Volcano Plot Proteomics: How to Create and Interpret Them Effectively

Volcano plots are a standard visualization for differential expression analysis in proteomics, plotting statistical significance against fold change for every quantified protein. This article provides a practical protocol for generating volcano plots from mass spectrometry data, selecting appropriate thresholds, handling multiple testing corrections, and interpreting results without overstating biological conclusions. The guidance applies to label-free quantification, isotopic labeling workflows, and cross-omics integration, with emphasis on reproducible reporting and common analytical pitfalls.

At a Glance

Understanding Volcano Plot Structure

A volcano plot displays each quantified protein as a single point on a scatter plot. The x-axis represents the magnitude of change between two conditions, typically expressed as log2 fold change. The y-axis represents statistical significance, usually the negative log10 of the p-value. Proteins with large fold changes and high statistical significance appear in the upper left and upper right corners of the plot, forming the characteristic two arms that give the volcano plot its name.

The plot serves a screening function instead of a confirmatory one. It allows researchers to quickly identify candidate proteins that merit further investigation, but it does not by itself establish biological relevance. A protein can appear significant in a volcano plot yet have no functional connection to the condition being studied, particularly when the underlying data contain technical artifacts or when the statistical model fails to account for important sources of variation.

The visual appeal of volcano plots has made them ubiquitous in proteomics publications, but their interpretive value depends entirely on the quality of the input data and the appropriateness of the statistical thresholds applied. A well-constructed volcano plot from poor data remains poor data. The plot is a communication tool, not an analytical method in itself.

Data Inputs for Volcano Plot Generation

Quantification Matrices

The starting point for any volcano plot is a protein expression matrix. This matrix contains rows for each identified protein and columns for each sample, with values representing abundance measurements. The quantification method determines the nature of these values. Label-free approaches such as MaxLFQ or spectral counting produce intensity values that require normalization before analysis. Isotopic labeling approaches, including trypsin-catalyzed 18O labeling, generate ratios between labeled and unlabeled peptides that need correction for labeling efficiency before they can be interpreted as true expression changes.

The choice of quantification approach affects downstream analysis in ways that are not always obvious. Spectral counts and intensity-based measurements have different statistical distributions and different sensitivities to missing values. An extensive evaluation of differential expression workflows, involving tens of thousands of combinatorial experiments on spike-in datasets, found that no single workflow performs optimally across all data types. The researchers demonstrated that ensemble inference, which integrates results from multiple top-performing workflows, can expand differential proteome coverage and resolve inconsistencies between individual approaches.

Experimental Design Considerations

Volcano plots require at least two groups for comparison. The simplest design compares a treatment condition against a control, with multiple biological replicates in each group. The number of replicates directly influences the reliability of the statistical tests that feed into the volcano plot. With only two or three replicates per group, the variance estimates for each protein are unstable, and the resulting p-values can be misleading.

Technical replicates do not substitute for biological replicates. Technical variation reflects measurement noise, while biological variation captures the actual differences between individuals or samples. A volcano plot generated from technical replicates alone will show artificially tight variance and may flag proteins as significant when the observed differences simply reflect batch effects or sample preparation artifacts.

Normalization and Transformation

Raw intensity values in proteomics span several orders of magnitude and are heavily right-skewed. Log2 transformation addresses this skew and makes the fold change scale symmetric, so that a two-fold increase and a two-fold decrease appear equidistant from zero. Most differential expression tools apply this transformation internally, but researchers generating volcano plots from custom pipelines must apply it explicitly.

Normalization adjusts for systematic differences in total protein amount, digestion efficiency, or instrument sensitivity between samples. Common approaches include median centering, quantile normalization, and variance stabilizing normalization. The choice of normalization method can change which proteins appear significant in the volcano plot, so the method should be documented and justified in any report.

Statistical Foundations for Volcano Plots

Choosing the Statistical Test

The statistical test used to generate p-values for the volcano plot must match the data structure. For experiments with multiple replicates per condition, moderated t-tests are generally preferred over ordinary t-tests because they borrow information across proteins to stabilize variance estimates. This is particularly important in proteomics, where the number of replicates is often small and the number of proteins measured is large.

