Contents
On this page
- The three workflows supported by NetworkAnalyst
- Limitations of raw data pooling and the function of batch-effect correction
- Translation of a combined gene list into a protein-interaction network
- Normalization choices and the validity of a meta-analysis
- Circumstances in which this type of tool is appropriate
Academy · Research Data, Analytics & AI · NetworkAnalyst Framework
NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks
NetworkAnalyst is a free, browser-based platform that enables bench biologists to combine several gene expression datasets, conduct a meta-analysis across them and map the result onto a protein-protein interaction network without writing code.
NetworkAnalyst is a web-based platform for the statistical, visual and network-based meta-analysis of gene expression data. Rather than requiring the installation of specialist software or the writing of R scripts, it accepts raw or pre-processed expression matrices through a browser, guides the user through normalisation and batch-effect correction, and produces publication-ready heatmaps, volcano plots and protein-protein interaction (PPI) networks from the combined result. Its distinguishing feature is that it was designed specifically to make cross-study meta-analysis of gene expression, a statistically demanding task when performed manually, accessible to researchers without formal bioinformatics training.
Learning objectives
- The functions of NetworkAnalyst across its three core workflows, and the place of meta-analysis among them
- The reasons that combining raw expression values across studies produces misleading results, and the approach the platform adopts instead
- How a combined gene list is translated into a protein-protein interaction network for biological interpretation
Common misconceptions
- “You can just pool the raw expression values from several studies and average them.” Batch effects (systematic, non-biological differences between datasets arising from differences in platform, laboratory or processing) frequently exceed the biological signal of interest, which is why both the original empirical Bayes batch-correction method and the NetworkAnalyst pipeline treat batch-effect adjustment as a mandatory step before any cross-study comparison rather than an optional refinement.
The three workflows supported by NetworkAnalyst
NetworkAnalyst is organised around three distinct but related workflows, and identifying the workflow that a task requires is the first practical decision for the user. The first is network analysis and visualisation beginning from a single gene or protein list, which is useful when a researcher already has a shortlist of candidate genes and wishes to examine how they interact. The second is meta-analysis that accounts for multiple experimental metadata parameters within one or more datasets, allowing covariates such as treatment, time point or tissue type to be modelled explicitly. The third, and the most directly relevant to the combination of independent studies, is meta-analysis across multiple separate gene expression datasets1. A later expansion of the platform, NetworkAnalyst 3.0, extended this third workflow with additional normalisation options and direct integration with major public repositories, enabling users to import comparable public datasets without reformatting each one manually2. The protocol paper describing the original workflow estimated that a complete meta-analysis, from the upload of raw data to a finished network figure, requires approximately 1.5 hours for a user following the documented steps; this estimate is notable because it implies that the platform performs most of the statistically technical work automatically1. Identifying the appropriate workflow is also important because each expects a different starting input: the gene-list workflow assumes that candidate genes have already been identified by other means, the metadata-aware workflow assumes a single dataset with richly annotated covariates, and the cross-dataset meta-analysis workflow assumes several independently collected datasets addressing a comparable biological question. Selection of the wrong entry point is therefore a common early error, which typically wastes time rather than producing an incorrect result.
Limitations of raw data pooling and the function of batch-effect correction
A dedicated meta-analysis tool is required, rather than simple concatenation of expression matrices from different studies, because gene expression datasets generated on different platforms, in different laboratories or even in different batches within the same laboratory carry systematic technical variation unrelated to biology. This is the batch-effect problem, which was formally addressed by an empirical Bayes method, commonly known by the name of its implementation, ComBat, that models and removes location and scale differences between batches while preserving the biological signal of interest3. A widely cited review of the reuse of public gene expression data found that failure to account for such technical variation is one of the most common reasons that secondary analyses of public expression data produce unreliable or contradictory results4. Methodological guidance specific to microarray meta-analysis goes further, setting out a structured checklist of decisions (platform matching, probe-to-gene mapping, cross-platform normalisation, and choice of effect-size or p-value combination method) that must be made correctly before a meta-analysis is statistically defensible5. A separate methodological survey of combination strategies catalogued the range of statistical approaches in use across the field, emphasising that no single method is universally correct and that the choice must be matched to the comparability of the underlying studies6.
Translation of a combined gene list into a protein-interaction network
Once a meta-analysis has produced a consensus list of differentially expressed genes, the biological question shifts from “which genes changed” to “how do these genes relate to each other functionally.” NetworkAnalyst addresses this by mapping the gene list onto a protein-protein interaction network, drawing on curated interaction databases, and by providing integrative visual exploration tools for the resulting network7. The conceptual justification for this step derives from network biology more broadly: cellular components do not act as independent units, and a trait or disease phenotype typically emerges from a connected module of interacting genes and proteins rather than from any single gene in isolation8. The interaction data underlying these networks commonly derive from resources such as the STRING database, which aggregates physical and functional protein associations from experimental evidence, curated databases and computational prediction across thousands of organisms9. For visualisation, NetworkAnalyst follows the same general design principles as Cytoscape, the widely used open-source platform for integrating biomolecular interaction networks with high-throughput expression data, allowing users to rearrange, filter and colour a network to highlight the connections most relevant to a given hypothesis10.
