Biology subjects

Salzberg, S. L.

Publications and source records attributed to Salzberg, S. L..

At least 19 recordsLinked to original sources

bioRxiv2026

Improving Metagenomics Classification with Kmask: Entropy-Based Masking of Low-Complexity Regions

Abstract Accurate taxonomic classification in metagenomics is often compromised by low-complexity sequences, which lead to chance matches that in turn cause sequences to be misclassified. Here we present Kmask, an entropy-based masking tool implemented for use either standalone or as part of Kraken [1,2] database construction, which replaces low-entropy regions with Ns. Using a sliding window size aligned with Kraken's default k-mer length and parameters optimized across 12 control bacterial genomes spanning a broad range of GC content, Kmask efficiently removes low-complexity sequences while retaining high-complexity regions. To benchmark performance, we applied Kmask to a newly constructed database, Microbial2025, that contains over 71,000 bacterial, archaeal, viral, and fungal genomes, and we then classified human reads against both masked and unmasked versions of the database using KrakenUniq [2]. We found that Kmask substantially reduced misclassifications, driving down the false positive rate to 5.78% from 7.52%. Notably, Kmask performed comparably to an SDUST-masked [3] database, achieving a similar false positive rate (5.78% vs. 5.17%) while masking out fewer bases (1.33% vs. 1.85%). We also tested Kmask on a database of human cancer sequences, where we found that it eliminated many false positives caused by low-complexity matches between bacterial genomes and human DNA. These results demonstrate that Kmask is an effective method for masking low-complexity sequences in large microbial databases, thus improving the accuracy of metagenomic classification.

bioinformatics↗

bioRxiv2026

Large-scale analysis of transcript data reveals thousands of recursive splicing events in human introns

Recursive splicing (RS) is a process in which an intron is removed from a nascent RNA molecule in two or more splicing events rather than one. We introduce a novel approach for detecting recursive splice sites (RSSs), the intronic loci at which RS events occur, based on alignment of total RNA-seq data to short, customized "target" sequences. We applied this approach to a data set from a recent study of gene expression in the human brain, using parameters corresponding to a very low false discovery rate, and found 3,022 RSSs that appear in 2,775 distinct introns from 2,407 genes. 2,891 (96%) of these RSSs are in protein-coding genes. The median length of recursively spliced introns from this set is 10,114 base pairs, which is substantially longer than the median human intron, but much shorter than average RS intron lengths reported in prior studies. Our work dramatically increases the number of known RSSs in the human genome and provides a generalizable bioinformatics pipeline for annotating RSSs from total RNA-seq data.

genetics↗

bioRxiv2025

Comprehensive Transcriptome Annotation of Thousands of HIV-1 Genomes

Alternative splicing in HIV-1 has been a central focus of decades of research, uncovering key mechanisms of viral gene regulation, immune evasion, and therapeutic response - yet, no reference resource has existed to support transcriptome-wide analysis, limiting adoption of modern computational methods. We present HIV Atlas (https://ccb.jhu.edu/HIV_Atlas), the first reference-quality annotation of HIV-1 and SIV transcriptional diversity. We manually curated transcriptomes for HIV-1HXB2 and SIVmac239 and developed Vira, an automated annotation-transfer method specifically designed to address unique challenges of viral genome biology, to generate high-quality annotations for 2,077 complete HIV-1 genomes. Using the resources presented in our work, we evaluated conservation of splice sites, revealing near-perfect preservation of major donors and acceptors. Furthermore, using several public datasets, we demonstrate how HIV Atlas enhances methodology, improves the quality and novelty of results, and opens novel avenues for research, supporting more accurate and comprehensive analyses of bulk, single-cell, and spatial RNA-seq in HIV-1 studies.

genomics↗

bioRxiv2025

Evidence of off-target probe binding in the 10x Genomics Xenium v1 Human Breast Gene Expression Panel compromises accuracy of spatial transcriptomic profiling

