IsofunGO

IsofunGO differentiates isoform functions using Gene Ontology (GO) embedding and an attention-based multi-instance learning framework to provide isoform-level functional annotations of proteoforms resulting from alternative splicing.

Key Features:

• Gene Ontology (GO) embedding: Models the hierarchical structure of GO terms as an attributed hierarchical network and embeds GO terms into compact low-dimensional vectors that preserve semantic relationships and hierarchy while reducing computational complexity during prediction.
• Attention-based multi-instance learning (MIL): Employs an attention-based MIL network that integrates genomics and transcriptomics data to predict functions of individual isoforms by referencing the compressed GO annotations from the embedding process.
• Compressed GO annotations: Uses the GO embeddings to produce compressed annotations that serve as targets for isoform-level prediction.
• Enhanced interpretability and performance: Combines GO embedding and attention mechanisms to improve prediction accuracy and to highlight relevant isoform features driving functional assignments.

Scientific Applications:

• Isoform-level functional annotation: Differentiates functions of alternative isoforms to enable analysis of proteoform-specific roles in biological processes.
• Functional genomics of complex disease: Supports dissection of molecular mechanisms underlying complex diseases by providing higher-resolution functional annotations at the isoform level.
• Benchmark evaluation: Enables experimental validation and performance assessment on benchmark datasets for isoform-function prediction.

Methodology:

Constructs an attributed hierarchical network to embed GO terms into low-dimensional vectors and applies an attention-based multi-instance learning network that integrates genomics and transcriptomics data to predict isoform functions using the compressed GO annotations.