Differential gene expression analysis

Question

[zhoub]

Hello,every one.I have been analyzing differentially expressed genes between tumors and normals(not paired) with RNAseq data and confused about the methods. Applying a fold-change cut-off is commonly used while a threshold for expression Z scores is also used. I would like to know which method is better for unpaired RNAseq data. Thanks so much!

[VHahaut]

Hi!

Which software did you use so far? (for mapping, counting and then calling differentially expressed genes)?

[zhoub]

I downloaded TCGA level 3 RNAseqV2 RSEM data for lung cancer,removed genes with low expression and analyzed by edgeR. Then I searched cBioPortal for further validation and found that expression Z-scores were used for expression analysis in the database, which gave the different result.

[BioinfGuru]

Usually the cut off is +/- 2 log2(rpkm). I'm not familiar with using Z scores.

Answer 1

[Persistent LABS]

It is important to use both statistical significance and effect size while filtering the deferentially expressed genes. Statistical significance level (for example: a p-value or better adjusted p-value) will tell you whether the observed difference is due to the data in hand or it appeared just by chance, while the effect size (for example: fold-change) will tell you the magnitude of difference between the two groups. You can have a look at this published article (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3444174/) which talks about "Using Effect Size—or Why the P Value Is Not Enough".

[zhoub]

Thanks. I did use fold-change and adjusted p-value for analysis by edgeR. However, some of my results were different from those in cBioPortal database using expression Z-scores. For example,in TCGA lung adenocarcinoma dataset, CAV1 was downregulated in tumors with a logFC -3.2 and p-value E-104, but CAV1 was identified to overexpress in 6% tumors using Z-score absolute value > 2. I was wondering about what leads the opposite results and which method should be used for differential analysis.

[Persistent LABS]

This paragraph from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4156192/ might help you to understand your findings. "A positive correlation exists between the upregulation of Cav-1 and the clinical features of primary lung cancer. Although Cav-1 levels in lung tumor tissues are significantly lower than in tumor-free lung tissues (91,92,100–102), the expression of Cav-1 in lung tumor tissues is markedly higher in patients with lymph node metastasis (92,93,100) and advanced tumor stage (93,100,103)." You have considered only 3 samples for your analysis. I am not sure about the tumor stages of these samples.