Bioinformatics advances最新文献

Motivation: Dysregulation of Ca²⁺-signaling genes has been shown in some types of cancer; however, it is virtually unknown in hepatitis B-derived hepatocellular carcinoma (HBV-HCC). Here, we evaluate the transcriptional and epigenetic regulation of Ca²⁺-signaling genes in HBV-HCC and whether their expression is associated with cancer hallmarks, and prognostic potential.

Results: We identified 432 differentially expressed Ca²⁺-signaling genes in HBV-HCC, including 134 that are specific to this condition, and were not found in non-HBV HCC. Fifty-three of these genes were associated with cancer hallmarks, of which 17 exhibited potential prognostic value by Cox multivariate analyses. We also provide new evidence for epigenetic regulation by post-transcriptional histone modifications and DNA methylation at the promoter of some of these genes. Finally, using Least Absolute Shrinkage and Selection Operator (LASSO) regression, we identified a four-gene prognostic signature (FBLN1, STC2, C1R, and F2RL2) that robustly stratified patient outcomes. This study presents the first integrative transcriptomic and epigenetic analysis of Ca²⁺-signaling genes in HBV-HCC, introducing a novel four-gene signature with prognostic potential. These findings highlight the relevance of a dysregulation of a subset of Ca²⁺-signaling genes as a distinctive feature of HBV-HCC.

Availability and implementation: All data generated or analyzed during this study are included in this article.

Motivation: Cell annotation is fundamental for single-cell data interpretation. Accurate annotation allows us to identify cell types, understand their functions, trace developmental trajectories, and pinpoint alterations associated with a condition of interest. However, this complex process demands extensive manual curation, domain expertise, and proficiency across diverse bioinformatics tools. These challenges impede reproducibility and consistency.

Results: We have developed a new approach for semi-automatic cell type annotation, powered by large language models (LLMs). Given the input single-cell data, we first perform dimension reduction, clustering, and differential analysis to identify distinct cell groups and their respective markers. Next, we utilize Meta's Llama and structured prompting to infer potential cell types. This approach greatly reduces manual labor from researchers while maintaining biological accuracy through enforced ontology, tissue context, and marker gene signatures. Our solution is freely accessible through our web-based platform named CytoAnalyst, hosted on a high-performance infrastructure with optimized networking and storage capabilities. CytoAnalyst also offers capabilities for quality control, embedding analysis, clustering, differential analysis, gene set analysis, cell enrichment, cell type annotation, and pseudo-time trajectory inference.

Availability and implementation: CytoAnalyst is freely available at https://cytoanalyst.tinnguyen-lab.com/. The CytoAnalyst handbook, including step-by-step tutorials and example case studies, is available at https://cytoanalyst.tinnguyen-lab.com/docs/.

Motivation: Alternative splicing (AS) effects on cellular functions can be captured by studying changes in the underlying protein-protein interactions (PPIs). Because AS results in the gain or loss of exons, existing methods for predicting AS-related PPI changes utilize known inter-protein exon-exon interactions (EEIs), which cover <0.5% of known human PPIs. Hence, there is a need to extend the limited EEI knowledge to advance the functional understanding of AS. Here, we explore whether existing 3-dimensional (3D) protein structure-based computational PPI interface prediction (PPIIP) methods, originally designed to predict inter-protein residue-residue interactions (RRIs), can be utilized to predict EEIs.

Results: We evaluate the PPIIP methods for the RRI- and EEI-prediction tasks using all known experimentally determined 3D structures of human protein heterodimers from the Protein Data Bank available at the time of data collection. From these heterodimers, we determined $\sim 230\hspace{0.17em}000$ RRIs and $\sim 20\hspace{0.17em}400$ EEIs as ground truth. We provide the first evidence of the adaptability of existing PPIIP methods to predict EEIs, with a performance score of up to $\sim 76\%$ based on the area under the receiver operating characteristic curve. Insights, data, and computational pipelines from our study can guide future developments of computational methods for solving the task of predicting EEIs.

Availability and implementation: Data and source code are available at https://github.com/lieboldj/EEIpred.

