Automation of gene function prediction through modeling human curators' decisions in GO phylogenetic annotation project

TL;DR

This paper introduces an automated pipeline that models human curator decisions to streamline gene function annotation in the GO-PAINT project, reducing manual effort and increasing efficiency.

Contribution

It presents a novel automated pipeline that simulates curator decisions, enabling faster and scalable gene function annotation in phylogenetic analysis.

Findings

Automated pipeline successfully models curator decisions.

Over 4000 phylogenetic families annotated manually.

Pipeline available for public use and integration.

Abstract

The Gene Ontology Consortium launched the GO-PAINT project (Phylogenetic Annotation and INference Tool) 9 years ago and is currently being used in the GO Reference Genome Annotation Project to support inference of GO function terms (molecular function, cellular component and biological process) by homology. PAINT uses a phylogenetic model to infer gene function by homology, a process that requires manual curation of experienced biocurators. Tremendous amount of time and efforts have been spent on the GO-PAINT project yielding more than 4000 fully annotated phylogenetic families with more than 170,000 annotations. These preliminary data have thus enabled potential algorithmic representation and automatic solvation of the additional 9000 unannoated phylogenetic families. Here we present an automated pipeline for phylogenetic annotation and inference, which simulates the standard annotation procedures of curators and models the curators' decisions during the manual curation process. The pipeline has been built into the newest version of PAINT software available at http://www.pantherdb.org/downloads/index.jsp. The standalone automation pipeline and datasets are available at https://github.com/haimingt/GO-PAINT-automation