ECRECer

ECRECer predicts enzyme commission (EC) numbers from protein sequences to associate proteins with the biochemical reactions they catalyze.

Key Features:

• Hierarchical Dual-Core Multitask Learning Framework (HDMLF): A dual-core framework that implements hierarchical, multi-objective multitasking for EC prediction.
• Embedding core (protein language models): Transforms protein sequences into numerical embeddings using protein language models.
• Learning core (GRU-based EC prediction): Predicts EC numbers using a gated recurrent unit (GRU) architecture.
• Attention layer: Integrates an attention layer within HDMLF to optimize prediction weighting.
• Greedy integration strategy: Employs a greedy strategy to integrate and fine-tune model components.
• Performance improvements: Demonstrates comparative gains reported as ~60% improvement in accuracy and ~40% increase in F1 score versus four representative methods, with enhanced performance on recently discovered proteins.

Scientific Applications:

• Enzyme promiscuity discovery: Enables identification of potential promiscuous activities, exemplified by predicting tyrB compensating for loss of aspartate aminotransferase aspC.
• Enzymology research: Facilitates assignment of EC numbers to support studies of enzyme function.
• Metabolic engineering: Supports metabolic engineering efforts by linking protein sequences to biochemical reactions via EC annotation.

Methodology:

Implements the HDMLF composed of an embedding core that uses protein language models to generate sequence embeddings and a learning core that uses a GRU for EC prediction; includes an attention layer and employs a greedy strategy for model integration and fine-tuning within a multi-objective hierarchical multitasking framework.