Biological knowledgebases facilitate discovery across the life sciences by structuring experimental findings into human-readable and computable formats. These essential resources are maintained by a small number of professional biocurators worldwide and face combined chronic underfunding and the exponential growth of the literature. In this perspective, we review how artificial intelligence, particularly large language models and agentic systems, can augment literature-curation workflows. Applications include literature recommendation, entity recognition, data extraction, summarization, ontology development, and quality control with emphasis on published use cases at Global Core BioData Resources and ELIXIR Core Data Resources. We identify key challenges, including the scarcity of training data, difficulty in extracting complex relationships, and concerns about error propagation. To address these challenges, we propose a human-in-the-loop framework where generative artificial intelligence approaches accelerate routine tasks while curators provide critical evaluation and domain expertise. We also propose practical recommendations for the community, including the creation of shared benchmark datasets, harmonized evaluation frameworks, and best-practice guidelines for transparent human-in-the-loop AI deployment in biocuration. These synergistic partnerships will be critical to ensure biological rigour, accelerating knowledge integration while maintaining the quality essential for trusted biological resources.

Motivation: Community detection methods are applied to single cell RNA sequencing (i.e. scRNA-seq) and mass cytometry data to efficiently identify major cell types and their subtypes, but their computational demands increase, particularly given the substantial growth in dataset sizes. The Leiden algorithm, an emerging method in this field, offers inherent parallelism that remains underutilized due to the limited parallel processing capabilities offered by today's modern multi-core CPUs, which have fewer than 100 cores (typically 32-64 CPUs). However, Leiden can achieve significant performance gains when implemented on GPUs. GPUs offer high memory bandwidth and an extensive array of parallel processing units that map well to the parallelism in Leiden. As far as we know, cuGraph is the only implementation that has mapped the Leiden algorithm to GPUs, using a blend of Python and C languages. However, it only supports undirected graphs, potentially discarding the valuable information carried by edge directionality. In addition, this Python implementation for GPUs is comparatively slower than a C/C++ based implementation, reducing the significant performance gains provided by a GPU-based speedup. Conversely, a C/C++ based implementation optimizes performance more effectively, ensuring an accurate baseline comparison when performing GPU acceleration.

Results: We developed a tool named gLeiden, a lightweight CUDA C++ based GPU implementation of the Leiden algorithm and, to the best of our knowledge, the very first GPU implementation that supports directed graphs, which generally demands nearly twice the computational time and memory resources compared to undirected graphs. The results show that our directed gLeiden outperforms the directed cLeiden version and shows 11× and 12× speedup on very large datasets. Our undirected ucLeiden and ugLeiden implementations significantly outperform the original Java version, with up to 42× speedup on large datasets. However, when comparing the undirected ugLeiden version with cuGraph, ugLeiden performance is comparable on smaller datasets and 58% faster on larger datasets. These results position our GPU-based Leiden implementation as a high-performance alternative to existing state-of-the-art community detection tools.

Availability and implementation: The source code and sample data are available at: https://github.com/Beenishgul/Leiden and https://figshare.com/s/3b51e463a56e2a374bdf.

Motivation: Cancer screening using liquid biopsy technology has become standard in modern clinical and preventive oncology. This method analyzes cell-free DNA (cfDNA) circulating in a patient's bloodstream. While mutation-based diagnostics using deep exome sequencing are highly sensitive and specific, an alternative approach involves examining cfDNA fragment size distribution profiles. This method is less expensive and can be derived from low-depth whole genome sequencing (WGS).

Results: Our study presents DeepFRAG: a new cancer detection method based on deep learning analysis of cfDNA fragment size distribution profiles using wavelet transform. We utilized two independent cohorts comprising 73 patients with stage III and IV cancers (breast, colorectal, pancreatic, lung, and liver) and 80 healthy individuals. We introduced an original data augmentation technique specific to WGS fragment size data, ensuring sufficient data for training the deep learning model. The proposed method demonstrated high accuracy, with a median test AUROC (area under the receiver operating characteristic curve) of 0.974 and a sensitivity of 96.1% at 98.8% specificity. Our approach offers several advantages, including high accuracy, cost-effectiveness, robustness, and suitability for detecting major cancer types. This method represents a promising advancement in cancer screening technology, expanding the options available for noninvasive cancer detection, with the goal of improving patient outcomes.

Availability and implementation: Data and source code are available at https://github.com/andreykoch/DeepFRAG.

Motivation: Predicting the impact of missense mutations on protein structure and function is a fundamental challenge for cancer research and clinical applications. Despite all the computational advances and, more recently, the use of artificial intelligence (AI), assessing the functional consequences of residue substitutions remains a challenging task. Proteins have complex three-dimensional structures, where the maintenance of their functionality depends on chemical interactions between amino acid residues. Single substitutions can affect these interactions, leading to more profound structural changes that are difficult to visualize.

