Metabolomics Data Analysis Workflow: From Raw Data to Biological Insight

Metabolomics data analysis transforms raw instrument signals from mass spectrometry or nuclear magnetic resonance platforms into interpretable biological knowledge. This workflow covers the complete pipeline from data acquisition through preprocessing, statistical analysis, and pathway interpretation, with practical guidance for researchers, students, and analysts working with metabolomics datasets. The workflow applies to both targeted and untargeted experiments, with emphasis on quality control, reproducibility, and transparent reporting of analytical decisions.

Scope and Reader Context

This workflow serves researchers who have collected raw metabolomics data and need a structured approach to convert those files into publishable biological findings. The content addresses common decisions at each pipeline stage, including software selection, parameter choices, quality assessment, statistical modeling, and metabolite identification. Readers should have basic familiarity with chromatography, mass spectrometry, or NMR principles, but the workflow explains analytical choices in practical terms that support informed decision making.

The workflow covers liquid chromatography-mass spectrometry (LC-MS), gas chromatography-mass spectrometry (GC-MS), and NMR-based metabolomics. Each platform generates distinct data structures that require platform-specific preprocessing approaches. The workflow emphasizes untargeted analysis because it presents the greatest analytical challenges, while noting where targeted analysis follows a simpler path.

At a Glance

Data Inputs and Experimental Design Considerations

Raw Data Formats and Acquisition Parameters

Metabolomics instruments generate vendor-specific raw files that require conversion to open formats for cross-platform analysis. The mzML format serves as the standard interchange format for mass spectrometry data, enabling downstream tools to read files regardless of the acquisition instrument. NMR data typically exports as FID or processed spectra files that feed into dedicated NMR analysis pipelines.

Before beginning computational analysis, verify that acquisition parameters match the research question. Untargeted metabolomics with liquid chromatography coupled to high-resolution mass spectrometry enables measurement of a wide range of physicochemically diverse molecules, making it the most common approach for profiling studies. Hydrophilic interaction liquid chromatography (HILIC) is often applied before high-resolution accurate mass analysis when the study targets energy pathways associated with mitochondrial metabolism. The choice of chromatography and ionization mode directly affects which metabolite classes appear in the final feature table.

Batch Structure and Quality Control Samples

Experimental design determines the validity of downstream statistical analysis. Metabolomics experiments generate large, complex data files that are impractical to analyze manually, so automated pipelines are essential. However, automation cannot compensate for poor experimental design.

Include pooled quality control samples that represent the average composition of the study samples. These QC samples should be injected at regular intervals throughout the analytical run to monitor instrument drift, retention time shifts, and signal intensity changes. Blank samples identify background contamination and carryover. The injection order should be randomized to avoid confounding biological variation with analytical drift.

Sample Types and Preparation Effects

The biological matrix influences both data quality and interpretation. Plasma, urine, cerebrospinal fluid, tissue extracts, and cell culture media each present distinct challenges for metabolite extraction and chromatography. Sample preparation steps introduce variability that must be controlled through standardized protocols and appropriate replication.

Document all preparation steps, including extraction solvent composition, derivatization procedures, and internal standards added. Internal standards serve as critical references for normalization and quality assessment. The choice of internal standards should reflect the chemical diversity of the metabolites under investigation.

Preprocessing Workflow

Raw Data Conversion and Inspection

Convert vendor files to open formats using available conversion tools. After conversion, inspect total ion chromatograms across all samples to identify obvious acquisition problems, including signal saturation, baseline drift, and injection failures. Compare QC sample chromatograms across the run to detect temporal trends in signal intensity or retention time.

Retention time stability is a primary quality indicator. Large shifts indicate column degradation or gradient problems that will complicate peak alignment. Signal intensity trends across the injection sequence reveal ion source contamination or detector sensitivity changes. These inspections should occur before any computational processing because problematic runs may require reacquisition.

Peak Detection and Feature Extraction

Peak detection algorithms identify mass spectral features, defined as a specific mass-to-charge ratio at a specific retention time with associated intensity. Feature extraction is the fundamental step for current data analysis tools, and different methods produce different feature extraction results in practical applications. This variability can puzzle users when selecting adequate tools for their data.

