ProPhylER Dataflow and Analysis Pipeline I

Overview

ProPhylER's data were generated by a dataflow that consists of three main parts:

1. Generation of clusters of orthologs and closely related paralogs, curated
2. Construction of alignments and trees, curated
3. Computation of rates, constraints, and predictions, automated

Cluster generation is based on Smith-Waterman alignments of pairs of sequences and is tuned to identify likely orthologs and close paralogs from whole-genome proteomes, to the exclusion of distant paralogs. Two manual curation steps ensured that clusters contain sufficient sequences to be informative, but that divergent paralogs would not be in the same cluster. After genome protein clusters are built, they are augmented by UniProt sequences, which contribute a substantial additional amount of information because of their annotations and because many UniProt sequences come from species whose genomes have not been sequenced. Another manual curation step at this point ensured integrity and specificity of the clusters.

From this point on, we require robust alignments of orthologs and close paralogs that

• have no highly divergent sequences that poison the alignment
• are unlikely to contain paralogs for which a change in substrate or binding specificity since its last common ancestor constraints would have resulted in a shift in constraints

This was enforced by manual curation of the alignments in which sequences were removed that violated ProPhylER's high standards for alignment specificity.

For each alignment, trees are then built automatically. After treebuilding, curator input may also be required. ProPhylER's post-treebuilding algorithms are then applied to

• Annotate the trees for gene duplications, which establishes orthology and paralogy
• Calculate evolutionary rates and infer evolutionarily constrained regions ("ECRs") across the alignment
• Generate consensus physicochemical profiles across the alignment
• Generate impact scores for all possible missense mutations for each sequence in the alignment

A number of large clusters have undergone additional curation in which subclusters were generated that can be viewed in the same ProPhylER session, allowing comparisons between different phylogenetic scopes or paralogs.

Detailed Dataflow Descriptions

1. From Individual Protein Sequences to Clusters of Closely Related Homologs

2. From Clusters of Close Homologs to Alignments and Trees

3. From Alignments and Trees to Profiles, Constrained Regions, and Mutation Impact Scores

Last updated 8/25/08