Highly accurate ab initio gene annotation with ANNEVO

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate gene annotation is crucial for understanding genome function.
• Existing methods face challenges in modeling complex gene transmission, including vertical inheritance and horizontal gene transfer.
• There is a need for advanced computational tools to improve ab initio gene prediction.

Purpose of the Study:

• To introduce ANNEVO, a novel genomic language model for precise ab initio gene annotation.
• To develop a model capable of directly learning from diverse genomes and evolutionary relationships.
• To overcome limitations of current gene annotation methods in handling complex evolutionary patterns.

Main Methods:

• ANNEVO utilizes a mixture of experts-based approach.
• The model directly learns distal sequence dependencies and joint evolutionary relationships from genomic data.
• It operates independently of external evidence, enabling ab initio annotation.

Main Results:

• ANNEVO demonstrates substantial performance improvements over existing ab initio gene annotation methods.
• Its performance is comparable to state-of-the-art annotation pipelines.
• ANNEVO provides more complete annotations than reference annotations and corrects existing errors across a wide range of species.

Conclusions:

• ANNEVO represents a significant advancement in ab initio gene annotation.
• The model's ability to integrate evolutionary insights enhances genome sequence interpretation.
• ANNEVO offers a powerful framework for precise and comprehensive gene prediction in diverse organisms.