Function Prediction of Proteins from their Sequences with BAR 3.0

Abstract

Protein functional annotation requires time and effort, while sequencing technologies are fast and cheap. For this reason, the development of software tools aimed at predicting protein function from sequences can help in protein annotation.

In this paper, we describe how to use our recently implemented Bologna Annotation Resource (BAR) version 3.0, a tool based on over 30 million protein sequences for protein structural and functional annotation. In BAR 3.0, sequences are arranged in a similarity graph and then clustered together when they share at least 40% sequence identity over 90% of sequence alignment, for a total of 1,361,773 clusters.

Protein sequences with known function transfer their annotation to other sequences in the same cluster after statistical validation. Sequences with unknown function and new sequences entering in a cluster inherit its statistically validated annotations.

The method well compares to other techniques in the Critical Assessment of protein Function Annotation algorithms (CAFA). The CAFA experiment tests the performances of different predictors on a dataset that accumulates annotations over time. BAR predictions have been submitted to all the instances of CAFA through the years (BAR Plus in CAFA, BAR++ in CAFA2 and BAR 3.0 in CAFA3). The benchmarking indicates that in the field improvement is still possible and that our BAR scores among the top performing methods.

This work focuses on how the tool can transfer statistically significant features to poorly annotated or new sequences derived from transcrptomics or proteomics experiments.