Biocuration and machine learning platform for improved crop gene annotation
A combined biocuration and machine learning approach that improves functional gene annotations for large crop gene families, enabling discovery of novel agronomic genes and validation of gene expression, activity, and regulation. Targets limitations of current Gene Ontology pipelines that misannotate duplicate and clade-specific genes.
Overview
Accurate gene functional annotations are essential for connecting genotypes to phenotypes, interpreting transcriptomes, and advancing translational crop research. Current Gene Ontology (GO) pipelines rely heavily on gene-orthology-based projections, which often produce inaccurate annotations for large gene families. This offering combines expert manual biocuration with machine learning and text mining to produce high-quality functional annotations for large crop gene families, helping researchers discover native crop genes and validate gene or protein expression, activity, and regulation.
The approach addresses a critical gap: duplicate genes within expanded families are frequently assigned misleading functional labels. By integrating curated experimental evidence with computational analysis, the platform supports identification of clade-specific novel genes associated with agronomic traits such as stress tolerance, developmental processes, and yield.
Technical specifications
Core methodology:
- Manual biocuration of selected gene families to build a Gold Standard Gene Set, drawing on genomic data, transcriptome profiles, mutant phenotypes, trait associations, and clade-specific expansion patterns
- Machine learning models trained on curated data to automate extraction of functional information from scientific literature
- Text mining for large-scale literature analysis to capture gene-trait and gene-pathway relationships
- Transcriptome data analysis to identify evolutionarily conserved expression patterns and gene-gene interactors
- Differential expression analysis under developmental and environmental conditions to distinguish functional roles among duplicate genes
Validation context:
- Demonstrated on the SDRLK subfamily of Receptor-Like Kinases, where GO annotations incorrectly assign self-incompatibility associations to genes that are ubiquitously expressed and functionally unrelated
- Cross-species evidence from Arabidopsis and rice supports reclassification of misannotated family members
Technology readiness level
The methodology has been validated through published research demonstrating the limitations of existing GO annotations for the SDRLK gene family. The biocuration approach has been applied to characterize gene families in published studies. Future validation plans include curating five gene families comprising approximately 500 genes to create a Gold Standard Gene Set, followed by machine learning and text mining expansion to scale annotations across additional families. The platform is positioned for collaborative refinement and application to additional crop gene families.
About Oregon State University
Oregon State University is a comprehensive public research university and Oregon’s land‑grant institution, with a main campus in Corvallis and a statewide footprint. Industry partners tap a statewide Extension network and county offices to pilot and scale solutions with communities and companies across Oregon. A coastal marine science campus in Newport anchors ocean research and provides access to open‑ocean wave‑energy test ranges and grid‑connected infrastructure under development nearby, enabling sea‑to‑shore prototyping. Field stations and university‑managed research forests support long‑term trials and product validation in real‑world environments. A dedicated technology transfer office and the OSU Advantage programs—including the Advantage Accelerator—streamline IP, licensing, startup formation, and industry agreements.
delivered to your inbox.