DeepNOG

DeepNOG assigns protein sequences to orthologous groups using deep convolutional neural networks and the eggNOG 5 database to enable rapid, alignment-free orthology assignment for protein function prediction.

Key Features:

• Deep convolutional neural networks: Uses deep convolutional neural networks to perform alignment-free sequence classification for orthologous group assignment.
• eggNOG 5 integration: Employs the eggNOG 5 database as the reference for orthologous group labels.
• Alignment-free processing: Bypasses sequence alignment to increase computational efficiency compared to alignment-based workflows.
• Comparable accuracy to profile HMMs: Achieves accuracy levels comparable to profile hidden Markov model approaches such as HMMER.
• Computational efficiency: Reduces computation time by an order of magnitude on CPUs and offers optional GPU acceleration.
• Benchmarking against alignment-free methods: Demonstrated higher precision and recall than the alignment-free method DeepFam on COG and eggNOG 5 benchmarks.
• Comparison to fast aligners: Provides greater computational efficiency relative to alignment-based tools such as HMMER and DIAMOND.

Scientific Applications:

• Evolutionary analysis: Enables large-scale orthology assignments to inform comparative and evolutionary studies across diverse lineages.
• Functional annotation: Supports protein function prediction by mapping sequences to annotated orthologous groups.
• Metabolic pathway modeling: Facilitates assignment of proteins to orthologous groups used in reconstructing metabolic pathways.
• Large-scale genomic analyses: Suited for high-throughput processing of proteomes where rapid orthology assignment is required.

Methodology:

Performs alignment-free orthology assignment using deep convolutional neural networks trained with eggNOG 5 references, evaluated against alignment-based methods (profile HMMs/HMMER and DIAMOND) and the alignment-free method DeepFam on COG and eggNOG 5 benchmarks, and supports CPU execution with optional GPU acceleration.