Figure 1 : Comparison of ribosome profiling normalization approaches.

From: Comparative survey of the relative impact of mRNA features on local ribosome profiling read density

Figure 1

(a) A stylized footprint density profile for MTIF3 gene transcript from ‘Andreev’ data set (left) is transformed into a binary function with RUST (centre). Each sequence feature, such as AAA codon in the case shown, could be characterized by its frequency as 1 or 0 (right). (b) The distributions of normalized codon densities for all AAA codons in ‘Andreev’ data set using different approaches, conventional normalization CN (left), ribosome residence time, RRT (top right) and logarithmic mean normalization, LMN (right). Note that due to intrinsic differences the scale of possible normalized densities (axis x) varies among the methods and that due to the selection criteria of each approach the number of datapoints used (axis y) is also variable. (c) Performance of five normalization approaches (RUST, CN of transcripts with average gene density >1/nucleotide (CN>1) and CN of all expressed transcripts (CN>0), LMN and RRT) at estimating codon dwell times The box plots show the distribution of log values of the estimated/simulated dwell times for all 61 codons. The deviations of these values from 0 occur due to under or overestimation of simulated dwell times. The better methods are those that have distributions with a smaller variance. Each subpanel represents a specific scenario. The simulation scenarios differ by coverage that reduces from top to the bottom and the level of noise modelled as high peaks of density that increases from left to right, except for the right-most column where noise is modelled as missing data at 20% of the coordinates. Asterisks used to indicate insufficient data for CN>1.