Several advanced tools are available for ultrahigh performance liquid chromatography-high resolution mass spectrometry data analysis, including MS-DIAL, XCMS, MZmine, AntDAS, Progenesis QI, and Compound Discoverer. Evaluations using both mixtures of standards and complex plant matrices show that AntDAS provides the most acceptable feature extraction, compound identification, and quantification results in targeted compound analysis. For complex plant datasets, both MS-DIAL and AntDAS provide more reliable results than other tools.

Peak picking software performance varies substantially. In comparative studies examining reproducibility and output quality, MS-DIAL showed the best recovery of spiked analytes. Reproducibility across multiple projects was good among all tested software. The total number of features found was consistent for data-dependent acquisition and full scan acquisition in MS-DIAL, but full scan data led to considerably more features in MZmine and Progenesis QI.

Peak Alignment Across Samples

After feature detection, align peaks across samples so that the same metabolite appears as the same feature in every sample. Retention time shifts between runs require correction through alignment algorithms. Alignment quality directly affects the accuracy of the final feature table.

Feature linearity serves as a useful quality parameter for evaluating alignment and detection performance. In comparative studies, features detected by MS-DIAL and MZmine showed good linearity, while Progenesis QI produced large variation, especially in full scan data. Peak width also proved to be a powerful filtering criterion, revealing many features in MZmine and Progenesis QI with questionable peak width. Full scan data appears to produce a disproportionately higher number of short features.

Missing Value Handling

Feature tables typically contain missing values from peaks that fell below detection limits in some samples or from alignment failures. Missing data handling requires careful consideration because the approach affects downstream statistics. Common strategies include removing features with excessive missingness, imputing missing values from the minimum detected intensity, or applying more sophisticated imputation methods.

The choice of missing value threshold depends on the study design and the expected metabolite detection rate. Features missing in a large proportion of samples likely represent unreliable detections or contaminants. However, metabolites present at low concentrations in specific biological conditions may show condition-dependent missingness that carries biological information.

Normalization and Data Transformation

Normalization corrects for technical variation in total signal intensity between samples. Options include normalization to total ion current, normalization to internal standards, median normalization, and probabilistic quotient normalization. The appropriate method depends on the dominant sources of technical variation in the dataset.

Data transformation stabilizes variance and improves the performance of statistical models. Common transformations include log transformation, Pareto scaling, and autoscaling. The choice of transformation affects the relative influence of abundant versus rare metabolites in multivariate analyses.

Statistical Analysis

Univariate Methods

Univariate statistical tests assess each feature independently for differences between experimental groups. Common approaches include t-tests for two-group comparisons, analysis of variance for multi-group designs, and non-parametric alternatives such as the Kruskal-Wallis test when normality assumptions fail. The SECIMTools suite includes t-tests, analysis of variance, and Kruskal-Wallis non-parametric tests as building blocks for metabolomics workflows.

Multiple testing correction is mandatory when analyzing thousands of features simultaneously. The false discovery rate approach controls the expected proportion of false positives among features declared significant. Without correction, the number of false positives becomes unmanageable.

Multivariate Methods

Multivariate methods capture patterns across all features simultaneously and often reveal group separation that univariate tests miss. Principal component analysis provides an unsupervised overview of data structure, revealing outliers, batch effects, and natural clustering. Partial least squares discriminant analysis is a supervised method that maximizes separation between predefined groups.

The SECIMTools suite includes principal component analysis, partial least squares discriminant analysis, hierarchical cluster heatmaps, and modularity clustering for visualization and exploration. Advanced classification methods include random forest and support vector machines. Variable selection tools include least absolute shrinkage and selection operator (LASSO) and Elastic Net.

Machine Learning Approaches

Machine learning methods have become increasingly common in metabolomics for classification and biomarker discovery. Deep learning models trained on metabolomics data can achieve high predictive performance, but the important question is defining which factors contribute to the model predictive power. Explainable AI methods help address this question by identifying the features that drive model predictions.

