Predicting functional residues of protein sequence alignments as a feature selection task

Determining which residues within a multiple alignment of protein sequences are most responsible for protein function is a difficult and important task in bioinformatics. Here, we show that this task is an application of the standard Feature Selection (FS) problem. We show the comparison of standard FS techniques with more specialised algorithms on a range of data sets backed by experimental evidence, and find that some standard algorithms perform as well as specialised ones. We also discuss how considering the discriminating power of combinations of residue positions, rather than the power of each position individually, has the potential to improve the performance of such algorithms.