NAR Genomics and Bioinformatics最新文献

Whole exome sequencing (WES) is a well-established tool for clinical diagnostics, is more cost-effective and faster to analyse than whole genome sequencing and has been implemented to uplift diagnostic rates in human disease. However, challenges remain to achieve comprehensive and uniform coverage of targets, and high sensitivity and specificity. Differences in genomic target regions and exome capture mechanism between kits may lead to differences in overall coverage uniformity and capture efficiency. Here, we analyse the efficiency of a range of off-the-shelf exome sequencing (ES) kits in capturing their reported targets and the consensus coding sequence (CCDS) regions. Our results show Twist Custom Exome, Twist Human Comprehensive Exome, and Roche KAPA HyperExome V1 perform particularly well at capturing their target regions at 10X and 20X coverage and achieve the highest capture efficiency of CCDS regions upon read downsampling. This was the case despite both Twist kits targeting less than 37Mb in the genome. Our analysis highlights the impact of kit target design on capture efficiency in WES, with kit target size and uniformity of coverage impacting the capture efficiency of CCDS regions. This benchmark will help researchers to make an informed decision based on their needs.

Advances in Oxford Nanopore Technologies (ONT) with the introduction of the r10.4.1 flow cell have reduced the sequencing error rates to <1%. When a reference sequence is known, this allows for accurate variant calling comparable with what is known from the second-generation short-read sequencing technologies, such as Illumina. Additionally, the longer sequence reads provided by ONT enable more efficient mappings, which means the amount of multimapping reads is reduced. However, when the correct reference is not known in advance, and the target reference is highly similar to other references, the multimapping problem is still a concern. Although the ConClave algorithm has provided an accurate solution to the multimapping problem of the second-generation short-read sequencing technologies, it is less effective when resolving the multimapping problems arising from third-generation long-read sequencing technologies. To overcome this problem, we are introducing proximity scoring of alleles, which aids the ConClave algorithm to accurately assign specific alleles from databases containing loci with a high degree of redundancy. Using multilocus sequence typing as a test case, we show that this approach matches the results obtained from sequencing data of Illumina while using limited computational resources that essentially correspond to that of today's smartphones.

The translatability of patient-derived xenograft (PDX)-generated clinical data into patient-specific outcomes for therapeutic guidance is limited by the challenges in generalizability of models across patients, treatments, and cancer types. Previously, machine learning (ML) models have been developed for the two most abundant cancer types, i.e. breast cancer and colorectal cancer, but these are unusable in other cancer types because each treatment/cancer type requires a different model to be trained. Here, we provide an ML framework to train a single pan-cancer, pan-treatment model for predicting treatment outcomes. We show that such models give promising results for all cancer types considered and reproduce the accuracy levels of individually trained cancer types. In the proposed model, all PDX genomic profiles from all cancer types are used as the training data, and instead of partitioning them into cancer types for each model, the cancer type and treatment name are appended as the input features of the training model. Using genomic-only and treatment-only embeddings and combining them with principal component analysis-based dimensionality reduction, our models show promising results and provide a framework for further improvements and real-time use for best treatment selections for cancer patients.

Oxidative damage can introduce G>T mutations upon DNA replication. When this damage occurs ex vivo, sequenced DNA exhibits strand asymmetry, whereby sequence alignment yields G>T mismatches without corresponding C>A mismatches on the complementary strand at a given locus. Strand asymmetry is used to identify potential sequencing artefacts in somatic variant calls in cancer sequencing projects. Consistent with previous studies, we found that the strandedness of this asymmetry is frequently shared across targeted capture regions. However, while some exome sequencing datasets displayed consistent asymmetry relative to the forward reference strand, some surprisingly showed asymmetry relative to the transcription strand. Though oxidation is the principle cause of artefactual G>T mutations, we propose that the asymmetry stems from the use of single-stranded exome capture probes, as we did not find consistent asymmetry in matched whole genome sequencing. We further propose that high levels of asymmetry can be indicative of oxidation artefacts in the reported somatic variant calls of some samples. While most analysed cohorts showed low to moderate asymmetry, in one cohort (testicular germ cell tumour), approximately half of the reported G>T somatic mutations were likely to be oxidative damage artefacts, as indicated by the extent of asymmetry in mismatches and variants.