In a study using Ultra Performance Liquid Chromatography Mass Spectrometry-based metabolomics data to predict Alzheimer disease, LASSO selected 21 metabolic biomarkers from 150 candidate metabolites. A deep learning model with two layers and 18 neurons achieved an accuracy of 0.881, an F1-score of 0.892, and an area under the curve of 0.873. Several metabolomic biomarkers involved in glucose and lipid metabolism, particularly bile acid metabolites, were associated with clinical biomarkers and cognitive assessments.

Explainable AI pipelines integrate multiple attribution methods to identify important features. The OmiXAI pipeline incorporates gradient-based techniques including Integrated Gradients, InputXGradients, Guided Backpropagation, and Deconvolution, as well as Saliency Maps and GNNExplainer for graph neural networks. In a case study of functional genomic element prediction, OmiXAI enabled feature engineering that reduced the critical feature set from almost 2,000 to just 50 features.

Model Validation and Overfitting Prevention

Any supervised model requires rigorous validation to assess generalizability. Split data into training and validation sets, with typical splits around 70 percent training and 30 percent validation. Cross-validation provides more robust performance estimates by repeatedly partitioning the data.

Overfitting occurs when a model learns noise specific to the training data instead of generalizable biological patterns. High performance on training data with substantially lower performance on validation data indicates overfitting. Feature selection must be performed within the cross-validation framework to avoid selection bias.

Metabolite Identification

Identification Confidence Levels

Biological interpretation of metabolomics data hinges on the ability to accurately identify metabolites. The range of confidence associated with identifications is often overlooked, yet it fundamentally determines the validity of biological conclusions. Untargeted metabolomics workflows exhibit many valuable attributes, but challenges inherent to the approach remain, particularly in the identification process of LC-MS-based studies in mammalian systems.

Identification confidence spans a continuum from putative annotation based on accurate mass alone to definitive confirmation using authentic standards with matched retention time and fragmentation spectra. Accurate mass matching provides only tentative identification because many metabolites share the same nominal mass. MS/MS spectral matching against reference libraries increases confidence but requires careful consideration of acquisition parameters. The gold standard remains comparison with authentic standards analyzed under identical conditions.

Spectral Library Searching

Reference spectral libraries enable automated matching of experimental MS/MS spectra against curated databases. Library matching scores reflect the similarity between experimental and reference spectra, but high scores do not guarantee correct identification. Isomeric compounds can produce similar fragmentation patterns, and library entries may contain errors.

The Human Metabolome Database and Kyoto Encyclopedia of Genes and Genomes serve as primary resources for metabolite information and pathway context. Automated searches of these databases can identify biomarkers and their associated biological pathways, as demonstrated in an integrated statistical pipeline combining mid-infrared spectroscopy and targeted metabolomics for colorectal cancer biomarker discovery.

Computational Annotation Tools

Several software tools provide automated annotation capabilities. MS-DIAL and AntDAS demonstrated more reliable compound identification than other tools in comparative evaluations. The maxiM/Ze application uses image recognition algorithms to process liquid chromatography-high resolution mass spectrometry data, implementing an automated sequential pipeline including mass detection, extracted ion chromatogram generation, peak alignment, and data visualization.

Validation of maxiM/Ze against Compound Discoverer 3.4 and MZmine 4.8.30 using eight replicate pooled plasma samples demonstrated competitive feature detection with 12,067 features, annotation of 219 unique compounds, and reproducibility with a median coefficient of variation of 35.8 percent across platforms. The image recognition approach converts extracted ion chromatograms into PNG images and applies image processing techniques including Canny edge detection, watershed segmentation, and Pearson correlation-based clustering to align peaks across samples with minimal user input.

Biological Interpretation

Pathway Enrichment Analysis

Once significant metabolites are identified, pathway enrichment analysis determines whether specific metabolic pathways are overrepresented among the significant features. This analysis requires mapping identified metabolites to pathway databases and comparing the observed enrichment against what would be expected by chance.

Pathway-level interpretation provides biological context that individual metabolite changes cannot convey. Coordinated changes in multiple metabolites within a pathway suggest pathway-level dysregulation instead of isolated biochemical perturbations. However, pathway analysis depends entirely on the quality and completeness of metabolite identification.