The evaluation of differential expression workflows mentioned earlier found that the choice of statistical test is one of the most consequential decisions in the entire pipeline. Tests that assume equal variance across all proteins perform poorly when the actual variance structure is heterogeneous. Tests that are too conservative reduce sensitivity and may miss genuine biological changes.

Multiple Testing Correction

When thousands of proteins are tested simultaneously, the probability of false positives increases dramatically. A p-value threshold of 0.05 applied to 5,000 proteins would be expected to produce 250 false positives by chance alone. Multiple testing correction controls this error rate.

The Benjamini-Hochberg procedure controls the false discovery rate, which is the expected proportion of false positives among all proteins declared significant. This is generally preferred over more conservative methods such as the Bonferroni correction, which controls the family-wise error rate and can be overly restrictive for exploratory proteomics studies. The false discovery rate approach allows researchers to accept a small proportion of false positives in exchange for greater sensitivity, which is appropriate for hypothesis-generating analyses.

The corrected p-values, often called q-values, should be used for the y-axis of the volcano plot. Some software packages plot the raw p-values and apply the correction only when determining which proteins are colored as significant. This practice can be misleading because the visual impression of significance does not match the statistical reality.

Setting Thresholds for Significance

Fold Change Thresholds

The fold change threshold determines the minimum magnitude of change required for a protein to be considered differentially expressed. A common default is a two-fold change, corresponding to a log2 fold change of 1. This threshold is arbitrary and should be adjusted based on the biological context and the technical variability of the assay.

For datasets with low technical variability, smaller fold changes may be biologically meaningful. For datasets with high variability, even two-fold changes may not be reproducible. One approach is to examine the distribution of fold changes across all proteins and set the threshold at a level that captures a reasonable number of candidates without being so permissive that the list becomes unmanageable.

P-value Thresholds

The p-value threshold for the volcano plot should reflect the multiple testing correction applied. A common choice is a false discovery rate of 0.05, meaning that 5% of the proteins declared significant are expected to be false positives. More stringent thresholds, such as 0.01, reduce the false positive rate but also reduce sensitivity.

The combination of fold change and p-value thresholds defines the four quadrants of the volcano plot. Proteins in the upper left and upper right quadrants pass both thresholds and are considered differentially expressed. Proteins in the lower quadrants may have large fold changes but lack statistical significance, often because of high variability between replicates. Proteins in the upper center may be statistically significant but have small fold changes that may not be biologically meaningful.

Threshold Selection as a Decision

Threshold selection should be documented as an analytical decision with justification. The thresholds chosen for one experiment may not be appropriate for another, even within the same laboratory. Factors that influence threshold selection include the sample size, the expected magnitude of biological effects, the technical reproducibility of the assay, and the downstream validation resources available.

Software Tools for Volcano Plot Creation

R-Based Workflows

The R programming environment offers the most flexible options for volcano plot generation. The ggplot2 package provides complete control over plot aesthetics, including point colors, sizes, labels, and threshold lines. The EnhancedVolcano package, built on ggplot2, is specifically designed for this purpose and includes features for labeling the most significant proteins.

R-based workflows require some programming proficiency but offer the greatest reproducibility. Scripts can be version-controlled, shared with collaborators, and rerun when data or analysis parameters change. The learning curve is steep for researchers without programming experience, but the investment pays off in analytical rigor.

Web-Based Applications

Several web-based tools have been developed to make volcano plot generation accessible to researchers without programming skills. EasyPubPlot is a Shiny web application that generates publication-ready volcano plots, heatmaps, box plots, dot plots, and bubble plots with minimal steps. The tool includes step-by-step tutorials and can be used through a web browser or installed locally as an R package.

xOmicsShiny is another Shiny-based application that supports volcano plot generation as part of a broader cross-omics analysis platform. It includes principal component analysis, Venn diagrams, heatmaps, weighted gene co-expression network analysis, and advanced clustering. The pathway mapping function covers WikiPathways, Reactome, and KEGG databases, allowing researchers to move from volcano plot candidates to pathway-level interpretation within the same tool.

CURTAIN is a web-based tool designed specifically for mass spectrometry proteomics data. It allows non-mass spectrometry experts to interactively explore volcano plots, visualize replicates in bar charts or violin plots, and export publication-ready figures. The tool also provides access to protein domain structures, AlphaFold predicted structures, reported interactors, and disease links for selected protein hits.