Normalization choices and the validity of a meta-analysis
Batch correction addresses differences between datasets; within each individual dataset, however, a separate normalisation step is required to make expression values comparable across samples and probes, and the choice of method has documented consequences for downstream meta-analysis. For microarray data, Rafael Irizarry and colleagues developed the Robust Multi-array Average (RMA) method specifically to correct for background noise and to summarise probe-level intensities into a single, comparable expression value per gene per sample; it remains one of the most widely used normalisation approaches because it was shown to reduce variance without introducing the systematic bias that simpler averaging methods can produce11. For RNA-sequencing (RNA-seq) data, which measure expression differently and are increasingly combined with legacy microarray datasets in meta-analyses, Mark Robinson and Alicia Oshlack developed the trimmed mean of M-values (TMM) normalisation method specifically to correct for the fact that total sequencing depth and library composition can otherwise make raw RNA-seq counts misleading when compared directly across samples12. A meta-analysis that combines microarray and RNA-seq datasets without first applying the normalisation appropriate to each platform is one of the less conspicuous sources of error in this type of analysis, since the error is not evident in the way that a missing file or a failed script would be; it produces a plausible result that reflects platform artefacts rather than biology.
Circumstances in which this type of tool is appropriate
A browser-based meta-analysis and network platform such as NetworkAnalyst is best suited to a specific situation: a researcher has, or can locate, several independent gene expression datasets addressing a related biological question and requires a statistically defensible combined signal, together with a means of interpreting that signal functionally, without building a custom bioinformatics pipeline. It is less suited to highly customised analyses that require a statistical model not built into the platform, or to raw sequencing data that still require alignment and quantification before expression values exist. Used within its intended scope, with normalisation and batch correction applied as mandatory steps, it converts what would otherwise be a programming task of several weeks into a workflow that a biologically trained researcher can complete and interpret directly. The limitations of the platform should also be stated explicitly: it cannot substitute for domain knowledge of the biological system under study, and a statistically sound meta-analysis applied to a poorly matched set of studies, for instance datasets from different tissue types or disease stages combined without acknowledgement of that difference, will still produce a misleading result, however correctly the normalisation and batch-correction steps were executed. The guided workflow reduces the statistical and computational burden of cross-study gene expression analysis, but the scientific judgement as to which datasets belong together in the same meta-analysis remains, and should remain, the responsibility of the researcher rather than the software.
Key points
Key points
A cross-study gene expression meta-analysis is only as trustworthy as its batch-effect correction and normalization; NetworkAnalyst's value is bundling those statistically demanding steps into a guided workflow that still ends in an interpretable protein-interaction network.
Practical application
Practical application
- Consistent annotation of every dataset to be combined should be confirmed. Platform, tissue, treatment and time-point metadata must be captured correctly before any meta-analysis step is run, because the correction methods depend on identifying which samples belong to which batch.
- Batch-effect correction should be treated as mandatory rather than optional. An empirical Bayes correction or an equivalent method should be applied before any cross-dataset result is interpreted, and the before-and-after plots generated by the platform should be inspected to confirm that the correction performed as expected.
- The resulting network should be interpreted as a hypothesis rather than a conclusion. A PPI network constructed from meta-analysed expression data identifies candidate functional modules that merit experimental validation; it does not in itself demonstrate a causal biological mechanism.
Frequently asked questions
No. It is designed as a browser-based graphical workflow specifically so that researchers without bioinformatics training can conduct a cross-study meta-analysis and generate a network visualisation directly.
The methodological guidance for such combinations emphasises careful cross-platform normalisation and probe-to-gene mapping beforehand, since the two technologies measure expression differently; combination is possible but requires deliberate handling rather than simple pooling.
Systematic technical differences between datasets can produce a result that resembles a strong biological signal but is in fact an artefact of the laboratory or platform that generated the samples; this is the central problem that the empirical Bayes correction method was developed to address.
No. The network is constructed from curated public interaction databases that aggregate prior experimental and computational evidence; genes or edges that appear important in the network still require targeted follow-up experiments to confirm their role in the biology under study.
The protocol describing the original workflow reports approximately 1.5 hours from raw data upload to a finished network figure, although this depends on dataset size and on the amount of manual metadata cleaning required beforehand.
