HyperGen: Compact and Efficient Genome Sketching using Hyperdimensional Vectors.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate whole-genome similarity estimation, like Average Nucleotide Identity (ANI), is computationally intensive for large datasets.
• Genome sketching offers a faster, memory-efficient alternative by distilling representative k-mers.
• Existing methods face challenges in balancing accuracy, speed, and memory usage for massive genome collections.

Purpose of the Study:

• Introduce HyperGen, a novel genome sketching method.
• Enhance accuracy, runtime performance, and memory efficiency for large-scale ANI estimation.
• Leverage hyperdimensional computing for improved genomic data representation.

Main Methods:

• Encode genomes into quasi-orthogonal vectors (Hypervectors, HV) using hyperdimensional computing (HDC).
• Utilize compact HV sketches that preserve more information than traditional k-mer hashes.
• Employ vector multiplication and optimized general matrix multiply (GEMM) routines for efficient ANI estimation.

Main Results:

• HyperGen achieves comparable or superior ANI estimation accuracy and linearity versus existing sketch-based methods.
• Demonstrates state-of-the-art performance in both genome sketching and database search speed.
• Generates memory-efficient sketches while maintaining high ANI estimation accuracy across various dataset scales.

Conclusions:

• HyperGen offers a significant advancement in scalable genomic similarity analysis.
• The HDC-based approach provides a robust and efficient solution for massive genome collections.
• HyperGen is available as an open-source Rust implementation.