Literature-Based Contextualization

Natural language processing platforms help researchers efficiently search and digest the vast scientific literature relevant to identified metabolites. These tools provide literature-based contextualization of metabolic features that decreases the time and expert-level subject knowledge required during the prioritization, identification, and interpretation steps in the metabolomics data analysis pipeline.

Four workflow types combine metabolomics data with NLP-based literature searches. The first workflow creates a list of metabolites interesting for follow-up based on initial annotation and prioritization. The second workflow finds literature evidence of metabolite and metabolic pathway activity governing the biological condition at a systems biology level. The third workflow identifies candidate biomarkers. The fourth workflow looks for metabolic conditions or drug-repurposing targets that two diseases have in common. These procedures can be used in isolation or consecutively, depending on the research questions, and typically take one to four hours or more to complete depending on software processing time.

Multi-Omics Integration

Metabolomics data gains additional interpretive power when integrated with other omics layers. Cells are the fundamental units of biological systems and exhibit unique development trajectories and molecular features. Single-cell omics technologies have expanded to include single-cell genome, epigenome, proteome, and metabolome, while also integrating multiple omics data and incorporating additional information such as spatial transcriptomics and CRISPR screening.

Multi-omics integration requires careful attention to data harmonization and the different measurement scales across omics layers. Metabolites represent the downstream products of genomic, transcriptomic, and proteomic changes, making metabolomics potentially most predictive of phenotype. Upstream biological disruptions result in a series of metabolomic changes, and the metabolome holds a wealth of information thought to be most predictive of phenotype.

Reproducibility and Data Management

FAIR Data Principles

Applying good data management and FAIR data principles in research projects helps disentangle knowledge discovery, study result reproducibility, and data reuse in future studies. The FAIR principles for research data establish guidance for making data Findable, Accessible, Interoperable, and Reusable. FAIR principles for research software were recently proposed based on the original concepts.

FAIR Digital Objects enable discovery and reuse of Research Objects, including computational workflows, for both humans and machines. A multi-omics data analysis workflow implementing FAIR practices was developed as a modular pipeline in the Nextflow workflow manager, including containers with software dependencies. The workflow adhered to software development practices like version control, documentation, and licensing, and was described with rich semantic metadata, packaged as a Research Object Crate, and shared via WorkflowHub.

Containerization and Workflow Engines

Most computational proteomics and metabolomics tools are designed as single-tiered software applications where analytics tasks cannot be distributed, limiting the scalability and reproducibility of data analysis. The combination of software containers with workflow environments enables large-scale metabolomics and proteomics analysis. BioContainers provides containerized software packages, while Galaxy and Nextflow serve as popular workflow environments.

The PhenoMeNal platform provides an advanced solution for setting up Infrastructure-as-a-Service that brings workflow-oriented, interoperable metabolomics data analysis platforms into the cloud. PhenoMeNal integrates a wide array of existing open-source tools that are tested and packaged as Docker containers through continuous integration and deployed based on a Kubernetes orchestration framework. It provides standardized, automated, and published analysis workflows in the user interfaces Galaxy, Jupyter, Luigi, and Pachyderm.

The Workflow4Metabolomics 3.0 Galaxy online infrastructure enables researchers to create, run, share, publish, and reference LC-MS, FIA-MS, GC-MS, and NMR data analysis workflows. This infrastructure supports the full analysis lifecycle from workflow creation through publication and citation.

Data Repositories and Public Data Reuse

Public data repositories enable data sharing and reuse that accelerates scientific progress. MetaboLights serves as one of the leading metabolomics data repositories. Analysis of confirmed reuse of datasets from MetaboLights found 272 unique publications that reused at least one dataset. Reuse is dominated by method and tool development, with smaller contributions from secondary biological analysis and data integration or meta-analysis.

LC-MS datasets account for the majority of reuse, whereas NMR and GC-MS contribute at a lower level. Data reuse has increased over time with noticeable acceleration in recent years. At the dataset level, reuse follows a long-tail distribution where a small subset of datasets accounts for repeated reuse, mainly as community benchmarks. Reuse is under-detected when dataset identifiers are not consistently reported, highlighting the need for standardized dataset citation to improve traceability and recognition of reuse.

