DeepGOZero

DeepGOZero predicts protein functions using ontology-based zero-shot learning to infer Gene Ontology (GO) classes for proteins from sequence data.

Key Features:

• Zero-Shot Learning Capability: Leverages zero-shot learning to predict GO classes with minimal or no experimental annotations using ontology-derived embeddings.
• Ontology Axioms Utilization: Integrates formal Gene Ontology axioms and logical relationships into the prediction model to enhance inference.
• Neural Network Integration: Combines ontology embeddings with neural networks to map protein sequence data to GO class predictions.
• Generic Zero-Shot Prediction Methodology: Employs a prediction approach that can be generalized to infer associations with other ontology classes beyond protein function.

Scientific Applications:

• Protein function prediction for unannotated GO terms: Predicts functions for GO classes that lack experimental annotations to extend functional annotations.
• Functional annotation of uncharacterized proteins: Generates hypotheses about molecular functions, biological processes, and cellular components for proteins from sequence data.
• Ontology-based association inference: Applies the zero-shot ontology-informed approach to infer associations between entities and ontology classes in other domains.

Methodology:

Uses a model-theoretic approach to learn ontology embeddings from Gene Ontology axioms and combines those embeddings with neural networks to perform zero-shot predictions from protein sequence data.