BMC Bioinformatics最新文献

Background: Transcriptomic profiling technologies have advanced the analysis of biological and toxicological responses. However, substantial differences in probe design, dynamic range, gene coverage, and preprocessing pipelines across platforms introduce artifacts that limit cross-study integration and hinder the reuse of historical datasets. We aim to develop computational methods for accurate cross-platform translation to maximize the value of legacy resources.

Results: We present TransPlatformer a deep learning framework for translating gene expression profiles across heterogeneous toxicogenomics platforms. TransPlatformer employs a novel attention-based architecture to map high-dimensional fold-change vectors from legacy microarray technologies to current platforms. Models are trained and evaluated using DrugMatrix, spanning three technological generations. We investigate mixed-tissue, single-tissue, and cross-tissue training paradigms and benchmark performance against multilayer perceptron and matrix-completion baselines. In mixed-tissue training, TransPlatformer achieves a greater than 50% reduction in mean absolute error (0.043 vs. 0.09) and nearly doubles Pearson correlation (≈ 0.71 vs. 0.37) relative to baseline methods. Importantly, TransPlatformer preserves rare but biologically meaningful over- and under-expressed signals, with mean absolute error below 0.22. Single-tissue models yield further improvements for well-represented organs, such as a 10% reduction in liver mean absolute error, while underscoring the need for data augmentation strategies in low-sample tissues.ra CONCLUSIONS: TransPlatformer provides an effective and scalable computational solution for cross-platform transcriptomic translation. By enabling biologically faithful harmonization of gene expression data, the proposed approach facilitates the reuse of legacy toxicogenomics datasets, enhances downstream biomarker discovery, and supports more reproducible predictive modeling in toxicology.

Background: Advancements in spatially resolved single-cell technologies are transforming our understanding of tissue architecture and disease microenvironments. However, analyzing the resulting high-dimensional, gigabyte-scale datasets remains challenging due to fragmented workflows, intensive computational requirements, and a lack of accessible, user-friendly tools for non-technical researchers.

Results: We introduce SPAC (analysis of SPAtial single-Cell datasets), a scalable, web-based platform for efficient and reproducible single-cell spatial analysis. SPAC employs a four-tier architecture that includes a modular Python-based analysis engine, seamless integration with high-performance computing (HPC) and GPU acceleration, an interactive browser interface for no-code workflow configuration, and a real-time visualization layer powered by Shiny for Python dashboards. This design empowers distinct user roles: data scientists can extend and customize analysis modules, while bench scientists can execute complete workflows and interactively explore results without coding. Built-in reproducibility features and collaborative workflow support ensure that analyses are transparent and easily shared across research teams. Using a 2.6-million-cell multiplex imaging dataset from a 4T1 breast tumor model as a benchmark, SPAC reduced unsupervised clustering time from ~3 hours on a CPU to under 10 minutes with GPU acceleration, achieving more than a 20-fold speedup. It also enabled fine-grained spatial profiling of distinct tumor microenvironment compartments, demonstrating the platform's scalability and performance.

Conclusions: SPAC addresses major barriers in single-cell spatial analysis by uniting an intuitive, user-friendly interface with scalable, high-performance computation in a robust and reproducible framework. By streamlining complex analyses and bridging the gap between experimental and computational researchers, SPAC fosters collaborative workflows and accelerates the transformation of large-scale spatial datasets into actionable biological insights.

Background: DNA sequences are fundamental carriers of genetic information, and their accurate classification is essential for understanding gene regulation, disease mechanisms, and translational genomics. Existing encoding methods often fail to capture both local and long-range dependencies simultaneously.

Results: We introduce EDEN (Expected Density of Nucleotide Encoding), a unified multiscale encoding framework based on kernel density estimation. EDEN captures position-specific and context-dependent nucleotide patterns and integrates them into a hybrid deep learning architecture. Across sixteen benchmark datasets covering promoter detection, core promoter detection, and transcription factor binding prediction, EDEN achieves the best average performance while using orders of magnitude fewer parameters compared with state-of-the-art models. All source code, pretrained models, and datasets are publicly available at: https://github.com/zabihis/EDEN .

Conclusions: EDEN provides an efficient, biologically informed, and interpretable multiscale representation for genomic sequence classification. Its favorable parameter-performance ratio and robust consistency across tasks underscore its practicality for large-scale genomic applications.

Background: Multiplex Substrate Profiling by Mass Spectrometry (MSP-MS) is a powerful method for determining the substrate specificity of proteolytic enzymes, which is essential for developing protease inhibitors, diagnostics, and protease-activated therapeutics. However, the complex datasets generated by MSP-MS pose significant analytical challenges and have limited accessibility for non-specialist users.

