HTG EdgeSeq assay provides a toolkit doing differential expression by DESeq2. I compared results coming by this software and when I am doing DESeq2 in R version 3.5.1 manually like
dds=DESeqDataSetFromMatrix(countData = df, colData = mycols, design = ~ condition) dds <- DESeq(dds)
Doing DESeq2 manually gives 200 significant genes but software gives no significant genes. In description of assay says;
This differential expression analysis has been completed using the DESeq2 package (version 1.14.1) available from Bioconductor. The DESeq2 package provides methods for estimating and testing differential expression using negative-binomial generalized linear models. Empirical Bayes methods are used to estimate dispersion and log2(fold change) with data-driven prior distributions. See http://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html for more information. No pre-filtering is applied to the data prior to analysis. The DESeq2 model corrects for library size using the median ratio method from Anders and Huber (2010). Dispersions are estimated with the Cox Reid-adjusted profile likelihood method developed by McCarthy et al. (2012). Log2 fold change is estimated via Tikhonov/ridge regularization with a zero-centered normal prior distribution with variance calculated using the observed distribution of maximum likelihood coefficients (see DESeq2 documentation for details). DESeq2 performs independent filtering on probes prior to applying the false discovery rate p-value adjustment in order to increase power. This will cause some probes to have no p-value.
I am just wondering different results of software from HTG company and my manually analysis is because company doing different setting in DESeq2 or why?
Both me and IT staff in HTG company comparing raw read counts of same cancer samples Vs normal samples but results is 100% different
This is few lines of results that HTG IT sent to me
Differential Expresssion Outputs: Mean Normalized (group): The estimated mean value of each probe for each group after normalization. AveExpr: The log2-transformed average expression of each probe accross all groups after normalization. Fold Change (group1.vs.group2): The estimated fold change between the two groups (transformed from log-fold change). rawP (group1.vs.group2): The unadjusted p-value for the comparison of a single probe accross two groups. adjP (group1.vs.group2): The p-value for each probe after adjustement using the Benjamini and Hochberg (1995) method for controlling the false discovery rate. Probe Mean normalized 1 Mean normalized 2 AveExpr Fold Change 2.vs.1 rawP 2.vs.1 adjP 2.vs.1 A2M 5446.19 4264.01 12.20 -1.23 0.13 0.62 AADAT 338.63 314.73 8.34 -1.06 0.71 0.91 ABCB1 396.92 334.81 8.49 -1.16 0.19 0.69 ABCB11 138.67 168.31 7.29 1.18 0.18 0.69