NAR Genomics and Bioinformatics最新文献

In this computational study, we address the challenge of predicting protein functions following mutations by fine-tuning protein language models (PLMs) using a novel tokenization strategy, hint token learning (HTL). To evaluate the effectiveness of HTL, we benchmarked this approach across four pretrained models with varying architectures and sizes on four diverse protein mutational datasets. Our results showed significant improvements in weighted F1 scores in most cases when HTL was applied. To understand how HTL enhances protein mutational predictions, we trained sparse autoencoders on embeddings derived from the fine-tuned PLMs. Analysis of the latent spaces revealed that the number of activated residues within functional protein domains increased by PLM training with HTL. These findings indicate that PLMs fine-tuned with HTL may capture more biologically relevant representations of proteins. Our study highlights the potential of HTL to advance protein function prediction and provides insights into how HTL enables PLMs to capture mutational impacts at the functional level. All data and code are available at: https://github.com/ejp-lab/EJPLab_Computational_Projects/tree/master/HintTokenLearning.

Privacy is paramount in health data, particularly in human genetics, where information extends beyond individuals to their relatives. Metagenomic datasets contain substantial human genetic material, necessitating careful handling to mitigate data leakage risks when sharing or publishing. The same applies to genetic datasets from the environment or datasets from contaminated laboratory samples, although to a lesser extent. Completely removing human sequence data while retaining unbiased nonhuman reads is not achievable currently, but several tools exist. To address these topics, we developed nf-core/detaxizer, a nextflow-based pipeline that employs Kraken2 and bbmap/bbduk for taxonomic classification, identifying and optionally filtering Homo sapiens reads. Due to its generalized design, other taxa can also be classified and filtered. We benchmark its filtering efficacy for human reads against Hostile and CLEAN, demonstrating its utility for secure data preprocessing. The comparison showed that the choice of tool and database can result in differences of up to an order of magnitude in both the amount of human data not removed and the amount of microbial data mistakenly removed. As part of the nf-core initiative, nf-core/detaxizer adheres to best practices, leveraging containerized dependencies for streamlined installation. The source code is openly available under the MIT license: https://github.com/nf-core/detaxizer.

Unraveling the genetic complexity of the human immunoglobulin heavy (IGH) chain locus provides valuable insights into the mechanisms underlying the efficacy and specificity of the adaptive immune response. Despite its crucial role, the IGH locus remains insufficiently characterized, with its allelic diversity and polymorphisms inadequately investigated. In this study, we present an analysis of the human IGH locus, incorporating 15 human genome assemblies from diverse ancestries, including African, European, Asian, Saudi, and mixed backgrounds. Through our examination of both maternal and paternal assemblies, we uncover novel IGH alleles, copy number variations (CNV), and polymorphisms, particularly within the variable (IGHV) region. Our findings reveal extensive and previously uncharacterized genetic variability in the constant (IGHC) region and distinct IMGT CNV forms across individuals. This research contributes to a significant enrichment of the IMGT^® IGH reference directory, databases, tools and web resources, and lays the groundwork for an IMGT^® haplotype database which can be progressively enriched as additional datasets become available. Such a resource promises to propel personalized immunogenomics forward, with exciting applications in cancer immunotherapy, COVID-19, and other immune-related diseases.

Somatic copy number alterations (CNAs) are hallmarks of cancer. Current algorithms that call CNAs from whole-genome sequenced (WGS) data have not exploited deep learning methods owing to computational scaling limitations. Here, we present a novel deep-learning approach, araCNA, trained only on simulated data that can accurately predict CNAs in real WGS cancer genomes. araCNA uses novel transformer alternatives (e.g. Mamba) to handle genomic-scale sequence lengths (∼1M) and learn long-range interactions. Results are extremely accurate on simulated data, and this zero-shot approach is on par with existing methods when applied to 50 WGS samples from the Cancer Genome Atlas. Notably, our approach requires only a tumour sample and not a matched normal sample, has fewer markers of overfitting, and performs inference in only a few minutes. araCNA demonstrates how domain knowledge can be used to simulate training sets that harness the power of modern machine learning in biological applications.

Next-generation sequencing has greatly advanced genomics, enabling large-scale studies of population genetics and complex traits. Genomic DNA (gDNA) from white blood cells has traditionally been the main data source, but cell-free DNA (cfDNA), found in bodily fluids as fragmented DNA, is increasingly recognized as a valuable biomarker in clinical and genetic studies. However, a direct comparison between cfDNA and gDNA has not been fully explored. In this study, we analyzed cfDNA and gDNA from 186 healthy individuals, using the same sequencing platform. We compared sequencing quality, variant detection, allele frequencies (AF), genotype concordance, population structure, and genomic association results (genome-wide association study and expression quantitative trait locus). While cfDNA showed higher duplication rates and lower effective sequencing depth, both DNA types displayed similar quality metrics at the same depth. We also observed that significant depth differences between cfDNA and gDNA were mainly found in centromeric regions. While gDNA identified more variants with more uniform coverage, AF spectra, population structure, and genomic associations were largely consistent between the two DNA types. This study provides a detailed comparison of cfDNA and gDNA, highlighting the potential of cfDNA as an alternative to gDNA in genomic research. Our findings could serve as a reference for future studies on cfDNA and gDNA.