Results: We developed mspms, a Bioconductor R package with an accompanying graphical interface, to streamline the analysis of MSP-MS data. Mspms standardizes workflows for data preparation, processing, statistical analysis, and visualization. The tool is designed for accessibility, serving advanced users through the R package and broader audiences through a web-based interface. We validated mspms using data from four well-characterized cathepsins (A-D), demonstrating that it reliably captures expected substrate specificities.

Conclusions: mspms is the first publicly available, comprehensive platform for MSP-MS data analysis downstream of peptide identification and quantification. It integrates preprocessing, normalization, statistical testing, and visualization into a single, transparent, and user-friendly framework, making it a valuable resource for the protease research community. The package is distributed via Bioconductor, and a graphical interface is available online for interactive use.

Background: High-throughput sequencing technologies generate massive amounts of FASTQ data comprising nucleotide sequences, quality scores, and read identifiers, necessitating efficient compression to alleviate storage and transmission burdens. Compared to general-purpose compressors, specialized FASTQ compressors achieve higher compression performance by exploiting the inherent redundancy in FASTQ files. However, existing FASTQ-specialized compressors often suffer from limited data applicability and tend to over-optimize either compression ratio or compression speed at the expense of the other.

Results: We present zDUR, a reference-free FASTQ compressor designed for efficient and scalable handling of next-generation sequencing data across diverse platforms and sequencing data types. Benchmarking against six reference-free compressors on 15 representative datasets spanning four sequencing data types demonstrates that zDUR achieves a favorable overall balance between compression ratio and speed, with broad applicability across data types. In particular, on single-cell RNA-seq and spatial transcriptomics datasets, zDUR achieves over a tenfold increase in runtime performance while maintaining higher compression ratios than SPRING, one of the state-of-the-art reference-free FASTQ compressors.

Conclusions: zDUR offers a scalable and efficient solution for reference-free FASTQ compression, balancing performance, speed, and usability across diverse datasets.

Background: Large, rare copy number variants (CNVs) are a main source of genetic variation in the genome and are important in both evolution and disease risk. CNVs can be detected using different data sources, including genome sequencing, genotyping arrays and quantitative PCR experiments, but in most large cohorts, genotyping arrays remain the most prevalent source. Current methods to call CNVs from genotyping array data suffer from high false positive rates and while multiple approaches, including QC filtering, visual inspection of intensity tracks, and wet-lab validation are commonly applied to counter this problem, such methods are often non-specific (QC filtering) or inefficient (visual and wet-lab validations) at a genome-wide scale.

Results: We have assembled the largest collection of human-verified CNV calls using visual validation, totalling almost 60,000 calls from 22,500 samples from three cohorts genotyped on several different arrays. Across all cohorts our visual validation found the majority of CNV calls to be false positive (53.7%) or unclear (9.7%). The false positive fraction varied substantially across datasets and genomic regions, and we show that existing filtering methods based on QC metrics are inefficient in removing false calls. Given the supremacy of visual validation over existing filtering methods in controlling the false positive fraction, we used a subset of our visual validation dataset to train a convolutional neural network to automate the validation of CNVs through machine vision. We tested the efficacy of the model using the remainder of the dataset and found the performance exceeded 90% in most measures, approximating that of a human analyst. Orthogonal validation with genome sequencing data found our visual validation to be highly accurate, with only 1.7% of calls supported by the sequencing dataset deemed as false by the human analyst, and a further 7.5% deemed as unclear.

Conclusions: Visual inspection is the only effective validation approach for CNV calls. Our model is capable of automating this task at scale with very high accuracy, as shown by testing both within-sample and out-of-sample. The software is available as an R package at https://github.com/SinomeM/CNValidatron_fl.

Background: Microbiome sequencing data are often collected from several body sites and exhibit dependencies. Our objective is to develop a model that enables joint analysis of data from different sites by capturing the underlying cross-site dependencies. The proposed model incorporates (i) latent factors shared across sites to explain common subject effects and to serve as the source of correlation between the sites and (ii) mixtures of latent factors to allow heterogeneity among the subjects in cross-site associations.

Results: Our simulation studies demonstrate that stronger associations between two sites lead to greater efficiency loss in regression analysis when such dependence is ignored in modeling. In a case study involving samples collected from a study on the female urogenital microbiome with aging, our model leads to the detection of covariate associations of the vaginal and urine microbiomes that are otherwise not statistically significant under a similar regression model applied to the two sites separately.

Conclusions: We propose a latent factor model for microbiome sequencing data collected from multiple sites. It captures the presumptive underlying cross-site associations without compromising estimation accuracy or inference efficiency in the absence of such associations. In addition, our proposed model improves predictive performance by enabling the prediction of microbial abundance at one site based on observations from another. We also provide an extended framework that allows for clustering of subjects (samples) and cluster-specific levels of paired association. Under this extended framework, clusters can be classified according to their association strengths.