I would like to measure the conservation of a particular subset of human splice sites and compare them to all other splice sites and matched sites that have a canonical splice-site sequence, but are not spliced. Which scores should I use? UCSC has both phasCons and PhyloP scores for hg38 from alignments of 100 vertebrates, and there is also GERP (although unfortunately UCSC doesn't have GERP for hg38). I'm not sure I understand which of these would be best.
The following text is from UCSC's page on PhastCons / PhyloP (LINK). I think for the size of splice-sites that the fine grained PhyloP score might be most useful.
"PhastCons is a hidden Markov model-based method that estimates the probability that each nucleotide belongs to a conserved element, based on the multiple alignment. It considers not just each individual alignment column, but also its flanking columns. By contrast, phyloP separately measures conservation at individual columns, ignoring the effects of their neighbors. As a consequence, the phyloP plots have a less smooth appearance than the phastCons plots, with more "texture" at individual sites. The two methods have different strengths and weaknesses. PhastCons is sensitive to "runs" of conserved sites, and is therefore effective for picking out conserved elements. PhyloP, on the other hand, is more appropriate for evaluating signatures of selection at particular nucleotides or classes of nucleotides (e.g., third codon positions, or first positions of miRNA target sites)."
@Emily Ensembl has point out that Ensembl has hg38 GERP scores here: ftp://ftp.ensembl.org/pub/current_compara/conservation_scores/88_mammals.gerp_conservation_score/