Composite hypothesis testing using summary statistics is a well-established approach for assessing the effect of a single marker or gene across multiple traits or omics levels. Numerous procedures have been developed for this task and have been successfully applied to identify complex patterns of association between traits, conditions, or phenotypes. However, existing methods often struggle with scalability in large datasets or fail to account for dependencies between traits or omics levels, limiting their ability to control false positives effectively. To overcome these challenges, we present the qch_copula approach, which integrates mixture models with a copula function to capture dependencies between traits or omics and provides rigorously defined P-values for any composite hypothesis. Through a comprehensive benchmark against eight state-of-the-art methods, we demonstrate that qch_copula controls Type I error rates effectively while enhancing the detection of joint association patterns. Compared to other mixture model-based approaches, our method notably reduces memory usage during the EM algorithm, allowing the analysis of up to 20 traits and 10⁵-10⁶ markers. The effectiveness of qch_copula is further validated through two application cases in human and plant genetics. The method is available in the R package qch, accessible on CRAN.

The dynamics of transcriptional elongation influence many biological activities, such as RNA splicing, polyadenylation, and nuclear export. To quantify the elongation rate, a typical method is to treat cells with drugs that inhibit RNA polymerase II (Pol II) from entering the gene body and then track Pol II using Pro-seq or Gro-seq. However, the downstream data analysis is challenged by the problem of identifying the transition point between the gene regions inhibited by the drug and not, which is necessary to calculate the transcription rate. Although the traditional hidden Markov model (HMM) can be used to solve it, this method is complicated with its hidden variable and many parameters to be estimated. Hence, we developed the R package proRate, which identifies the transition point with a novel least sum of squares (LSS) method and calculates the elongation rate accordingly. In addition, proRate also covers other functions frequently used in transcription dynamic study, including metagene plotting, pause index calculation, gene structure analysis, etc. The effectiveness of this package is proved by its performance on three Pro-seq or Gro-seq datasets, showing higher accuracy than HMM. proRate is freely available at https://github.com/yuabrahamliu/proRate or https://github.com/FADHLyemen/proRate.

Meiotic crossovers promote correct chromosome segregation and the shuffling of genetic diversity. However, the measurement of crossovers remains challenging, impeding our ability to decipher the molecular mechanisms that are necessary for their formation and regulation. Here we demonstrate a novel repurposing of the single-nucleus Assay for Transposase Accessible Chromatin with sequencing (snATAC-seq) as a simple and high-throughput method to identify and characterize meiotic crossovers from haploid testis nuclei. We first validate the feasibility of obtaining genome-wide coverage from snATAC-seq by using ATAC-seq on bulk haploid mouse testis nuclei, ensuring adequate variant detection for haplotyping. Subsequently, we adapt droplet-based snATAC-seq for crossover detection, revealing >25 000 crossovers in F₁ hybrid mice. Comparison between the wild type and a hyper-recombinogenic Fancm-deficient mutant mouse model confirmed an increase in crossover rates in this genotype, however with a distribution which was unchanged. We also find that regions with the highest rate of crossover formation are enriched for PRDM9. Our findings demonstrate the utility of snATAC-seq as a robust and scalable tool for high-throughput crossover detection, offering insights into meiotic crossover dynamics and elucidating the underlying molecular mechanisms. It is possible that the research presented here with snATAC-seq of haploid post-meiotic nuclei could be extended into fertility-related diagnostics.

Genome graphs provide a powerful reference structure for representing genetic diversity. Their structure emphasizes the polymorphic regions in a collection of genomes, enabling network-based comparisons of population-level variation. However, current tools are limited in their ability to quantify and compare structural features across large genome graphs. We introduce GViNC, Genome graph Visualization, Navigation, and Comparison, a novel framework that enables partitioning genome graphs into interpretable subgraphs, mapping linear coordinates to graph nodes, and summarizing both local and global structural variation using new metrics for variability, hypervariability, and graph distances. We applied GViNC to multiple pan-genomic and population-specific genome graphs constructed with over 85M variants in 2504 individuals from the 1000 Genomes Project. We found that genomic complexity varied by ancestry and across chromosomes, with rare variants increasing variability by 10-fold and hypervariability by 50-fold. GViNC highlighted key regions of the human genome, such as Human Leukocyte Antigen and DEFB loci, and many previously unreported high-diversity regions, some with population-specific signatures in protein-coding and regulatory genes. By bridging sequence-level variation and graph-level topology, GViNC enables scalable, quantitative exploration of genome structure across populations. GViNC's versatility can aid researchers in extensively investigating the genetic diversity of different cohorts, populations, or species of interest.

[This corrects the article DOI: 10.1093/nar/lqaf063.].