The accuracy of spatial gene expression profiles generated by probe-based in situ spatially-resolved transcriptomic technologies depends on the specificity with which probes bind to their intended target gene. Off-target binding, defined as a probe binding to something other than the target gene, can distort a genes true expression profile, making probe specificity essential for reliable transcriptomics. Here, we investigated off-target binding affecting the 10x Genomics Xenium technology. We developed a software tool, Off-target Probe Tracker (OPT), to identify putative off-target binding via alignment of probe sequences and assessing whether mapped loci corresponded to the intended target gene across multiple reference annotations. Applying OPT to a Xenium human breast gene panel, we identified at least 14 out of the 313 genes in the panel potentially impacted by off-target binding to protein-coding genes. To substantiate our predictions, we leveraged a Xenium breast cancer dataset generated using this gene panel and compared results to orthogonal spatial and single-cell transcriptomic profiles from Visium CytAssist and 3' single-cell RNA-seq derived from the same tumor block. Our findings indicate that for some genes, the expression patterns detected by Xenium demonstrably reflect the aggregate expression of the target and predicted off-target genes based on Visium and single-cell RNA-seq rather than the target gene alone. We further applied OPT to identify potential off-target binding in custom gene panels and integrate tissue-specific RNA-seq data to assess effects. Overall, this work enhances the biological interpretability of spatial transcriptomics data and improves reproducibility in spatial transcriptomics research.

bioinformatics↗

bioRxiv2025

OpenSpliceAI: An efficient, modular implementation of SpliceAI enabling easy retraining on non-human species

The SpliceAI deep learning system is currently one of the most accurate methods for identifying splicing signals directly from DNA sequences. However, its utility is limited by its reliance on older software frameworks and human-centric training data. Here we introduce OpenSpliceAI, a trainable, open-source version of SpliceAI implemented in PyTorch to address these challenges. OpenSpliceAI supports both training from scratch and transfer learning, enabling seamless retraining on species-specific datasets and mitigating human-centric biases. Our experiments show that it achieves faster processing speeds and lower memory usage than the original SpliceAI code, allowing large-scale analyses of extensive genomic regions on a single GPU. Additionally, OpenSpliceAIs flexible architecture makes for easier integration with established machine learning ecosystems, simplifying the development of custom splicing models for different species and applications. We demonstrate that OpenSpliceAIs output is highly concordant with SpliceAI. In silico mutagenesis (ISM) analyses confirm that both models rely on similar sequence features, and calibration experiments demonstrate similar score probability estimates.

genomics↗

bioRxiv2024

WheresWalker: a pipeline for rapid mutation mapping using whole genome sequencing

Forward genetic screening is a powerful approach to assign functions to genes and can be used to elucidate the many genes whose functions remain unknown. Chemical mutagenesis is an unbiased and efficient method for generating point mutations in the founding generation of animals in a forward genetic screening experiment. Missense and nonsense mutations induced by chemical mutagenesis can lead to the generation of partial function, gain-of-function, or null alleles that underlie compelling phenotypes, but positional cloning of the underlying causative single base pair changes can be laborious and time-consuming, especially in large polymorphic genomes. Current methods use a bioinformatic mapping-by-sequencing approach which often identifies large genomic regions which contain an intractable number of candidate genes for testing. Here, we describe WheresWalker, a modern mapping-by-sequencing algorithm that identifies a mutation-containing interval and then supports positional cloning to refine the interval which drastically reduces the number of potential candidates allowing for extremely rapid mutation identification. We validated this method using mutants from a forward genetic mutagenesis screen in zebrafish for modifiers of ApoB-lipoprotein metabolism. WheresWalker correctly maps and identifies novel zebrafish mutations in mttp, apobb.1, and mia2 genes, as well as a previously published mutation in maize. Further, we use WheresWalker to identify a previously unappreciated ApoB-lipoprotein metabolism-modifying locus, slc3a2a.

genetics↗

bioRxiv2024

Comprehensive analysis of microbial content in whole-genome sequencing samples from The Cancer Genome Atlas project

In recent years, a growing number of publications have reported the presence of microbial species in human tumors and of mixtures of microbes that appear to be highly specific to different cancer types. Our recent re-analysis of data from three cancer types revealed that technical errors have caused erroneous reports of numerous microbial species found in sequencing data from The Cancer Genome Atlas (TCGA) project. Here we have expanded our analysis to cover all 5,734 whole-genome sequencing (WGS) data sets currently available from TCGA, covering 25 distinct types of cancer. We analyzed the microbial content using updated computational methods and databases, and compared our results to those from two major recent studies that focused on bacteria, viruses, and fungi in cancer. Our results expand upon and reinforce our recent findings, which showed that the presence of microbes is far smaller than had been previously reported, and that many species identified in TCGA data are either not present at all, or are known contaminants rather than microbes residing within tumors. As part of this expanded analysis, and to help others avoid being misled by flawed data, we have released a dataset that contains detailed read counts for bacteria, viruses, archaea, and fungi detected in all 5,734 TCGA samples, which can serve as a public reference for future investigations. One-sentence summaryAnalysis of microbial content in 5,734 whole-genome sequencing samples from TCGA yields a comprehensive new resource for investigating the role of microbes in cancer.

