The Spatial Atlas of Human Anatomy (SAHA): A Multimodal Subcellular-Resolution Reference Across Human Organs
A team of leading researchers released their new pre-print on the Spatial Atlas of Human Anatomy (SAHA). This work represents the most comprehensive subcellular-resolution reference of healthy adult human tissues to date. Integrating spatial transcriptomics, proteomics, and histology across over 15 million cells from more than 100 donors, SAHA captures both shared and organ-specific cellular niches in gastrointestinal and immune systems. It identifies rare immune populations, maps microenvironments, and reveals spatial disruptions in disease states such as colorectal cancer and IBD. The dataset includes high-plex RNA and protein measurements validated across multiple platforms, preserving spatial context often lost in dissociated single-cell studies. SAHA’s open-access portal supports computational modeling and clinical applications, serving as a benchmark for understanding tissue organization, pathology, and therapeutic response. This openly available resource is structured to support widespread use in research and diagnostics.
Publication Details: https://doi.org/10.1101/2025.06.16.658716
Improving spliced alignment by modeling splice sites with deep learning
A new preprint highlights the release of Minisplice. This lightweight tool is designed to improve spliced alignment by learning splice site patterns with a compact convolutional neural network. Trained across diverse vertebrate and insect genomes, the model distinguishes real GT..AG splice sites from background signals, boosting accuracy even in species without annotations. The tool integrates directly into minimap2 and miniprot, refining alignment decisions by scoring potential splice junctions. In benchmarks using long-read RNA-seq and cross-species protein alignments, minisplice consistently reduces junction error rates, especially for noisy or divergent sequences. Despite its small size (just over 7,000 parameters), minisplice performs competitively against more complex methods, offering a practical balance between performance and simplicity.
Publication details: https://doi.org/10.48550/arXiv.2506.12986
FastGA: Fast Genome Alignment
Researchers created FastGA, a genome alignment tool that achieves over tenfold speed improvements over prior methods while matching their sensitivity. It relies on cache-coherent most-significant-digit radix sorts and merges, a linear merge algorithm for adaptive seed detection, and a tailored adaptive-wave dynamic programming method that tolerates up to 30% sequence variation. It bypasses the need for repetitive-sequence masking and encodes millions of alignments in a compact trace-point format, further compressed by ONEcode. In tests, FastGA aligned two bat genomes (2.2 Gbp and 2.5 Gbp) in just over two minutes on an eight-core M4 Max laptop using under 6 GB of memory, yielding 1.05 million alignments covering about 60% of each genome in a 66 MB file. The tool is open source on GitHub (www.github.com/thegenemyers/FASTGA)
Publication details: https://doi.org/10.1101/2025.06.15.659750
A team of leading researchers released their new pre-print on the Spatial Atlas of Human Anatomy (SAHA). This work represents the most comprehensive subcellular-resolution reference of healthy adult human tissues to date. Integrating spatial transcriptomics, proteomics, and histology across over 15 million cells from more than 100 donors, SAHA captures both shared and organ-specific cellular niches in gastrointestinal and immune systems. It identifies rare immune populations, maps microenvironments, and reveals spatial disruptions in disease states such as colorectal cancer and IBD. The dataset includes high-plex RNA and protein measurements validated across multiple platforms, preserving spatial context often lost in dissociated single-cell studies. SAHA’s open-access portal supports computational modeling and clinical applications, serving as a benchmark for understanding tissue organization, pathology, and therapeutic response. This openly available resource is structured to support widespread use in research and diagnostics.
Publication Details: https://doi.org/10.1101/2025.06.16.658716
Improving spliced alignment by modeling splice sites with deep learning
A new preprint highlights the release of Minisplice. This lightweight tool is designed to improve spliced alignment by learning splice site patterns with a compact convolutional neural network. Trained across diverse vertebrate and insect genomes, the model distinguishes real GT..AG splice sites from background signals, boosting accuracy even in species without annotations. The tool integrates directly into minimap2 and miniprot, refining alignment decisions by scoring potential splice junctions. In benchmarks using long-read RNA-seq and cross-species protein alignments, minisplice consistently reduces junction error rates, especially for noisy or divergent sequences. Despite its small size (just over 7,000 parameters), minisplice performs competitively against more complex methods, offering a practical balance between performance and simplicity.
Publication details: https://doi.org/10.48550/arXiv.2506.12986
FastGA: Fast Genome Alignment
Researchers created FastGA, a genome alignment tool that achieves over tenfold speed improvements over prior methods while matching their sensitivity. It relies on cache-coherent most-significant-digit radix sorts and merges, a linear merge algorithm for adaptive seed detection, and a tailored adaptive-wave dynamic programming method that tolerates up to 30% sequence variation. It bypasses the need for repetitive-sequence masking and encodes millions of alignments in a compact trace-point format, further compressed by ONEcode. In tests, FastGA aligned two bat genomes (2.2 Gbp and 2.5 Gbp) in just over two minutes on an eight-core M4 Max laptop using under 6 GB of memory, yielding 1.05 million alignments covering about 60% of each genome in a 66 MB file. The tool is open source on GitHub (www.github.com/thegenemyers/FASTGA)
Publication details: https://doi.org/10.1101/2025.06.15.659750