Choosing Between Tools

The choice of tool depends on the researcher's technical skills, the complexity of the analysis, and the need for reproducibility. Web-based tools lower the barrier to entry but may limit customization and make it harder to document the exact analysis steps. R-based workflows require more effort but provide complete control and reproducibility.

For collaborative projects or data sharing, tools that generate shareable links are valuable. The developers of CURTAIN advocate that mass spectrometry data published in volcano plot format include a shareable weblink, allowing readers to explore the underlying data interactively. This practice aligns with broader trends toward data transparency and reproducibility in proteomics.

Step-by-Step Protocol for Creating a Volcano Plot

Step 1: Prepare the Expression Matrix

Start with a protein expression matrix where rows are proteins and columns are samples. Ensure that protein identifiers are consistent and that each protein has a unique identifier. Remove contaminants, decoy hits, and proteins identified with very few peptides, as these are unreliable.

Document the number of proteins quantified in each sample and the proportion of missing values. This information is essential for deciding how to handle missing data and for interpreting the final volcano plot.

Step 2: Normalize the Data

Apply normalization to correct for systematic differences between samples. The specific method depends on the quantification approach. Label-free intensity data typically benefit from median or quantile normalization. Spectral count data may require different normalization strategies.

Record the normalization method and any parameters used. This information should appear in the methods section of any report or publication.

Step 3: Handle Missing Values

Missing values in proteomics data arise from two sources. Some proteins are genuinely absent or below the detection limit in certain samples. Others are present but were not detected due to stochastic sampling or technical issues. The distinction matters because the appropriate handling differs.

If missing values are concentrated in one condition, this may indicate biological regulation. If they are scattered randomly, imputation may be appropriate. Common imputation methods include replacing missing values with a small constant, using the minimum detected value, or using more sophisticated approaches based on the data distribution.

Document the proportion of missing values and the imputation method. If more than a quarter of the values are missing for a given protein, consider excluding that protein from the analysis.

Step 4: Apply Statistical Testing

Perform differential expression analysis using a method appropriate for the data structure. For label-free data with multiple replicates, use a moderated t-test or equivalent. For isotopic labeling data, account for labeling efficiency as described in the 18O labeling literature.

The output should include a fold change and a p-value for each protein. Apply multiple testing correction to obtain adjusted p-values.

Step 5: Generate the Plot

Create the volcano plot with log2 fold change on the x-axis and negative log10 of the adjusted p-value on the y-axis. Add threshold lines for the fold change and p-value cutoffs. Color points based on whether they pass both thresholds, pass only one threshold, or fail both.

Label the most significant proteins with their gene names or protein identifiers. Avoid labeling too many points, as this makes the plot unreadable. A common approach is to label the top 10 to 20 proteins by significance.

Step 6: Annotate and Export

Add axis labels that specify the comparison being made and the units of measurement. Include the number of proteins that pass each threshold in the plot or in the figure legend. Export the plot in a vector format such as PDF or SVG for publication, and in a raster format such as PNG for presentations.

Interpreting Volcano Plot Results

Reading the Plot Correctly

The position of a protein in the volcano plot encodes two pieces of information. The horizontal position indicates the direction and magnitude of change. The vertical position indicates the statistical confidence in that change. A protein in the upper right corner has increased in abundance with high confidence. A protein in the upper left corner has decreased in abundance with high confidence.

Proteins in the lower portion of the plot should not be interpreted as unchanged. They may have changed substantially but with high variability between replicates, or they may have small changes that the study was not powered to detect. The absence of significance is not evidence of absence of effect.

Biological Interpretation

The volcano plot identifies candidate proteins, but biological interpretation requires additional context. A protein that passes both thresholds may be differentially expressed for reasons unrelated to the experimental condition. Batch effects, sample handling differences, and contamination can all produce apparent differential expression.

Pathway analysis can help place candidate proteins in biological context. Tools such as xOmicsShiny provide pathway mapping across multiple databases, allowing researchers to identify whether the differentially expressed proteins cluster in specific biological pathways. This analysis can reveal coordinated changes that would not be apparent from examining individual proteins.

