SRA

The recent development of high throughput DNA provides an alternative new method for describing and quantifying transcriptome. During the analysis process, the user is nevertheless always confronted to a particular count data set (which each row corresponds to the number of reads assigned to one gene dispersed among the technical runs) and which statistical work consists to obtain reliable lists of differential genes (or other entities) between two conditions: Here we so applied 454 roche sequencing to trout larval trunk RNA samples and adult muscle samples. This fundamental stage of process leads us to interrogate about the repeatability of these lists such generated by using the sequencing approach. This paper proposes to give the most accurate answers to this question by comparing list of differential genes obtained from 454 sequencing technology to those one obtained by using more classical microarray technology as Agilent support [4*44K]. Basing on 7010 genes in common with unique correspondence between the two technologies (when it exists), the variability between replicates from one support to another have been studied corresponding to the first level of comparison; the divergences observed between the differential lists thus corresponding to the second level: If the descriptive analysis emphasizes the difficulty for contigs associated to low number of reads to be highly correlated (Spearman''s correlation <0.22 for genes mapping with less than 8 reads versus 0.80 otherwise), the differential analysis goes faster and underlines the problem of insufficient coverage for contigs generated by 454 technology (Here 1.069.535 reads for 55793 contigs). A flexible statistical tool based on the properties of hypergeometric distribution suggests that it is, in fact, not really possible to conclude in favour of differential effect for contigs as soon as they count strictly less than a total of 5 reads with a type I error rate of 5%. Isolated these specific genes from the analysis generates a sensitive improvement of the differential lists of genes in overlap from 7 to 20%. This tool shows finally that a differential analysis was in fact really possible for only 15% of the beginning genes from 454 sequencing. A most accurate study on the technical variability for RNA sequencing gives also evidence about the existence of a real between runs ''variability. Indeed about 70% of specific genes which are counted as differential by using the microarray support show at the same time a greater variability between runs than the other groups and plays an inhibitor roles for these genes to be re-find as differential with the RNA sequencing support. It plays so directly on the final interpretation and highlights the importance for this variability to be well taken into account by users in more adapted statistical models.