Biodata Mining最新文献

Background: Publicly available metagenomics datasets are crucial for ensuring the reproducibility of scientific findings and supporting contemporary large-scale studies. However, submitting a comprehensive metagenomics dataset is both cumbersome and time-consuming. It requires including sample information, sequencing reads, assemblies, binned contigs, metagenome-assembled genomes (MAGs), and appropriate metadata. As a result, metagenomics studies are often published with incomplete datasets or, in some cases, without any data at all. subMG addresses this challenge by simplifying and automating the data submission process, thereby encouraging broader and more consistent data sharing.

Results: subMG streamlines the process of submitting metagenomics study results to the European Nucleotide Archive (ENA) by allowing researchers to input files and metadata from their studies in a single form and automating downstream tasks that otherwise require extensive manual effort and expertise. The tool comes with comprehensive documentation as well as example data tailored for different use cases and can be operated via the command-line or a graphical user interface (GUI), making it easily deployable to a wide range of potential users.

Conclusions: By simplifying the submission of genome-resolved metagenomics study datasets, subMG significantly reduces the time, effort, and expertise required from researchers, thus paving the way for more numerous and comprehensive data submissions in the future. An increased availability of well-documented and FAIR data can benefit future research, particularly in meta-analyses and comparative studies.

Network representations of data are designed to encode relationships between concepts as sets of edges between nodes. Human biology is inherently complex and is represented by data that often exists in a hierarchical nature. One canonical example is the relationship that exists within and between various -omics datasets, including genomics, transcriptomics, and proteomics, among others. Encoding such data in a network-based or graph-based representation allows the explicit incorporation of such relationships into various biomedical big data tasks, including (but not limited to) disease subtyping, interaction prediction, biomarker identification, and patient classification. This review will present various existing approaches in using network representations and analysis of data in multiomics in the framework of deep learning and machine learning approaches, subdivided into supervised and unsupervised approaches, to identify benefits and drawbacks of various approaches as well as the possible next steps for the field.

Background: Over the past decade an increase in usage of electronic health data (EHR) by office-based physicians and hospitals has been reported. However, these data types come with challenge regarding completeness and data quality and it is, especially for more complex models, unclear how these characteristics influence the performance.

Methods: In this paper, we focus on joint models which combines longitudinal modelling with survival modelling to incorporate all available information. The aim of this paper is to establish simulation-based guidelines for the necessary quality of longitudinal EHR data so that joint models perform better than cox models. We conducted an extensive simulation study by systematically and transparently varying different characteristics of data quality, e.g., measurement frequency, noise, and heterogeneity between patients. We apply the joint models and evaluate their performance relative to traditional Cox survival modelling techniques.

Results: Key findings suggest that biomarker changes before disease onset must be consistent within similar patient groups. With increasing noise and a higher measurement density, the joint model surpasses the traditional Cox regression model in terms of model performance. We illustrate the usefulness and limitations of the guidelines with two real-world examples, namely the influence of serum bilirubin on primary biliary liver cirrhosis and the influence of the estimated glomerular filtration rate on chronic kidney disease.

Genetic toxicology is crucial for evaluating the potential risks of chemicals and drugs to human health and the environment. The emergence of high-throughput technologies has transformed this field, providing more efficient, cost-effective, and ethically sound methods for genotoxicity testing. It utilizes advanced screening techniques, including automated in vitro assays and computational models to rapidly assess the genotoxic potential of thousands of compounds simultaneously. This review explores the transformation of traditional in vitro and in vivo methods into computational models for genotoxicity assessment. By leveraging advances in machine learning, artificial intelligence, and high-throughput screening, computational approaches are increasingly replacing conventional methods. Coupling conventional screening with artificial intelligence (AI) and machine learning (ML) models has significantly enhanced their predictive capabilities, enabling the identification of genotoxicity signatures tied to molecular structures and biological pathways. Regulatory agencies increasingly support such methodologies as humane alternatives to traditional animal models, provided they are validated and exhibit strong predictive power. Standardization efforts, including the establishment of common endpoints across testing approaches, are pivotal for enhancing comparability and fostering consensus in toxicological assessments. Initiatives like ToxCast exemplify the successful incorporation of HTS data into regulatory decision-making, demonstrating that well-interpreted in vitro results can align with in vivo outcomes. Innovations in testing methodologies, global data sharing, and real-time monitoring continue to refine the precision and personalization of risk assessments, promising a transformative impact on safety evaluations and regulatory frameworks.