Validation Requirements

Volcano plot results should be considered hypothesis-generating until validated by independent methods. Validation approaches include targeted mass spectrometry assays such as selected reaction monitoring, western blotting, or enzyme-linked immunosorbent assays. The validation method should be orthogonal to the discovery method to reduce the risk of shared technical artifacts.

The proportion of candidates that survive validation provides a measure of the false discovery rate in practice. If a large fraction of candidates fail validation, the thresholds may need to be more stringent, or the data quality may need improvement.

Common Failure Patterns and How to Avoid Them

Overinterpreting Marginal Results

A common error is treating proteins near the threshold lines as robust findings. Proteins that barely pass the significance threshold are more likely to be false positives than proteins that are far from the threshold. The volcano plot visually encourages this error because the threshold lines create a sharp boundary between significant and non-significant regions.

Researchers should examine the distribution of p-values and fold changes instead of relying solely on the binary classification. Proteins with very small p-values and large fold changes are the most reliable candidates.

Ignoring Data Quality Issues

Volcano plots can look clean and convincing even when the underlying data are poor. Low peptide counts, high missing value rates, and poor chromatographic reproducibility all degrade data quality in ways that may not be visible in the final plot.

Before generating a volcano plot, assess data quality using metrics such as the number of proteins identified, the coefficient of variation for technical replicates, and the correlation between biological replicates. Tools such as CURTAIN provide correlation matrix and profile plot analysis for this purpose.

Using Inappropriate Statistical Tests

Applying statistical tests that do not match the data structure produces unreliable p-values and misleading volcano plots. Ordinary t-tests on small sample sizes are particularly problematic because they do not account for the uncertainty in variance estimates.

The evaluation of differential expression workflows found that the choice of statistical method substantially affects the number of differentially expressed proteins identified. Researchers should use methods designed for proteomics data instead of generic statistical tests.

Failing to Document Analysis Parameters

Volcano plots generated without documentation of normalization, imputation, and threshold choices cannot be reproduced or properly evaluated. This is a significant problem for scientific transparency and for the credibility of the reported findings.

Every volcano plot should be accompanied by a description of the analysis parameters, either in the methods section of a publication or in the metadata of a data repository. This documentation should include the software versions, the statistical methods, the thresholds, and the number of proteins at each stage of the analysis.

Quality Control and Reproducibility

Internal Controls

Quality control samples should be included in every proteomics experiment. These can be pooled samples that are analyzed at regular intervals throughout the instrument run, or standard proteins spiked into each sample. The quality control data provide a measure of technical variability that can be used to assess whether observed differences between conditions are likely to be biological.

The coefficient of variation for quality control samples should be monitored throughout the experiment. If the coefficient of variation increases substantially, the data quality may be compromised, and the volcano plot results should be interpreted with caution.

Reproducibility Checks

Reproducibility can be assessed at several levels. Technical reproducibility refers to the consistency of repeated measurements of the same sample. Biological reproducibility refers to the consistency of measurements across biological replicates. Both are important for the reliability of volcano plot results.

The correlation between biological replicates provides a quick check of data quality. High correlations indicate that the biological signal is consistent across samples. Low correlations suggest that the variability between samples is dominated by technical factors or that the biological effect is highly heterogeneous.

Reporting Standards

The FAIR Guiding Principles provide a framework for making research data findable, accessible, interoperable, and reusable. Applying these principles to proteomics data means depositing raw data, processed data, and analysis scripts in appropriate repositories. The National Center for Biotechnology Information provides data resources for this purpose, and the European Bioinformatics Institute offers training on data management and analysis.

For genomic data, the National Institutes of Health Genomic Data Sharing Policy establishes expectations for data sharing and confidentiality. While this policy specifically addresses genomic data, the principles of responsible data sharing apply to proteomics data as well.

Limitations of Volcano Plot Analysis

Statistical Limitations

Volcano plots inherit all the limitations of the statistical tests that generate the underlying p-values. When sample sizes are small, the p-values are imprecise, and the volcano plot may misrepresent the confidence in the observed changes. When the data violate the assumptions of the statistical test, the p-values are invalid regardless of how the plot is constructed.

The fold change axis also has limitations. Log2 fold changes compress large changes and expand small changes, which can make a two-fold change appear more similar to a four-fold change than the underlying biology warrants. The choice of transformation affects the visual interpretation of the plot.

