Fig. 2

Evaluation of FFPE library preparations and data normalization approaches. a) Two FFPE samples (S1-orange, S2-green) were each prepared using two different commercial library preparations (NuGEN-Ovation-blue, Illumina-Access-black) in triplicate and sequenced by a Illumina Hi-Seq 2500 sequencer. Disparate library methodologies (Ovation (Ova-red and purple)), Access (Acc-orange and green)) resulted in clear differences in b) average raw counts per gene. Results of both libraries were subjected separately to c) Loess and VST, and d) quantile and q-spline data normalization approaches. VST normalization was selected based upon equal performance in both libraries across the range of reads and demonstrated success in e) normalization of average counts per gene in both libraries. f) Unsupervised hierarchical clustering of (14,432) normalized gene counts from S1 and S2 samples in triplicate reveals highest level of sample separation based upon library preparation (blue and black), followed by sample source. The 500 genes with the highest overall mean expression are shown. g) Library-based variation was assessed by plotting the coefficient of variation (CV) from triplicate samples for each gene for each library preparation (S1-orange, S2-green). Average % CV and standard deviation of CV is presented along corresponding axes. 20% CV is indicated by red dashed lines.