Extended Data Fig. 2: Regional and global substitution models.
From: Identification of constrained sequence elements across 239 primate genomes
(a) Comparison of neutral background models with genome-wide random sampling of ancestral repeats from all autosomes (green) versus regional modeling of substitution rates at a 1 Mb scale (purple). The upper panel shows median phyloP scores in 1 Mb windows along chromosome 1, the lower panel the corresponding standard deviations. Median scores and dispersion are very similar between global and regional neutral models, values of larger discrepancy tend to fall within windows that containe a limited number of ancestral repeat sequences used to calibrate the regional model, resulting in less reliable estimates of local substitution rates (<50 kb, annotated as purple crosses). (b) Comparison of performance of global versus regional model at separating codon position 2 (amino acid-altering positions) versus 4-fold degenerate sites (synonymous positions), and promoters versus matched distal non-coding sequence. Global and regional models achieve similar performance on both coding and non-coding benchmarks.
