This WebApp is dedicated to diverse estimations of Co-Expression from GTEx data as a tool to understand global, organism-level co-expression and regulation.
The original estimations were performed by Miguel-Angel Cortes-Guzman & Victor Trevino (CoGTEx v1).
Then we extended the estimations and the website done mainly by Maria-Julia Teja-Urrutia.
Estimations:
- CoGTEx v1 integrate Coexpression by TPM levels (default) and Z-Scores (z-scores per tissue in user interface).
- Correlations per tissue : Pearson estimations per tissue.
- Number of correlations per tissue: Estimation of "global" coexpression by a simple sum of correlated tissues passing a threshold.
- Integrative Multi-Tissue: Chi-Square scoring of correlations irrespective of correlation sign.
- Mean per tissue: Pearson estimations from mean expression values per tissue-cluster.
- Mutual Exclusivity: Exploratory of perhaps "contrary" to correlated expression.

Gene expression per tissue-cluster. Methods for tissue cluster generation: ... GTEx v8 expression data was normalized, batch-processed, and filtered. Then, PCA, clustering, and tSNE stringent procedures were applied to generate 42 distinct and curated tissue clusters. Coexpression was estimated from these 42 tissue clusters. Read more

Coexpression using TPM or Z-Scores data from all tissue-cluster pulled together. TPM assumes mean expression level matters, while z-scores doesn't. ... The correlation was estimated from 33,445 genes, sampling 70 samples per tissue-cluster to avoid tissue over-representation. This process was repeated 20 times, extracting the minimum correlation as a robust estimation. Three correlation metrics were calculated (Pearson, Spearman, and G-statistic). G-statistic refer to a test very similar to Chi-Square contingency tables. Cut-off for contingency were estimated at 33% and 66% from its expression distribution. For TPM, data were used in Log10 scale and quantile-normalized pulling all tissues together then correlation was computed. For Z-Normalized, data were scaled to Z-Scores per tissue, then pulled together and correlation was estimated. Please refer to (Cortez-Guzmán 2023) for definition of clusters. Read more

Pearson correlation was calculated for the expression data of each of the 42 tissues-clusters using 70 samples per tissue-cluster and quantile-normalized TPM values in log10 scale (given that the estimation is per-tissue, a z-score would provide equivalent results). ... The table is built by selecting the desired tissues in the order they are clicked. The top 1000 correlated genes for each tissue are added. More than 1000 (or selected) genes are expected when selecting more than 1 tissue. By default, the table is sorted by the first selected tissue. In the scatter plot visualization, Pearson correlations can be stratified by sex and age. Read more

Estimation of the number of tissue-clusters whose correlation is higher than a cut-off. ... Estimated using quantile-normalized TPM. As above, 42 tissue-clusters were considered using 70 samples per tissue-cluster and quantile-normalized TPM values in log10 scale. The estimations are shown for +, -, or abs (||) values. The columns m0.3, m0.5, m0.7 show the lower of the + and - columns, which represent the number of "divergent" correlated tissues. Read more

Mutual Exclusivity score is estimated by a chi2-like statistic ... Using quantile-normalized TPM in log10 scale, the gene per person is defined as "expressed" or "not-expressed" whether it is higher or lower than 0.2 (or ~1.6 TPM). A contigency table is generated among gene pairs. The mutual exclusive score = min(A, B) * (1-f) / 100. A is the (observed-expected)^2/expected (as in the contigency table) of the expressed-not expressed coordinate. B is the equivalent for the not-expressed - expressed coordinate. A and B are set to zero if expected is larger than observed. f is the fraction of samples in expressed - expressed or not expressed-not expressed coordinates to compensate for those genes not expressed in some tissues or those expressed in some tissues. Read more

Many genes pairs show strong positive and negative correlations, which affects estimations of coexpression when pulled together. ... Therefore, Integrative coexpression uses the correlation per tissue-cluster to estimate an overall coexpression measure. For this, a p-value is estimated from a chi-square statistic. X2_all = sum( |Zr|^2 ), df = t-1. X2_expressed = sum( |Zr|^2 conditioned to expressed ), df = t_expressed - 1. t is the number of tissues. t_expressed is the number of tissues where the paired gene is expressed. X2_all consider all tissues irrespective of the expression level (Integrative All column). X2_expressed consider only expressed tissues in the paired gene (Integrative Expressed column). Zr is the Pearson correlation coefficient standardized for that tissue (Zr = (r-mean) / stddev ). p-values can be estimated from X2_all and X2_expressed and the minimum value is chosen. Then p-values are ranked. Read more

The genes within +/- 2 Mbp are included here.

Pearson "robust" correlation from the mean of each gene per tissue. The robust estimation is performed by removing the greatest and the lowest tissue-means per gene to avoid exceptional overexpression or underexpression. To visualize an approximation of this in the scatter plot, use option to show the "mean +/-" and remove all other options. You will see the means (and confidence intervals) per tissue-cluster.

In this space, you can revise specific genes manually. Click Open-Genes and cut-paste list of genes. Open Genes Close Genes

(1) Configuration: Top number of genes: 2050100200500    Filter for TF genes
(2) Select the order of genes from sections above:    
        Correlations:   CoGTEX v1 (TPM/Z-Score)      Per Tissue      Number of Tissues      Integrative      Means 
        Other:   Mutual Exclusivity      Neighbouring Genes      Manual