Quality Control and Common Failure Patterns

Quality Metrics and Filtering

Data quality and control parameters are becoming more important in metabolomics. Peak picking software output requires quality assessment before downstream analysis. Feature linearity serves as a good quality parameter for evaluating detection performance. Peak width provides a powerful filtering criterion that can reveal features of questionable quality.

Manual classification of true positive features in comparative studies showed that MS-DIAL performed significantly better in data-dependent acquisition data with 62 percent true positives compared to other software in either acquisition mode. Popular solutions MS-DIAL and MZmine perform well in targeted analysis of spiked analytes as well as in classic untargeted analysis.

Common Failure Patterns

Several recurring problems compromise metabolomics data analysis. Poor chromatographic stability produces retention time shifts that defeat alignment algorithms. Ion suppression from matrix components reduces signal for co-eluting metabolites. Batch effects introduce systematic variation that can masquerade as biological differences. Contaminants from plastics, solvents, and sample preparation materials appear as features that consume analytical effort and may be misinterpreted as biological signals.

Software parameter choices substantially affect results. Different feature extraction methods lead to different results in practical applications, which may puzzle users when selecting adequate data analysis tools. Default parameters rarely suit all datasets, and parameter optimization should be documented and reported.

Professional Escalation Criteria

Certain situations warrant consultation with specialized bioinformatics support or statisticians. These include persistent batch effects that resist normalization, poor reproducibility between technical replicates, unexpected feature counts that suggest detection problems, and difficulty achieving acceptable model validation performance. When metabolite identification confidence is low but biological conclusions depend on specific identifications, consider orthogonal confirmation through targeted analysis with authentic standards.

Tools Selection and Comparison

Open Source Versus Commercial Software

The choice between open source and commercial tools involves tradeoffs in usability, support, documentation, and feature sets. Open source tools including XCMS, MZmine, and MS-DIAL provide transparent algorithms and active development communities. Commercial tools such as Progenesis QI and Compound Discoverer offer polished interfaces and vendor support but require licensing fees.

The SECIMTools suite provides open access, easy to use analytic tools broadly accessible to the biological community. These Python applications are available both as standalone tools and wrapped for use in Galaxy, which promotes reproducibility by saving histories and enabling workflow sharing among scientists. The suite includes quality control metrics, visualization techniques, basic statistical analysis methods, advanced classification methods, and advanced variable selection tools.

Platform-Specific Considerations

NMR metabolomics requires different preprocessing approaches than mass spectrometry. A typical NMR spectrum with up to 64,000 data points of a complex biological fluid or an extract of cells and tissues consists of thousands of sharp signals mainly derived from small molecules. Advanced NMR spectroscopic methods are available for extracting information on high molecular weight compounds such as lipids or lipoproteins. Numerous data preprocessing, data reduction, and analysis methods have been developed and continue to evolve in the field of NMR metabolomics.

While NMR is less sensitive than mass spectrometry, NMR provides a wealth of complex and information-rich metabolite data. NMR data together with conventional statistics, modeling methods, and bioinformatics tools reveals biomarker and mechanistic information. Metabolic profiling involves the analysis of hundreds to thousands of molecules using mainly state-of-the-art mass spectrometry and NMR spectroscopy technologies.

Emerging Tools and Methods

New tools continue to address reproducibility challenges in metabolomics data analysis. Many current data-processing tools are time-consuming, require careful parameter selection, and depend heavily on user expertise, making reproducibility a challenge. The maxiM/Ze application addresses these challenges through automated image recognition-based processing with minimal user input.

Machine learning integration with metabolomics continues to advance. Medicinal plant-based drug discovery increasingly integrates machine learning, LC-MS-based metabolomics, and network pharmacology. Machine learning enables data-driven prioritization of candidate compounds and plant species beyond well-studied taxa, while metabolomics provides experimental validation through comprehensive chemical profiling and dereplication. Molecular docking and molecular dynamics further refine candidate selection by evaluating binding modes and stability, whereas network pharmacology offers systems-level insight into multitarget and multipathway effects.

