katalyx/minimap2fast-indexed-reference-genome-homo-sapiens-GRCh38

GRCh38 Indexed Reference Genome for mm2-fast

Dataset Description

This dataset contains the pre-computed learned index for the GRCh38 (hg38) human reference genome optimized for use with mm2-fast, an accelerated implementation of minimap2 that achieves up to 1.8x speedup on modern CPUs. The learned index uses a Recursive Model Index (RMI) based hash table for optimized seeding, significantly reducing alignment time compared to standard minimap2.[^1][^2][^3]

Dataset Contents

This dataset includes the following files for each preset configuration:

For ONT reads (map-ont preset):

• GRCh38.p14.genome.fa - Reference genome in FASTA format
• Learned index files (RMI data structures) for optimized seeding
• Configuration metadata for the map-ont preset

Additional presets may include:

• map-hifi - PacBio HiFi reads
• map-pb - PacBio CLR reads
• asm5 - Assembly-to-assembly alignment (≤5% divergence)
• asm20 - Assembly-to-assembly alignment (≤20% divergence)

The learned index was generated using mm2-fast's create_index_rmi.sh script, which takes approximately 2-3 minutes for the human genome.[^2]

Usage

Prerequisites

Install mm2-fast with learned hash table support:

git clone --recursive https://github.com/bwa-mem2/mm2-fast.git
cd mm2-fast
source build_rmi.sh  # Build binaries for creating/using learned index
make clean && make lhash=1  # Compile with learned hash table optimization

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("your-username/grch38-mm2fast-index")

Using with mm2-fast

Place the reference FASTA and learned index files in your working directory, then run mm2-fast directly:

# Align ONT reads
./minimap2 -ax map-ont GRCh38.p14.genome.fa reads.fastq.gz > alignments.sam

# Align PacBio HiFi reads (use map-hifi indexed files)
./minimap2 -ax map-hifi GRCh38.p14.genome.fa reads.fastq.gz > alignments.sam

# Align PacBio CLR reads (use map-pb indexed files)
./minimap2 -ax map-pb GRCh38.p14.genome.fa reads.fastq.gz > alignments.sam

Important: You must use the mm2-fast binary compiled with lhash=1 and have the corresponding learned index files in the same directory as the reference FASTA.[^2]

Integration with Bioinformatics Pipelines

This indexed reference can be integrated into Nextflow or Snakemake workflows:

# Nextflow example
process MM2FAST_ALIGN {
    input:
    path reference_fa
    path learned_index_files
    path reads
    
    output:
    path "*.sam"
    
    script:
    """
    ./minimap2 -ax map-ont ${reference_fa} ${reads} > aligned.sam
    """
}

Technical Specifications

• Reference Version: GRCh38.p14 (Genome Reference Consortium Human Build 38, patch 14)
• Assembly Type: Primary assembly (chromosomes 1-22, X, Y, MT)
• Index Format: RMI-based learned hash table for mm2-fast
• mm2-fast Version: Based on minimap2 v2.24
• Index Generation Time: ~2-3 minutes for human genome
• Genome Size: ~3.1 billion base pairs
• Memory Requirement: ~30GB for human genome alignment
• CPU Requirements: x86_64 with AVX512 or AVX2 support[^2]

Performance Benefits

mm2-fast with learned index provides significant performance improvements over standard minimap2:[^3]

• Learned index enables 3-4x faster minimizer lookups during seeding
• Combined with SIMD-parallel chaining and optimized alignment, achieves up to 1.8x end-to-end speedup
• Maintains 100% identical output to minimap2 (with --max-chain-skip=infinity)

For example, mapping 100K reads from HG002 ONT data against the human genome takes 54 seconds with mm2-fast compared to 92 seconds with standard minimap2 on Intel Xeon Platinum 8280 CPUs.[^2]

Use Cases

This indexed reference is suitable for:

• Oxford Nanopore Technologies (ONT) long-read sequencing alignment
• PacBio long-read sequencing (HiFi and CLR) alignment
• Whole genome sequencing analysis with accelerated throughput
• High-volume genomics projects requiring faster turnaround
• Cancer genomics and transcriptomics workflows
• Structural variant detection from long reads
• Research workflows on modern multi-core CPUs with AVX512/AVX2

Index Generation

The learned index was created using mm2-fast's indexing workflow:

source build_rmi.sh
./create_index_rmi.sh GRCh38.p14.genome.fa map-ont

This generates the RMI-based data structures that are loaded automatically when mm2-fast runs with the lhash=1 compilation flag.[^2]

Citation

If you use this dataset in your research, please cite:

GRCh38 Reference:

• Schneider VA, et al. (2017). "Evaluation of GRCh38 and de novo haploid genome assemblies demonstrates the enduring quality of the reference assembly." Genome Research, 27(5), 849-864.

mm2-fast:

• Kalikar S, Jain C, Md V, Misra S. (2022). "Accelerating minimap2 for long-read sequencing applications on modern CPUs." Nature Computational Science, 2, 78-83. https://doi.org/10.1038/s43588-022-00201-8

Minimap2:

• Li H. (2018). "Minimap2: pairwise alignment for nucleotide sequences." Bioinformatics, 34(18), 3094-3100.

Important Notes

• The learned index files must be present in the same directory as the reference FASTA file during alignment
• mm2-fast must be compiled with make lhash=1 to enable learned hash table optimization
• Different presets (map-ont, map-hifi, map-pb, etc.) require separate learned indexes
• For correctness verification, mm2-fast should be run with --max-chain-skip=1000000 to match standard minimap2 output
• mm2-fast requires modern CPUs with AVX512 or AVX2 SIMD instruction support

License

The GRCh38 reference genome is available under the public domain. Please refer to the Genome Reference Consortium for specific usage terms.[^4]

Support

For issues or questions regarding this dataset, please open an issue on the dataset repository. For mm2-fast specific questions, refer to the mm2-fast GitHub repository.