Biological Limitations

A volcano plot cannot distinguish between changes that are causally related to the experimental condition and changes that are secondary effects. A protein that changes in abundance may be a direct response to the treatment, an indirect response through a signaling cascade, or a consequence of changes in cell composition.

The plot also cannot distinguish between changes in protein abundance and changes in protein modification. A protein with the same abundance but different phosphorylation status would not appear as differentially expressed in a standard volcano plot, even though the modification may be biologically significant. Tools such as CURTAIN-PTM address this limitation by allowing analysis of post-translational modification sites.

Technical Limitations

The depth of proteome coverage affects volcano plot interpretation. Proteins that are not detected cannot appear in the plot, regardless of their biological importance. Low-abundance proteins are particularly likely to be missed, which can bias the set of candidates toward abundant proteins.

The dynamic range of mass spectrometry also limits the ability to detect changes in proteins that are present at very different abundances. A two-fold change in an abundant protein may be easily detected, while the same relative change in a low-abundance protein may be missed entirely.

Integrating Volcano Plots with Broader Analysis

Cross-Omics Integration

Volcano plots are not limited to proteomics data. The same visualization can be applied to transcriptomics, metabolomics, and lipidomics data. Tools such as xOmicsShiny support cross-omics data integration, allowing researchers to compare volcano plots across data types and identify concordant changes.

The integration of multiple omics data types can reveal biological insights that are not apparent from any single data type. For example, a study of gut microbiota and metabolic features in patients with prolactinoma used multi-omics profiling to identify distinct trends in microbial composition and metabolic alterations. The integration of metagenomic and metabolomic data provided a more complete picture than either approach alone.

Pathway-Level Interpretation

Moving from individual proteins to pathways requires additional analysis. The differentially expressed proteins identified by the volcano plot can be mapped to known biological pathways using databases such as WikiPathways, Reactome, and KEGG. Tools such as xOmicsShiny and EasyPubPlot provide pathway mapping functionality.

Pathway analysis can reveal coordinated changes that would not be apparent from examining individual proteins. For example, a set of proteins that individually show modest changes may collectively indicate activation of a specific signaling pathway. This type of analysis adds biological meaning to the statistical findings of the volcano plot.

Mendelian Randomization and Causal Inference

In some research contexts, volcano plot findings can be integrated with causal inference methods. Mendelian randomization uses genetic variants as instrumental variables to assess causal relationships between exposures and outcomes. This approach has been applied to study the relationship between chronic obstructive pulmonary disease and lung cancer, identifying blood metabolites that mediate the effect.

While Mendelian randomization is beyond the scope of standard proteomics analysis, the integration of differential expression findings with genetic evidence can strengthen causal claims. Researchers should be aware of these possibilities when designing studies and interpreting volcano plot results.

Professional Escalation Criteria

When to Seek Statistical Consultation

Researchers should consult a statistician or bioinformatics specialist when the experimental design is complex, when the data violate statistical assumptions, or when the results are unexpected. Specific situations that warrant consultation include experiments with very few replicates, data with high missing value rates, and analyses where different statistical methods produce substantially different results.

Statistical consultation is particularly important when the volcano plot results will inform regulatory decisions, clinical studies, or other high-stakes applications. The cost of a statistical error in these contexts can be substantial.

When to Reconsider the Experimental Design

If the volcano plot shows very few significant proteins, the experiment may be underpowered. Increasing the number of replicates, improving the sample preparation, or using a more sensitive quantification approach may be necessary. If the volcano plot shows an implausibly large number of significant proteins, the data may contain systematic artifacts that need to be investigated.

The decision to repeat an experiment should be based on data quality metrics instead of on the desire to obtain significant results. Repeating an experiment with the same design and the same technical issues will likely produce the same problems.

When to Escalate Data Quality Concerns

If quality control metrics indicate problems with the instrument, the samples, or the analysis pipeline, these issues should be escalated to the appropriate personnel. Instrument maintenance, sample preparation protocols, and analysis software should all be reviewed when data quality is compromised.

The decision to exclude data from analysis should be documented and justified. Excluding samples because they are outliers can introduce bias if the exclusion is not based on pre-specified criteria.

Frequently Asked Questions

What is the difference between a volcano plot and a MA plot?