Reporting Standards and Documentation

Methods Reporting

Transparent reporting of analytical methods enables reproducibility and proper interpretation. The methods section should describe sample preparation, chromatography conditions, mass spectrometry parameters, preprocessing software and versions, parameter settings, statistical methods, and identification confidence levels. Software versions matter because algorithm behavior changes between releases.

The protocol for untargeted metabolomic analysis from sample preparation to data processing describes each step in detail, from preparation of solvents for chromatography to selecting parameters during data processing. Following such protocols ensures consistency and supports method comparison across studies.

Data Availability Statements

Deposit raw data and processed feature tables in public repositories to enable data reuse and verification. Include dataset identifiers in publications to facilitate tracking of data reuse. Standardized dataset citation improves traceability and recognition of reuse, addressing the current problem where reuse is under-detected when dataset identifiers are not consistently reported.

Limitations and Interpretation Caveats

Metabolomics data analysis has inherent limitations that should be acknowledged in reporting. Untargeted metabolomics provides relative instead of absolute quantification. Coverage is limited to metabolites that ionize under the chosen conditions and extract with the chosen solvent. Identification confidence varies across features, and many features remain unidentified. Statistical associations do not establish causation, and pathway interpretations depend on incomplete knowledge of metabolic networks.

The field of metabolomics is still maturing. The community has leveraged proteomics experience when applicable and developed a range of sample preparation and instrument methodology along with myriad data processing and analysis approaches. Research focuses have shifted toward a fundamental understanding of the biology responsible for metabolomic changes.

Frequently Asked Questions

What is the difference between targeted and untargeted metabolomics data analysis?

Targeted metabolomics quantifies a predefined set of metabolites using authentic standards and optimized acquisition methods. The analysis pipeline is simpler because features are known in advance and calibration curves enable absolute quantification. Untargeted metabolomics aims to measure all detectable metabolites without prior selection, generating thousands of unknown features that require computational detection, alignment, and identification. Untargeted workflows are hypothesis generating and exhibit many valuable attributes, but challenges inherent to the approach remain, particularly in the identification process.

How many quality control samples should be included in a metabolomics experiment?

Quality control samples should be injected at regular intervals throughout the analytical run to monitor instrument drift, retention time shifts, and signal intensity changes. The exact number depends on total sample count and run length. More QC injections provide better drift characterization but consume instrument time and reduce throughput. Pooled QC samples representing the average composition of study samples are standard practice for untargeted metabolomics.

Which metabolomics data analysis software should I choose?

Software selection depends on your platform, research question, and technical expertise. Comparative evaluations show that MS-DIAL and AntDAS provide reliable feature extraction and compound identification for complex datasets. MZmine and XCMS are widely used open source options with active communities. Commercial tools such as Progenesis QI and Compound Discoverer offer polished interfaces but may show different performance characteristics. Consider running a test dataset through multiple tools to compare feature counts, reproducibility, and identification rates before committing to a primary pipeline.

How do I handle missing values in my metabolomics feature table?

Missing values arise from peaks below detection limits or alignment failures. Common strategies include removing features with excessive missingness, imputing missing values from minimum detected intensity, or applying more sophisticated imputation methods. The choice depends on the missingness pattern and the downstream analysis. Features missing in a large proportion of samples likely represent unreliable detections. Condition-dependent missingness may carry biological information and should not be removed without consideration.

What confidence level should I assign to metabolite identifications?

Identification confidence spans a continuum from putative annotation based on accurate mass alone to definitive confirmation using authentic standards with matched retention time and fragmentation spectra. Accurate mass matching provides only tentative identification because many metabolites share the same nominal mass. MS/MS spectral matching increases confidence but requires careful consideration of acquisition parameters. The gold standard remains comparison with authentic standards analyzed under identical conditions. Report confidence levels explicitly in publications.

How do I prevent overfitting in machine learning models for metabolomics?

Split data into training and validation sets, with typical splits around 70 percent training and 30 percent validation. Cross-validation provides more robust performance estimates by repeatedly partitioning the data. Feature selection must be performed within the cross-validation framework to avoid selection bias. High performance on training data with substantially lower performance on validation data indicates overfitting. Report validation performance instead of training performance.

