Density Peak clustering of protein sequences associated to a Pfam clan reveals clear similarities and interesting differences with respect to manual family annotation

Abstract Background The identification of protein families is outstanding practical importance for in silico annotation and at the basis several bioinformatic resources. Pfam possibly most well known family database, built many years work by domain experts with extensive use manual curation. This approach generally very accurate, but it quite time consuming may suffer from a bias generated hand-curation itself, which often guided available experimental evidence. Results We introduce procedure that aims to identify automatically putative families. based on Density Peak Clustering uses as input only local pairwise alignments between sequences. In experiment we present here, ran algorithm about 4000 full-length proteins least one classified belonging Pseudouridine synthase Archaeosine transglycosylase (PUA) clan. obtained 71 automatically-generated sequence clusters 100 members. While our were largely consistent classification, showing good overlap either single or multi-domain architectures, also observed some inconsistencies. latter inspected using structural evidence, suggested automatic classification captured evolutionary signals reflecting non-trivial features architectures. Based this analysis identified novel pre-PUA alternative boundaries few PUA PUA-associated As first indication was unlikely be clan-specific, performed same P53 clan, obtaining comparable results. Conclusions clustering described takes advantage information contained large set successfully identifies architectures an unsupervised manner. Comparison highlights significant points interesting differences, suggesting new could have potential applications related classification. Testing hypothesis, however, will require further experiments diverse datasets.