New reference genomes and transcriptomes are increasingly available across the tree of life, opening new avenues to tackle exciting questions. However, there are still challenges associated with annotating genomes and inferring evolutionary processes and with a lack of methodological standardisation. Here, we propose a new workflow designed for evolutionary analyses to overcome these challenges, facilitating the detection of recombination suppression and its consequences in terms of rearrangements and transposable element accumulation. To do so, we assemble multiple bioinformatic steps in a single easy-to-use workflow. We combine state-of-the-art tools to detect transposable elements, annotate genomes, infer gene orthology relationships, compute divergence between sequences, infer evolutionary strata (i.e. footprints of stepwise extension of recombination suppression) and their structural rearrangements, and visualise the results. This workflow, called EASYstrata, was applied to reannotate 42 published genomes from Microbotryum fungi. We show in further case examples from a plant and an animal that we recover the same strata as previously described. While this tool was developed with the goal to infer divergence between sex or mating-type chromosomes, it can be applied to any pair of haplotypes whose pattern of divergence is of interest. This workflow will facilitate the study of non-model species for which newly sequenced phased diploid genomes are becoming available.

The transcription factor PAX8 is expressed in most ovarian tumors, being associated with increased tumorigenesis. Although recent studies have addressed the gene regulatory functions of PAX8 in ovarian cancer, an integrative analysis of multi-omic and patient data is required to identify the core regulatory network of PAX8 and its prognostic and therapeutic value. Here, we integrate PAX8 chromatin binding and accessibility data in ovarian cancer cells with transcriptomic and patients' data to gain insight into the core gene regulatory network orchestrated by PAX8 in ovarian tumors. Integration of differential chromatin accessibility, transcription factor binding, and gene expression upon PAX8 knockout provides a core regulatory network that explains most of the genes regulated by PAX8. We combine these target genes with patient expression data and find a PAX8 gene signature associated with tumor stemness, a property related to therapy resistance. Indeed, we show that the PAX8 gene signature predicts disease outcome and response to therapy in ovarian cancer patients. Finally, we validated experimentally our results from bioinformatic analyses, thus reassuring their robustness. Our findings uncover a PAX8 core network that represents a promising strategy for targeted antitumor therapies and open new pathways to fight against ovarian cancer resistance.

Myxococcus xanthus DZ2, a model myxobacterium, has three reported genome assemblies, including two recent complete assemblies (MxDZ2_Tam and MxDZ2_Nan) from the same culture stock. These assemblies misreported their circular nature and differed by 6.4 kb, raising questions about their accuracy. After removing duplicate ends, aligning genomes to the origin of replication, and circularization, this computational analysis revealed a minimal 32 bp difference, with MxDZ2_Tam being slightly larger. Forty sequence variations including 38 indels and two substitutions, were impacting 18 coding genes via frameshift mutations. Although PacBio-HiFi technology boasts a low error rate, it remains higher than the 454-platform used for the earlier MxDZ2_Kirby draft assembly. Therefore, using MxDZ2_Kirby as a reference, we constructed a "truly circular" genome for M. xanthus DZ2. Additionally, analysis of Mx-alpha regions, involved in antagonism via the toxin gene sitA, across 61 myxobacterial genomes identified their presence in five taxonomically polyphyletic species, potentially influencing their physiology, development, and ecological interactions beyond predation. Only M. xanthus DZ2 and DZF1 contained all three Mx-alpha regions, whereas M. xanthus DK1622 has only one. Overall, this study underscores the need for meticulous validation of sequencing-based genome assemblies and their variations and provides novel insights into Mx-alpha regions as potential adaptive elements in myxobacteria.

