differentially expressed genes were selected based on the fold change of gene expression between each groups of tumor samples and the control (normal group) under the cutoff of |log⁡2foldchange| > 1 [17 (link)]. The overexpression/underexpression frequency was calculated for each gene in each tumor group. For example, gene A was overexpressed in ER group as compared to the normal group, and then the proportion of tumor samples in ER group with expression value of gene A higher than the mean expression value of gene A in normal group was defined as the overexpression frequency of gene A in ER group.
CCLE expression data was obtained using Affymetrix U133 Plus 2.0 Arrays, with measurements of 18,926 genes. Differentially expressed genes were selected based on the fold change of gene expression between each cell line and the average of expression value of all the cell lines [17 (link)].
For the comparison between gene expression data from TCGA and CCLE, robust z-scores (median-centered expression values divided by the median absolute deviation) were derived separately for the two data sets from CCLE and TCGA, and only common genes were remained.