CRAFT

CRAFT generates compact alignment-free embeddings from k-mer co-occurrence statistics to enable efficient storage, retrieval, and fast comparison of large-scale genomic and metagenomic high-throughput sequencing (HTS) datasets.

Key Features:

• Compact representation: Transforms HTS and genomic data into a reduced embedding space by counting co-occurrences of adjacent k-mer pairs and learning compact embeddings.
• Dimensionality reduction and storage: Reduces sequence data dimensionality by 10^2–10^4-fold, decreasing memory and storage requirements for large datasets.
• Embedding-based sequence comparison: Performs fast sequence comparisons by calculating dissimilarities between learned embeddings of input and archived sequences.
• Benchmark performance: Has been benchmarked against six state-of-the-art alignment-free measures and shows comparable speed and accuracy.

Scientific Applications:

• Genome browsing: Locating best-matching genomes within massive sequence repositories using embedding-based similarity search.
• Metagenomics studies: Comparing metagenomic sequencing data against archived databases to analyze complex microbial communities.
• Comparative genomics: Rapidly comparing genomic sequences to identify similarities and differences across genomes.

Methodology:

Parse input genomic/metagenomic sequences to extract k-mers; count co-occurrences of adjacent k-mer pairs; learn compact embeddings from k-mer frequencies; compute dissimilarities between embeddings for sequence comparison; report best matches.