Modeling and cleaning RNA-seq data significantly improve detection of differentially expressed genes

BMC Bioinformatics

Table 3 Results of the cleaning of expression data from BGISEQ-500 platform [25]

Cleaning method	#DEGs* (p-value^#)	DEGs functional classification
Cleaning method	#DEGs* (p-value^#)	Cellular process	Metabolic process	Response to stimulus	Regulation of biological process
Author’s^$	1045	466	448	297	219
Raw data	1047 (6.77e-17)	501	458	314	211
RNAdeNoise	1215 (3.86e-18)	559	502	347	235
HTSFilter	860 (8.37e-17)	416	372	278	179
counts > 3	995 (5.58e-17)	477	434	301	201
counts > 5	988 (3.52e-17)	477	432	302	199
counts > 10	940 (4.97e-17)	450	409	293	183
FPKM > 0.3	790 (6.77e-17)	384	348	258	165

DEGs were detected by EdgeR and classified using DAVID. Filtering with RNAdeNoise increases the significance and adds 16% more DEGs and up to 20% DEGs in several functional classes compared to the original results
*Criteria for DEGs: |log2(FoldChange)|>1.0, p-value < 0.002
^#Average p-value of top 100 genes
^$First four functional classes from Fig. 3 [25] were taken for comparison

ISSN: 1471-2105