ISOGO

ISOGO predicts Gene Ontology (GO) functions for protein-coding splice isoforms by integrating protein domain information and expression correlation data to enable isoform-level functional annotation.

Key Features:

• Isoform-level GO imputation: Predicts GO annotations for individual protein-coding splice isoforms rather than for genes as a whole.
• Protein domain integration: Uses protein domain information to inform function prediction at the isoform level.
• Expression correlation data: Leverages expression correlation data derived from RNA-seq across 11,373 cancer patients to support functional assignment.
• Improved precision-recall performance: Achieves an area under the precision-recall curve (AUPRC) five times larger than earlier methods.
• High AUROC for assigned functions: Reports a median area under the receiver operating characteristic curve (AUROC) of 0.82 for assigned functions to genes.
• Isoform-specific validation: Validated against known isoform-specific functions of BRCA1, MADD, VAMP7, and ITSN1.
• Benchmark evaluation: Evaluated using data from the CAFA3 challenge.
• Main isoform assessment with APPRIS: Assesses whether the main isoform predicted by APPRIS is most likely to possess annotated gene functions, finding this in 99.4% of genes.

Scientific Applications:

• Isoform-specific functional annotation: Assigns GO biological process, molecular function, and cellular component terms to individual coding isoforms.
• Alternative splicing interpretation: Enables analysis of functional diversification arising from alternative splicing at the isoform level.
• Cancer transcriptomics: Uses cancer patient RNA-seq-derived expression correlations to inform function prediction in cancer-associated contexts.
• Benchmarking and method comparison: Provides isoform-level predictions suitable for benchmarking in challenges such as CAFA3.
• Main isoform prioritization: Supports identification of the most functionally relevant isoform per gene using APPRIS-based assessment.

Methodology:

Integrates protein domain annotations with expression correlation data derived from RNA-seq of 11,373 cancer patients and uses APPRIS main-isoform predictions to impute Gene Ontology functions for coding isoforms.