Fig. 2

Overview of the computational analysis to measure ∆LFE while controlling for other factors known to be under selection at different regions of the coding sequence and find factors correlated with it. a The variables and concepts involved in determining local folding strength and calculating ∆LFE. The effects of the compositional factors on the left side are removed in order to specifically measure the contribution of codon arrangements to the native folding energy. Blue arrows indicate possible selection forces. b The different steps in the computational pipeline used to estimate ∆LFE and the factors affecting it (see “Analysis” under the “Methods” section). For each genome, the CDSs are randomized based on each null model (CDS-wide and position specific), to calculate a mean ∆LFE profile based on that null model. At the next step, based on GLS, correlations between features of the ∆LFE profile and genomic/environmental features are computed. Input data sources (native CDS sequences, species trait values, species tree) are shown in green. c The distributions of some genomic properties within the dataset—CDS count, genomic GC-content, and genomic ENc′ (measure of CUB). The dataset was designed to represent a wide range of values (among other considerations; see “Species selection and sequence filtering” under the “Methods” section)