28 April 2014
The ever increasing number of sequences in protein databases usually turns out large numbers of homologs in sequence similarity searches. While information from homology can be very useful for functional prediction based on amino acid conservation, many of these homologs usually have high levels of identity among themselves, which hinders multiple sequence alignment computation and, especially, visualization. More generally, high redundancy reduces the usability of a protein set in machine learning applications and biases statistical analyses. We developed an algorithm to identify redundant sequence homologs that can be culled producing a streamlined FASTA file. As a difference from other automatic approaches that only aggregate sequences with high identity, our method clusters near full-length homologs allowing for lower sequence identity thresholds. Our method was fully tested and implemented in a web application called FASTA Herder, publicly available at http://fh.ogic.ca.