What are the FAIR principles and why do they matter for metabolomics?

The FAIR principles establish guidance for making research data Findable, Accessible, Interoperable, and Reusable. Applying these principles helps disentangle knowledge discovery, study result reproducibility, and data reuse in future studies. For metabolomics, FAIR practices include depositing raw data in public repositories, using open data formats, documenting analysis workflows, and providing rich metadata. FAIR Digital Objects enable discovery and reuse of computational workflows for both humans and machines.

How can I integrate metabolomics data with other omics data?

Multi-omics integration requires careful attention to data harmonization and the different measurement scales across omics layers. Metabolites represent the downstream products of genomic, transcriptomic, and proteomic changes, making metabolomics potentially most predictive of phenotype. Integration approaches range from simple correlation analyses to sophisticated machine learning models that combine multiple omics layers. Workflow managers such as Nextflow support modular pipelines that incorporate multiple omics data types with containerized software dependencies.

Related Bioinformatics Guides

• The KEGG Database and Pathway Analysis
• Alternative Splicing Analysis from RNA-Seq Data
• Hidden Markov Models in Biological Sequence Analysis
• The Reactome Pathway Knowledgebase
• MicroRNA Target Prediction Tools

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Single-cell omics: experimental workflow, data analyses and applications.. Science China. Life sciences, 2025.
• Untargeted Metabolomics Strategies-Challenges and Emerging Directions.. Journal of the American Society for Mass Spectrometry, 2016.
• Analysis of NMR Metabolomics Data.. Methods in molecular biology (Clifton, N.J.), 2020.
• A Protocol for Untargeted Metabolomic Analysis: From Sample Preparation to Data Processing.. Methods in molecular biology (Clifton, N.J.), 2021.
• Cognitive analysis of metabolomics data for systems biology.. Nature protocols, 2021.
• A multi-omics data analysis workflow packaged as a FAIR Digital Object.. GigaScience, 2024.
• Scalable Data Analysis in Proteomics and Metabolomics Using BioContainers and Workflows Engines.. Proteomics, 2020.
• PhenoMeNal: processing and analysis of metabolomics data in the cloud.. GigaScience, 2019.
• Fecal Extracellular Vesicle Metabolomics as a Non-Invasive Biomarker Source in Colorectal Cancer: TPOT AutoML Superiority over Tree-Based Models with SHAP and LIME Clinical Interpretability. 2026.
• maxiM/Ze: An Image Recognition Approach for Visualizing and Processing Mass Spectrometry Based Metabolomics Data. 2026.
• Modernising antiviral drug discovery: harnessing medicinal plants through machine learning and metabolomics to target the SARS-CoV-2 main protease.. 2026.
• Integrative analysis of circulating biomarkers of colorectal cancer combining mid-infrared spectroscopy and targeted metabolomics.. 2026.
• OmiXAI: An ensemble XAI pipeline for interpretable deep learning in omics data.. 2026.
• Revealing the Hidden Landscape of Public Metabolomics Data Reuse in MetaboLights. 2026.
• A comparison of feature extraction capabilities of advanced UHPLC-HRMS data analysis tools in plant metabolomics.. Analytica Chimica Acta, 2023.
• SECIMTools: a suite of metabolomics data analysis tools. BMC Bioinformatics, 2018.
• An Overview of Metabolomics Data Analysis: Current Tools and Future Perspectives. 2018.
• Impact of three different peak picking software tools on the quality of untargeted metabolomics data.. Journal of Pharmaceutical and Biomedical Analysis, 2024.
• Deep learning analysis of UPLC-MS/MS-based metabolomics data to predict Alzheimer's disease.. Journal of Neurological Sciences, 2023.
• Create, run, share, publish, and reference your LC-MS, FIA-MS, GC-MS, and NMR data analysis workflows with the Workflow4Metabolomics 3.0 Galaxy online infrastructure for metabolomics. International Journal of Biochemistry and Cell Biology, 2017.
• Mass-spectrometry based metabolomics: an overview of workflows, strategies, data analysis and applications. Proteome Science, 2025.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.