• GTEx，The Genotype-Tissue Expression (GTEx) project，首次被提出來是2013年，上百位科學家聯(lián)名在Nature Genetics雜志發(fā)表的文章首次介紹了“基因型-組織表達工程”，并成立了“基因型-組織表達研究聯(lián)盟”。The GTEx has catalogued gene expression in >9,000 samples across 53 tissues from 544 healthy individuals.
• TCGA，The cancer genome altas，https://cancergenome.nih.gov/ ，是由National Cancer Institute ( NCI, 美國國家癌癥研究所) 和  National Human Genome Research Institute (NHGRI, 國家人類基因組研究所) 合作建立的癌癥研究項目，通過收集整理癌癥相關的各種組學數(shù)據(jù)。The Cancer Genome Atlas (TCGA) has quantified gene expression levels in >12000 samples from >33 cancer types.

其實是沒辦法簡單的回答是否可以整合TCGA和GTEx數(shù)據(jù)庫，或者說該如何結合，這背后的統(tǒng)計學略微有點復雜，不僅僅是批次效應。發(fā)表在Sci Data. 2018; 的文章：Unifying cancer and normal RNA sequencing data from different sources 就比較詳細的說明了TCGA和GTEx數(shù)據(jù)庫的轉錄組數(shù)據(jù)的天然差異：

• sequencing platform and chemistry, personnel, details in the analysis pipeline, etc
• 基因表達量范圍：4-10 (log2 of normalized_count) for TCGA, and 0-4 (log2 of RPKM) for GTEx

全部代碼共享在：GitHub (https://github.com/mskcc/RNAseqDB).

最近一篇發(fā)表在SR，17 February 2020 的文章：Variability in estimated gene expression among commonly used RNA-seq pipelines 比較了常見轉錄組測序數(shù)據(jù)分析流程對定量拿到的表達矩陣的影響：

• We compared gene expression values from common samples (4,800 tumor samples from TCGA and 1,890 normal-tissue samples from GTEx) processed by the pipelines to     understand how gene expression quantification is impacted by differences in data processing.

TCGA和GTEX是兩個超級大的擁有RNA-seq數(shù)據(jù)的計劃，其中TCGA涵蓋33種癌癥，超1萬個樣品，而GTEX也有500多個病人的50多種組織的近1萬個樣品數(shù)據(jù)。它們各自的發(fā)起單位對RNA-seq數(shù)據(jù)處理不一樣，而且后續(xù)也有一些新的流程處理試圖統(tǒng)一兩個數(shù)據(jù)庫的RNA-seq數(shù)據(jù)分析結果，比較出名的5個流程分別是：

• TOPMed pipeline (https://github.com/broadinstitute/gtex-pipeline)
• recount2 pipeline (https://jhubiostatistics.shinyapps.io/recount/)

作者把這5個流程應用到TCGA和GTEX，得到10個不同組合的數(shù)據(jù)

• GDC (GDC-Xena/Toil, GDC-Piccolo, GDC-Recount2, GDC-MSKCC and GDC-MSKCC Batch).
• GTEx (GTEx-Xena/Toil, GTEx-Recount2, GTEx-MSKCC, GTEx-MSKCC Batch)

做了非常完善的比較，并且公布全部代碼在：https://github.com/sonali-bioc/UncertaintyRNA

整合TCGA和GTEx數(shù)據(jù)庫的文獻

非常多！

很多簡陋的數(shù)據(jù)挖掘，比如發(fā)表在PeerJ的 BIOINFORMATICS AND GENOMICS雜志的文章：Identification of four hub genes associated with adrenocortical carcinoma progression by WGCNA 也會涉及到TCGA數(shù)據(jù)庫和GTEx的整合。

首先下載TCGA和GTEx數(shù)據(jù)庫的TPM表達矩陣：

Gene transcripts per million (TPM) data were downloaded from the UCSC Xena database, which included ACC (The Cancer Genome Atlas, n = 77) and normal samples (Genotype Tissue Expression, n = 128).

然后差異分析流程是：

• Of the 60,498 genes in each sample, we removed genes with a mean TPM ≤ 2.5 (>1 is a common cutoff for determining if an isoform is expressed or not  in the cancer and normal samples and thus retained 13,987 genes.

• For those genes in the samples that showed significant changes, we used analysis of variance (ANOVA) in R  to determine the variance in genes between the two groups. ANOVA is a collection of statistical models useful for DEG analysis.

• We obtained 2,953 significant DEGs (Table S2) in ACC with a p < 0.001 and |log2 (fold-change)| > 1 cutoff.

差異分析結果是：1,181 up-regulated and 1,772 down-regulated genes.

可以看到，作者默認TPM這個轉錄組測序表達數(shù)據(jù)歸一化形式本身是具有跨平臺跨數(shù)據(jù)庫的特性，所以無需考慮批次效應，直接使用最簡單粗暴的ANOVA檢驗即可！

如果是甲基化數(shù)據(jù)

我們都知道，TCGA數(shù)據(jù)庫是目前最綜合最全面的癌癥病人相關組學數(shù)據(jù)庫，包括：

• DNA Sequencing
• miRNA Sequencing
• Protein Expression array
• mRNA Sequencing
• Total RNA Sequencing
• Array-based Expression
• DNA Methylation
• Copy Number array

知名的腫瘤研究機構都有著自己的TCGA數(shù)據(jù)庫探索工具，比如：

• Broad Institute     FireBrowseportal, The Broad Institute
• cBioPortalfor Cancer Genomics, Memorial Sloan-Kettering Cancer Center

對轉錄表達這個層面的信息來說，最優(yōu)選擇當然是整合TCGA和GTEx數(shù)據(jù)庫，但是對于甲基化數(shù)據(jù)，我們有沒有類似于GTEx數(shù)據(jù)庫的超級大隊列呢？

目前我還沒有接觸到，我前面分享過：這樣的診斷模型才優(yōu)秀，作者就是下載TCGA的結直腸癌甲基化位點信號矩陣文件：

• Tissue DNA methylation data were obtained from the TCGA (TCGA, TCGA-COAD, and TCGA-READ).

以及正常人的血液的甲基化信號值作為對照：

• Whole-blood DNA methylation profiles from healthy donors were generated in an aging study (GSE40279)

上面的兩個隊列是為了確定直腸癌特異性甲基化位點，做的是差異分析，確定了 top 1000 methylation markers

可以合理的推測應該是沒有人類各個正常組織的甲基化數(shù)據(jù)供使用，所以他們才會退而求其次使用正常人的血液的甲基化信號值作為對照吧！