DeepNOG: fast and accurate protein orthologous group assignment

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein orthologous group databases are crucial for evolutionary analysis and functional annotation.
• Alignment-based methods for orthology assignment present a significant computational bottleneck.

Purpose of the Study:

• To introduce DeepNOG, a novel, fast, and accurate alignment-free method for orthology assignment.
• To evaluate DeepNOG's performance against existing state-of-the-art methods.

Main Methods:

• DeepNOG utilizes deep convolutional networks for orthology assignment.
• The method was compared against HMMER, DIAMOND, and DeepFam on COG and eggNOG 5 databases.
• Performance was assessed based on precision, recall, and computational time.

Main Results:

• DeepNOG demonstrates significantly improved precision and recall compared to DeepFam on large datasets like eggNOG 5.
• DeepNOG achieves an order of magnitude faster computation time on CPUs compared to alignment-based methods.
• GPU acceleration further enhances DeepNOG's throughput.

Conclusions:

• DeepNOG provides a highly scalable and efficient solution for orthology assignment.
• The alignment-free approach offers a substantial speedup for large-scale genomic analyses.
• A user-friendly command-line tool facilitates the adoption of DeepNOG.