Acute myeloid leukemia (AML) is caused by proliferation of mutated myeloid progenitor cells. The standard chemotherapy regimen does not efficiently cause remission as there is a high relapse rate. Resistance acquired by leukemic stem cells is suggested to be one of the root causes of relapse. Therefore, there is an urgency to develop new drugs for therapy. Repurposing approved drugs for AML can provide a cost-friendly, time-efficient, and affordable alternative. The multiscale interactome network is a computational tool that can identify potential therapeutic candidates by comparing mechanisms of the drug and disease. Communities that could be potentially experimentally validated are detected in the multiscale interactome network using the algorithm CRank. The results are evaluated through literature search and Gene Ontology (GO) enrichment analysis. In this research, we identify therapeutic candidates for AML and their mechanisms from the interactome, and isolate prioritized communities that are dominant in the therapeutic mechanism that could potentially be used as a prompt for pre-clinical/translational research (e.g. bioinformatics, laboratory research) to focus on biological functions and mechanisms that are associated with the disease and drug. This method may allow for an efficient and accelerated discovery of potential candidates for AML, a rapidly progressing disease.

Background: Functional deterioration (FD) of various organ systems is the major cause of death in ICU patients, but few studies propose effective multi-task (MT) model to predict FD of multiple organs simultaneously. This study propose a MT deep learning model named inter-organ correlation based multi-task model (IOC-MT), to dynamically predict FD in six organ systems.

Methods: Three public ICU databases were used for model training and validation. The IOC-MT was designed based on the routine MT deep learning framework, but it used a Graph Attention Networks (GAT) module to capture inter-organ correlation and an adaptive adjustment mechanism (AAM) to adjust prediction. We compared the IOC-MT to five single-task (ST) baseline models, including three deep models (LSTM-ST, GRU-ST, Transformer-ST) and two machine learning models (GRU-ST, RF-ST), and performed ablation study to assess the contribution of important components in IOC-MT. Model discrimination was evaluated by AUROC and AUPRC, and model calibration was assessed by the calibration curve. The attention weight and adjustment coefficient were analyzed at both overall and individual level to show the AAM of IOC-MT.

Results: The IOC-MT had comparable discrimination and calibration to LSTM-ST, GRU-ST and Transformer-ST for most organs under different gap windows in the internal and external validation, and obviously outperformed GRU-ST, RF-ST. The ablation study showed that the GAT, AAM and missing indicator could improve the overall performance of the model. Furthermore, the inter-organ correlation and prediction adjustment of IOC-MT were intuitive and comprehensible, and also had biological plausibility.

Conclusions: The IOC-MT is a promising MT model for dynamically predicting FD in six organ systems. It can capture inter-organ correlation and adjust the prediction for one organ based on aggregated information from the other organs.

Digital twins in healthcare offer an innovative approach to precision diagnosis, prognosis, and treatment. SynTwin, a novel computational methodology to generate digital twins using synthetic data and network science, has previously shown promise for improving prediction of breast cancer mortality. In this study, we validate SynTwin using population-level data for different cancer types from the Surveillance, Epidemiology, and End Results (SEER) program from the National Cancer Institute (USA). We assess its predictive accuracy across cancer types of varying sample sizes (n = 1,000 to 30,000 records), mortality rates (35% to 60%), and study designs, revealing insights into the strengths and limitations of digital twins derived from synthetic data in mortality prediction. We also evaluate the effect of sample size (n = 1,000 to 70,000 records) on predictive accuracy for selected cancers (non-Hodgkin lymphoma, bladder, and colorectal cancers). Our results indicate that for larger datasets (n > 10,000) including digital twins in the nearest network neighbor prediction model significantly improves the performance compared to using real patients alone. Specifically, AUROCs ranged from 0.828 to 0.884 for cancers such as cervix uteri and ovarian cancer with digital twins, compared to 0.720 to 0.858 when using real patient data. Similarly, among the selected three cancers, AUROCs using digital twins exceeded AUROCs using real patients alone by at least 0.06 with narrowing variance in performance as the sample size increased. These results highlight the benefit of network-based digital twins, while emphasizing the importance of considering effective sample size when developing predictive models like SynTwin.