A volcano plot displays statistical significance against fold change, while an MA plot displays log intensity against log fold change. The MA plot is useful for assessing the relationship between abundance and fold change, while the volcano plot is designed for identifying statistically significant changes. Both plots can be generated from the same data and serve complementary purposes.

How many replicates are needed for a reliable volcano plot?

The number of replicates needed depends on the biological variability of the system and the magnitude of the expected changes. Three biological replicates per condition is a common minimum, but more replicates are needed when variability is high or when small changes are of interest. The statistical power of the experiment should be assessed before data collection begins.

Should I use raw p-values or adjusted p-values in the volcano plot?

Adjusted p-values should be used for determining which proteins are considered significant, because they account for the multiple testing problem. Some researchers plot raw p-values and apply the adjustment only in the significance classification, but this can be misleading. The y-axis should reflect the values used for significance determination.

How do I handle proteins with missing values in some samples?

The approach depends on the pattern of missingness. If missing values are concentrated in one condition, this may be biologically meaningful and should not be imputed. If missing values are scattered, imputation may be appropriate. Document the proportion of missing values and the imputation method used.

What fold change threshold should I use?

The fold change threshold should be based on the technical variability of the assay and the biological context. A two-fold change is a common starting point, but smaller thresholds may be appropriate for well-controlled experiments with low variability. The threshold should be justified in the methods section.

Can I compare volcano plots from different experiments?

Direct comparison of volcano plots from different experiments is problematic because the thresholds, normalization methods, and statistical tests may differ. If comparison is necessary, the analysis pipelines should be standardized across experiments. Cross-experiment comparisons should focus on the direction and consistency of changes instead of on the specific proteins that pass thresholds.

What should I do if my volcano plot shows no significant proteins?

A volcano plot with no significant proteins may indicate that the experiment is underpowered, that the biological effect is small, or that the data quality is poor. Review the quality control metrics, consider whether the sample size is adequate, and evaluate whether the statistical methods are appropriate for the data structure.

How should I report volcano plot results in a publication?

Report the number of proteins quantified, the number of proteins that pass each threshold, and the specific thresholds used. Provide the analysis parameters in the methods section, including the normalization method, the imputation approach, the statistical test, and the multiple testing correction. Deposit the data and analysis scripts in an appropriate repository to enable reproducibility.

Related Bioinformatics Guides

• The Rise of Omics: Genomics, Proteomics, and Metabolomics
• The Nagoya Protocol and Digital Sequence Information (DSI)
• Cryo-EM Density Map Interpretation and Computational Structure Fitting
• GWAS Manhattan Plot R: Structural Analysis and Computational Methodologies in Bioinformatics
• The Bioinformatics Revolution in Structural Proteomics and Computational Drug Discovery: A Unified Paradigm

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.
• xOmicsShiny: an R Shiny application for cross-omics data analysis and pathway mapping.. Bioinformatics advances, 2025.
• CURTAIN-A unique web-based tool for exploration and sharing of MS-based proteomics data.. Proceedings of the National Academy of Sciences of the United States of America, 2024.
• Visual Genomics Analysis Studio as a Tool to Analyze Multiomic Data.. Frontiers in genetics, 2021.
• EasyPubPlot: A Shiny Web Application for Rapid Omics Data Exploration and Visualization.. Journal of proteome research, 2025.
• Exploratory Pilot Multi-Omics Profiling of Gut Microbiota and Metabolic Features in Patients with Prolactinoma.. 2026.
• Chronic Obstructive Pulmonary Disease Influence on Lung Cancer Risk Through Blood Metabolite Mediation: A Two-Sample Mendelian Randomisation and Mediation Analysis.. 2026.
• Assessment of differences in immune responses to albumin compared with crystalloids for resuscitation in sepsis.. 2026.
• Development of K-CORE: a web-based platform for integrated clinico-genomic analysis.. 2026.
• Optimizing differential expression analysis for proteomics data via high-performing rules and ensemble inference. Nature Communications, 2024.
• Current trends in differential expression proteomics: isotopically coded tags.. Trends in Biotechnology, 2001.
• Improved Method for Differential Expression Proteomics Using Trypsin-catalyzed 18O Labeling with a Correction for Labeling Efficiency *S. Molecular & Cellular Proteomics, 2007.

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