cancer biology↗

bioRxiv2024

PSAURON: a tool for assessing protein annotation across a broad range of species

Evaluating the accuracy of protein-coding sequences in genome annotations is a challenging problem for which there is no broadly applicable solution. In this manuscript we introduce PSAURON (Protein Sequence Assessment Using a Reference ORF Network), a novel software tool developed to assess the quality of protein-coding gene annotations. Utilizing a machine learning model trained on a diverse dataset from over 1000 plant and animal genomes, PSAURON assigns a score to coding DNA or protein sequence that reflects the likelihood that the sequence is a genuine protein coding region. PSAURON scores can be used for genome-wide protein annotation assessment as well as the rapid identification of potentially spurious annotated proteins. Validation against established benchmarks demonstrates PSAURONs effectiveness and correlation with recognized measures of protein quality, highlighting its potential use as a general-purpose method to evaluate gene annotation. PSAURON is open source and freely available at https://github.com/salzberg-lab/PSAURON. One-Sentence SummaryPSAURON is a machine learning-based tool for rapid assessment of protein coding gene annotation.

bioinformatics↗

bioRxiv2024

Novel metagenomics analysis suggests a Vibrio species is associated with stony coral tissue loss disease

Stony coral tissue loss disease (SCTLD) has devastated coral reefs off the coast of Florida and continues to spread throughout the Caribbean. Although a number of bacterial taxa have consistently been associated with SCTLD, no pathogen has been definitively implicated in the etiology of SCTLD. Previous studies have predominantly focused on the prokaryotic community through 16S rRNA sequencing of healthy and affected tissues. Here, we provide a different analytical approach by applying a bioinformatics pipeline to publicly available metagenomic sequencing samples of SCTLD lesions and healthy tissues from four stony coral species. To compensate for the lack of coral reference genomes, we used data from apparently healthy coral samples to approximate a host genome and healthy microbiome reference. These reads were then used as a reference to which we matched and removed reads from diseased lesion tissue samples, and the remaining reads associated only with disease lesions were taxonomically classified at the DNA and protein levels. For DNA classifications, we used a pathogen identification protocol originally designed to identify pathogens in human tissue samples, and for protein classifications, we used a fast protein sequence aligner. To assess the utility of our pipeline, a species-level analysis of a candidate genus, Vibrio, was used to demonstrate the pipelines effectiveness. Our approach revealed both complementary and unique coral microbiome members compared to a prior metagenome analysis of the same dataset. Article SummaryStudies of stony coral tissue loss disease (SCTLD), a devastating coral disease, have primarily used 16S rRNA sequencing approaches to identify putative pathogens. In contrast, this study applied human tissue pathogen identification protocols to SCTLD metagenomic DNA samples. Diseased samples were filtered of host sequences using a k-mer based method since host genomes were unavailable. DNA and protein-level classifications from this novel approach revealed both complementary and unique microbiome members compared to a prior metagenome analysis of the same dataset.

bioinformatics↗

bioRxiv2023

A genome sequence for the threatened whitebark pine

Whitebark pine (WBP, Pinus albicaulis) is a white pine of subalpine regions in western contiguous US and Canada. WBP has become critically threatened throughout a significant part of its natural range due to mortality from the introduced fungal pathogen white pine blister rust (WPBR, Cronartium ribicola) and additional threats from mountain pine beetle (Dendroctonus ponderosae), wildfire, and maladaptation due to changing climate. Vast acreages of WBP have suffered nearly complete mortality. Genomic technologies can contribute to a faster, more cost-effective approach to the traditional practices of identifying disease-resistant, climate-adapted seed sources for restoration. With deep-coverage Illumina short-reads of haploid megametophyte tissue and Oxford Nanopore long-reads of diploid needle tissue, followed by a hybrid, multistep assembly approach, we produced a final assembly containing 27.6 Gbp of sequence in 92,740 contigs (N50 537,007 bp) and 34,716 scaffolds (N50 2.0 Gbp). Approximately 87.2% (24.0 Gbp) of total sequence was placed on the twelve WBP chromosomes. Annotation yielded 25,362 protein-coding genes, and over 77% of the genome was characterized as repeats. WBP has demonstrated the greatest variation in resistance to WPBR among the North American white pines. Candidate genes for quantitative resistance include disease resistance genes known as nucleotide-binding leucine-rich-repeat receptors (NLRs). A combination of protein domain alignments and direct genome scanning was employed to fully describe the three subclasses of NLRs (TNL, CNL, RNL). Our high-quality reference sequence and annotation provide a marked improvement in NLR identification compared to previous assessments that leveraged de novo assembled transcriptomes.