Results: Here, we present CaRinDB, a database that integrates cancer-associated missense mutation data, functional predictions, molecular features, allelic frequencies, and residue interaction network (RIN) parameters derived from Protein Data Bank structures and AlphaFold models. Users can access and explore variant information through an intuitive web portal, with custom plots and tables to visualize and analyze cancer-associated mutation data. CaRinDB is the first database that unites distinct annotation features of cancer-associated mutations and their structural impacts, utilizing RINs graph parameters and a source of compiled and processed data for the development of AI tools.

Availability and implementation: CaRinDB is freely available at https://bioinfo.imd.ufrn.br/CaRinDB/. The integrated development environment used was Jupyter notebooks, available on GitHub (https://github.com/evomol-lab/CaRinDB). CaRinDB web interface was implemented in R and Shiny.

Motivation: Ancient DNA studies rely heavily on the EIGENSTRAT genotype format (.geno, .ind, .snp) for standard population genetic analyses including PCA, f-statistics, and qpWave/qpAdm. However, there is limited software available for processing EIGENSTRAT format data. pygenstrat , a Python package, is presented here, providing a command-line interface for comprehensive EIGENSTRAT data processing with extensive filtering, subsetting, and conversion options. pygenstrat implements memory-efficient, chunked processing algorithms for handling large ancient DNA datasets with low memory usage. It supports comprehensive operations, including updating individual and SNP files, subsetting datasets by selecting individuals or SNPs, filtering by minor allele frequency and missingness, pseudo-haploidisation, allele polarization, as well as conversion between EIGENSTRAT (text) and ANCESTRYMAP (binary) formats. Its modular architecture and Python implementation enable rapid integration with custom pipelines and future extensions.

Results: Benchmarking on the Allen Ancient DNA Resource (v 62.0) shows 2×-15× speedups and 90%-95% memory reduction compared to convertf, while producing equivalent outputs for standard operations. These improvements reduce turnaround time in ancient DNA workflows and facilitate reproducible processing.

Availability and implementation: pygenstrat is open-source, available at https://github.com/dkoptekin/pygenstrat.

Motivation: Exploration of large-scale biological datasets remains a central challenge in computational biology. While many tools are available, they are often developed in isolation, leading to fragmented workflows, duplicated efforts, and limited reproducibility. There is a pressing need for flexible, standardized solutions that unify exploratory data analysis and biomarker discovery across diverse platforms.

Results: We present HDAnalyzeR, a user-friendly and extensible R package for the streamlined analysis of high-dimensional biological data. HDAnalyzeR provides modular, reproducible workflows that support a range of analyses, from quality control and dimensionality reduction to differential expression and enrichment analysis. The package features built-in visualization, metadata-aware modeling, and seamless integration with interactive apps and learning resources. We also present two case studies, where HDAnalyzeR dramatically reduced analysis time and code complexity while providing biologically meaningful insights, such as classification of blood cancer types with AUC = 1.0 and identification of thousands of solid tumor-associated genes. HDAnalyzeR is designed to support both beginner users and experienced bioinformaticians, promoting transparency, reproducibility, and publication-quality output.

Availability and implementation: HDAnalyzeR is freely available both as an open-source R package at https://github.com/kantonopoulos/HDAnalyzeR and a web application at https://hdanalyzer.serve.scilifelab.se.

Motivation: Advances in sequencing technologies have enabled researchers to sequence whole genomes rapidly and cheaply. However, despite improvements in genome assembly, structural genome annotation (i.e. the identification of protein-coding genes) remains challenging, particularly for eukaryotic genomes. It requires using several approaches (typically ab initio, transcriptomics, and homology search), which may give substantially different results. Deciding which gene models to retain in a consensus is far from trivial, and automated approaches tend to lag behind laborious manual curation efforts in accuracy.

Results: We present OMAnnotator, a novel approach to building a consensus annotation. OMAnnotator repurposes the OMA algorithm, originally designed to elucidate evolutionary relationships among genes across species, to integrate predictions from different annotation sources into a consensus annotation, using evolutionary information as a tie-breaker. During benchmarking on the Drosophila melanogaster reference, OMAnnotator's consensus improved upon its source annotations and two state-of-the-art pipelines used as annotation combiners with the same inputs. When applied to three recently published genomes, OMAnnotator gave substantial improvements in two cases, and mixed results in the third, which had already benefitted from extensive expert curation. This underlines the method's effectiveness and robustness for combining the results of disagreeing annotation softwares, strengthening the toolkit for eukaryotic genome annotation.

Availability and implementation: OMAnnotator is available on GitHub (https://github.com/DessimozLab/OMAnnotator).