FANTASIA

FANTASIA annotates protein sequences with Gene Ontology (GO) terms by leveraging embedding-space similarity from protein language models to infer functions across proteomes.

Key Features:

• Protein language models: Uses protein language models, including ProtT5, to extract functional signals from sequence embeddings.
• Embedding-space similarity: Employs embedding-space similarity as the primary basis for transferring functional annotations between proteins.
• GOPredSim implementation: Implements GOPredSim with the ProtT5 model for GO term prediction.
• GO term prediction: Predicts Gene Ontology (GO) terms for protein sequences to provide functional annotations.
• Large-scale proteome annotation: Applies annotation across extensive proteomes, including analyses of approximately 1,000 animal proteomes.
• Uncharacterized genes in non-model organisms: Targets functional annotation of uncharacterized protein-coding genes prevalent in non-model organisms.
• Transcriptomics recovery: Recovers functional information from transcriptomics experiments using proteome-derived embeddings.
• Validation on curated datasets: Validation and benchmarking performed on curated datasets demonstrating reliable performance.
• Comparative performance: Reported to outperform traditional deep learning methods in accuracy and informativeness across species and gene ontologies.

Scientific Applications:

• Functional annotation of proteomes: Assigns GO terms to proteins across full proteomes to enable genome-scale functional maps.
• Annotating non-model organisms: Provides functional predictions for uncharacterized protein-coding genes in non-model species.
• Transcriptomics interpretation: Aids recovery and interpretation of functional signals from transcriptomics datasets.
• Molecular evolution and biological process inference: Supports analyses that relate predicted functions to molecular evolution and biological processes.
• Cross-species comparative analyses: Enables comparisons of functional annotations across species and gene ontologies.

Methodology:

FANTASIA predicts GO terms by computing embedding-space similarity using GOPredSim with the ProtT5 protein language model.