An automated pipeline for multi-species protein function prediction from the UniProt Knowledgebase

A recent international challenge, CAFA (Critical Assessment of Functional Annotation of proteins), provided the unique opportunity of critically evaluate and compare state-of-the-art methods for automated protein function prediction (AFP) (1). As pointed out by the CAFA results, one of the key issues that characterizes the AFP problem is the integration of both heterogeneous experimental data and different prediction methods. These items pose serious computational problems because of the ever increasing rate at which bio-molecular data are made available in public databases and for the complexity of the the prediction tasks. In this contribution we describe an automatic pipeline that addresses both these issues by exploiting data directly available for the UniProt KB. We applied the pipeline to the second and still ongoing CAFA 2 challenge.