genomics↗

bioRxiv2023

Splam: a deep-learning-based splice site predictor that improves spliced alignments

The process of splicing messenger RNA to remove introns plays a central role in creating genes and gene variants. Here we describe Splam, a novel method for predicting splice junctions in DNA based on deep residual convolutional neural networks. Unlike some previous models, Splam looks at a relatively limited window of 400 base pairs flanking each splice site, motivated by the observation that the biological process of splicing relies primarily on signals within this window. Additionally, Splam introduces the idea of training the network on donor and acceptor pairs together, based on the principal that the splicing machinery recognizes both ends of each intron at once. We compare Splams accuracy to recent state-of-the-art splice site prediction methods, particularly SpliceAI, another method that uses deep neural networks. Our results show that Splam is consistently more accurate than SpliceAI, with an overall accuracy of 96% at predicting human splice junctions. Splam generalizes even to non- human species, including distant ones like the flowering plant Arabidopsis thaliana. Finally, we demonstrate the use of Splam on a novel application: processing the spliced alignments of RNA-seq data in order to identify and eliminate errors. We show that when used in this manner, Splam yields substantial improvements in the accuracy of downstream transcriptome analysis on both poly(A) and ribo-depleted RNA-seq libraries. Overall, Splam offers a faster and more accurate approach to detecting splice junctions, while also providing a reliable and efficient solution for cleaning up erroneous spliced alignments.

genomics↗

bioRxiv2023

Detecting differential transcript usage in complex diseases with SPIT

Differential transcript usage (DTU) plays a crucial role in determining how gene expression differs among cells, tissues, and different developmental stages, thereby contributing to the complexity and diversity of biological systems. In abnormal cells, it can also lead to deficiencies in protein function, potentially leading to pathogenesis of diseases. Detecting such events for single-gene genetic traits is relatively uncomplicated; however, the heterogeneity of populations with complex diseases presents an intricate challenge due to the presence of diverse causal events and undetermined subtypes. SPIT is the first statistical tool that quantifies the heterogeneity in transcript usage within a population and identifies predominant subgroups along with their distinctive sets of DTU events. We provide comprehensive assessments of SPITs methodology in both single-gene and complex traits and report the results of applying SPIT to analyze brain samples from individuals with schizophrenia. Our analysis reveals previously unreported DTU events in six candidate genes.

genomics↗

bioRxiv2023

Investigating Open Reading Frames in Known and Novel Transcripts using ORFanage

ORFanage is a system designed to assign open reading frames (ORFs) to both known and novel gene transcripts while maximizing similarity to annotated proteins. The primary intended use of ORFanage is the identification of ORFs in the assembled results of RNA sequencing (RNA-seq) experiments, a capability that most transcriptome assembly methods do not have. Our experiments demonstrate how ORFanage can be used to find novel protein variants in RNA-seq datasets, and to improve the annotations of ORFs in tens of thousands of transcript models in the RefSeq and GENCODE human annotation databases. Through its implementation of a highly accurate and efficient pseudo-alignment algorithm, ORFanage is substantially faster than other ORF annotation methods, enabling its application to very large datasets. When used to analyze transcriptome assemblies, ORFanage can aid in the separation of signal from transcriptional noise and the identification of likely functional transcript variants, ultimately advancing our understanding of biology and medicine.

bioinformatics↗

bioRxiv2022

CHESS 3: an improved, comprehensive catalog of human genes and transcripts based on large-scale expression data, phylogenetic analysis, and protein structure