References
- Xia J, Gill EE, Hancock REW. NetworkAnalyst for statistical, visual and network-based meta-analysis of gene expression data. Nature Protocols. 2015;10(6):823-844. read 10 October 2026. View on Nature Protocols
- Zhou G, Soufan O, Ewald J, Hancock REW, Basu N, Xia J. NetworkAnalyst 3.0: a visual analytics platform for comprehensive gene expression profiling and meta-analysis. Nucleic Acids Research. 2019;47(W1):W234-W241. read 10 October 2026. View on Nucleic Acids Research
- Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics. 2007;8(1):118-127. read 10 October 2026. View on Biostatistics
- Rung J, Brazma A. Reuse of public genome-wide gene expression data. Nature Reviews Genetics. 2013;14(2):89-99. read 10 October 2026. View on Nature Reviews Genetics
- Ramasamy A, Mondry A, Holmes CC, Altman DG. Key issues in conducting a meta-analysis of gene expression microarray datasets. PLOS Medicine. 2008;5(9):e184. read 10 October 2026. View on PLOS Medicine
- Tseng GC, Ghosh D, Feingold E. Comprehensive literature review and statistical considerations for microarray meta-analysis studies. Nucleic Acids Research. 2012;40(9):3785-3799. read 10 October 2026. View on Nucleic Acids Research
- Xia J, Benner MJ, Hancock REW. NetworkAnalyst - integrative approaches for protein-protein interaction network analysis and visual exploration. Nucleic Acids Research. 2014;42(W1):W167-W174. read 10 October 2026. View on Nucleic Acids Research
- Barabási AL, Oltvai ZN. Network biology: understanding the cell's functional organization. Nature Reviews Genetics. 2004;5(2):101-113. read 10 October 2026. View on Nature Reviews Genetics
- Szklarczyk D, Gable AL, Lyon D, et al. STRING v11: protein-protein association networks with increased coverage. Nucleic Acids Research. 2019;47(D1):D607-D613. read 10 October 2026. View on Nucleic Acids Research
- Shannon P, Markiel A, Ozier O, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Research. 2003;13(11):2498-2504. read 10 October 2026. View on Genome Research
- Irizarry RA, Hobbs B, Collin F, et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 2003;4(2):249-264. read 10 October 2026. View on Biostatistics
- Robinson MD, Oshlack A. A scaling normalization method for differential expression analysis of RNA-seq data. Genome Biology. 2010;11:R25. read 10 October 2026. View on Genome Biology
Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.
How to cite this article
Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.
Pubrica Academy. (2026, October 10). NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks. Pubrica. https://pubrica.com/academy/concepts-definitions/network-analyst-framework/
Pubrica Academy. “NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks.” Pubrica, 10 Oct. 2026, pubrica.com/academy/concepts-definitions/network-analyst-framework/.
Pubrica Academy (2026) NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks. Pubrica. Available at: https://pubrica.com/academy/concepts-definitions/network-analyst-framework/ (Accessed: 15 October 2026).
Pubrica Academy. “NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks.” Pubrica. October 10, 2026. https://pubrica.com/academy/concepts-definitions/network-analyst-framework/
Pubrica Academy, ‘NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks’ (Pubrica, 10 October 2026) <https://pubrica.com/academy/concepts-definitions/network-analyst-framework/> accessed 15 October 2026.
Pubrica Academy. NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks. [Internet]. Pubrica; 2026 Oct 10 [cited 2026 Oct 15]. Available from: https://pubrica.com/academy/concepts-definitions/network-analyst-framework/
Pubrica Academy, “NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks,” Pubrica, Oct. 10, 2026. [Online]. Available: https://pubrica.com/academy/concepts-definitions/network-analyst-framework/
@misc{/la_networkanalyst_gene_expression_meta_analysis_2026,
author = {{Pubrica Academy}},
title = {NetworkAnalyst: A Framework for Meta-Analyzing Gene Expression Networks},
howpublished = {Pubrica},
year = {2026},
month = {oct},
url = {https://pubrica.com/academy/concepts-definitions/network-analyst-framework/},
note = {Accessed: <span data-cite-acc="iso">2026-10-15</span>}
}This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.
Gene Expression Meta-Analysis Readiness Checklist
For use before combining gene expression datasets in NetworkAnalyst or a comparable platform.
[ ] Collected consistent metadata (platform, tissue, treatment, time point) for every dataset [ ] Confirmed probe-to-gene or transcript-to-gene mapping is consistent across platforms [ ] Applied an empirical Bayes or equivalent batch-effect correction before comparing datasets [ ] Inspected before/after correction plots to confirm batch effects were actually removed [ ] Chosen a combination method (effect size vs. p-value based) appropriate to how similar the studies are [ ] Treated the resulting PPI network as a hypothesis-generating output requiring experimental follow-up
Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.
How Pubrica can help
Pubrica’s data analytics team assists research groups in assembling comparable public and in-house expression datasets, applying the appropriate batch-effect correction and interpreting the resulting network output before its inclusion in a manuscript or grant report.