Single-cell RNA sequencing (scRNA-seq) has significantly deepened our understanding of cellular heterogeneity and cell type interactions, providing insights into how cell populations adapt to environmental variability. However, its lack of spatial context limits intercellular analysis. Similarly, existing spatial transcriptomics (ST) data often lack single-cell resolution, restricting cellular mapping. To address these limitations, we introduce ST-deconv, a deep learning-based deconvolution model that integrates spatial information. ST-deconv leverages contrastive learning to enhance the spatial representation of adjacent spots, improving spatial relationship inference. It also employs domain-adversarial networks to improve generalization and deconvolution across diverse datasets. Moreover, ST-deconv can generate large-scale, high-resolution spatial transcriptomic data with cell type labels from single-cell input, facilitating the learning of spatial cell type composition. In benchmarking experiments, ST-deconv outperforms traditional methods, reducing the root mean square error (RMSE) by 13% to 60%, with an RMSE as low as 0.03 for high spatial correlation datasets and 0.07 for low spatial correlation datasets across different transcriptomic contexts. Reconstructing real tissue structure, a purity of 0.68 on mouse olfactory bulb (MOB) and a cell type correlation of 0.76 on human pancreatic ductal adenocarcinoma (PDAC) were achieved. These advancements make ST-deconv a powerful tool for enhancing spatial transcriptomics and downstream analyses of intercellular interactions.

Tissue microarrays (TMAs) enable researchers to analyse hundreds of tissue samples simultaneously by embedding multiple samples into single arrays, enabling conservation of valuable tissue samples and experimental reagents. Moreover, profiling TMAs allows efficient screening of tissue samples for translational and clinical applications. Multiplexed imaging technologies allow for spatial profiling of proteins at single-cell resolution, providing insights into tumour microenvironments and disease mechanisms. High-plex spatial single-cell protein profiling is a powerful tool for biomarker discovery and translational cancer research; however, there remain limited options for end-to-end computational analysis of this type of data. Here, we introduce PRISM, a Python package for interactive, end-to-end analyses of TMAs with a focus on translational and clinical research using multiplexed proteomic data. PRISM leverages the SpatialData framework to standardize data storage and ensure interoperability with single-cell and spatial analysis tools. It consists of two main components: TMA Image Analysis for marker-based tissue masking, TMA dearraying, cell segmentation, and single-cell feature extraction; and AnnData Analysis for quality control, clustering, iterative cell-type annotation, and spatial analysis. Integrated as a plugin within napari, PRISM provides an intuitive and purely interactive graphical interface for real time and human-in-the-loop analyses. PRISM supports efficient multi-resolution image processing and accelerates bioinformatics workflows using efficient scalable data structures, parallelization and GPU acceleration. By combining modular flexibility, computational efficiency, and a completely interactive interface, PRISM simplifies the translation of raw multiplexed images to actionable clinical insights, empowering researchers to explore and interact effectively with spatial omics data.

In high-throughput sequencing (HTS) studies, sample-to-sample variation in sequencing depth is driven by technical factors, and not by variation in the scale (size) of the biological system. Typically a statistical normalization removes unwanted technical variation in the data or the parameters of the model to enable differential abundance analyses. We recently showed that all normalizations make implicit assumptions about the unmeasured system scale and that errors in these assumptions can dramatically increase false positive and false negative rates. We demonstrated that these errors can be mitigated by accounting for uncertainty using a scale model, which we integrated into the ALDEx2 R package. This article provides new insights focusing on the application to transcriptomic analysis. We provide transcriptomic case studies demonstrating how scale models, rather than traditional normalizations, can reduce false positive and false negative rates in practice while enhancing the transparency and reproducibility of analyses. These scale models replace the need for dual cutoff approaches often used to address the disconnect between practical and statistical significance. We demonstrate the utility of scale models built based on known housekeeping genes in complex metatranscriptomic datasets. Thus this work provides guidance on how to incorporate scale into transcriptomic data sets.