The original CHESS database of human genes was assembled from nearly 10,000 RNA sequencing experiments in 53 human body sites produced by the Genotype-Tissue Expression (GTEx) project, and then augmented with genes from other databases to yield a comprehensive collection of protein-coding and noncoding transcripts. The construction of the new CHESS 3 database employed improved transcript assembly algorithms, a new machine learning classifier, and protein structure predictions to identify genes and transcripts likely to be functional and to eliminate those that appeared more likely to represent noise. The new catalog contains 41,356 genes on the GRCh38 reference human genome, of which 19,839 are protein-coding, and a total of 158,377 transcripts. These include 14,863 novel protein-coding transcripts. The total number of transcripts is substantially smaller than earlier versions due to improved transcriptome assembly methods and to a stricter protocol for filtering out noisy transcripts. Notably, CHESS 3 contains all of the transcripts in the MANE database, and at least one transcript corresponding to the vast majority of protein-coding genes in the RefSeq and GENCODE databases. CHESS 3 has also been mapped onto the complete CHM13 human genome, which gives a more-complete gene count of 43,773 genes and 19,968 protein-coding genes. The CHESS database is available at http://ccb.jhu.edu/chess.

genomics↗

bioRxiv2022

The first gapless, reference-quality, fully annotated genome from a Southern Han Chinese individual

We used long-read DNA sequencing to assemble the genome of a Southern Han Chinese male. We organized the sequence into chromosomes and filled in gaps using the recently completed CHM13 genome as a guide, yielding a gap-free genome, Han1, containing 3,099,707,698 bases. Using the CHM13 annotation as a reference, we mapped all genes onto the Han1 genome and identified additional gene copies, generating a total of 60,708 genes, of which 20,003 are protein coding. A comprehensive comparison between the genes revealed that 235 protein-coding genes were substantially different between the individuals, with frameshifts or truncations affecting the protein-coding sequence. Most of these were heterozygous variants in which one gene copy was unaffected. This represents the first gene-level comparison between two finished, annotated individual human genomes.

bioinformatics↗

bioRxiv2022

3D-Beacons: Decreasing the gap between protein sequences and structures through a federated network of protein structure data resources

While scientists can often infer the biological function of proteins from their 3-dimensional quaternary structures, the gap between the number of known protein sequences and their experimentally determined structures keeps increasing. A potential solution to this problem is presented by ever more sophisticated computational protein modelling approaches. While often powerful on their own, most methods have strengths and weaknesses. Therefore, it benefits researchers to examine models from various model providers and perform comparative analysis to identify what models can best address their specific use cases. To make data from a large array of model providers more easily accessible to the broader scientific community, we established 3D-Beacons, a collaborative initiative to create a federated network with unified data access mechanisms. The 3D-Beacons Network allows researchers to collate coordinate files and metadata for experimentally determined and theoretical protein models from state-of-the-art and specialist model providers and also from the Protein Data Bank.

bioinformatics↗

bioRxiv2022

Metagenomic classification with KrakenUniq on low-memory computers

Kraken and KrakenUniq are widely-used tools for classifying metagenomics sequences. A key requirement for these systems is a database containing all k-mers from all genomes that the users want to be able to detect, where k = 31 by default. This database can be very large, easily exceeding 100 gigabytes (GB) and sometimes 400 GB. Previously, Kraken and KrakenUniq required loading the entire database into main memory (RAM), and if RAM was insufficient, they used memory mapping, which significantly increased the running time for large datasets. We have implemented a new algorithm in KrakenUniq that allows it to load and process the database in chunks, with only a modest increase in running time. This enhancement now makes it feasible to run KrakenUniq on very large datasets and huge databases on virtually any computer, even a laptop, while providing the same very high classification accuracy as the previous system.

bioinformatics↗

bioRxiv2021

A reference-quality, fully annotated genome from a Puerto Rican individual

Until 2019, the human genome was available in only one fully-annotated version, GRCh38, which was the result of 18 years of continuous improvement and revision. Despite dramatic improvements in sequencing technology, no other genome was available as an annotated reference until 2019, when the genome of an Ashkenazi individual, Ash1, was released. In this study, we describe the assembly and annotation of a second individual genome, from a Puerto Rican individual whose DNA was collected as part of the Human Pangenome project. The new genome, called PR1, is the first true reference genome created from an individual of African descent. Due to recent improvements in both sequencing and assembly technology, and particularly to the use of the recently completed CHM13 human genome as a guide to assembly, PR1 is more complete and more contiguous than either GRCh38 or Ash1. Annotation revealed 37,755 genes (of which 19,999 are protein-coding), including 12 additional gene copies that are present in PR1 and missing from CHM13. 57 genes have fewer copies in PR1 than in CHM13, 9 map only partially, and 3 genes (all non-coding) from CHM13 are entirely missing from PR1.

genomics↗