# Network Visualization in Proteomics: Best Practices for Creating Publication-Ready Figures with Cytoscape Proteomics experiments generate complex interaction datasets that require careful visualization to communicate biological findings effectively. This article provides a practical framework for creating publication-ready network figures using Cytoscape, covering data preparation, layout selection, node and edge styling, annotation strategies, and quality control. The guidance applies to researchers working with protein-protein interaction data, co-expression networks, and quantitative proteomics results derived from mass spectrometry experiments. ## Understanding the Role of Network Visualization in Proteomics Network visualization serves a distinct purpose in proteomics data analysis. Unlike bar charts or volcano plots that display statistical significance and fold changes, network figures reveal relationships between proteins, functional modules, and pathway context. A well-constructed network figure allows readers to grasp the organizational structure of a proteomics dataset at a glance, showing which proteins cluster together, which serve as hubs, and how quantitative changes distribute across functional groups. The challenge for most researchers is not generating the network data but transforming raw interaction information into a figure that meets journal standards and communicates findings without distortion. Mass spectrometry-based proteomics produces high-dimensional datasets with thousands of quantified proteins, and the interaction data derived from databases or experimental assays requires thoughtful filtering before visualization becomes meaningful. Cytoscape remains one of the most widely used platforms for biological network visualization and analysis. Its open-source architecture, extensive plugin ecosystem, and support for multiple data formats make it suitable for both novice and advanced users. The platform supports standard layout algorithms, customizable node and edge styling, and integration with annotation resources that enrich network interpretation. ## Data Inputs and Preparation for Network Construction ### Source Data Formats and Import Requirements Before opening Cytoscape, researchers must prepare their interaction data in a format the software can interpret. The most common approach involves a two-column edge list where each row represents an interaction between two proteins. A typical edge list contains at least three columns: source node identifier, target node identifier, and an optional interaction score or confidence value. For proteomics applications, the node identifiers should match the protein identifier system used throughout the analysis. UniProt accession numbers provide a stable and widely recognized identifier system, while gene symbols offer readability at the cost of potential ambiguity. The NCBI maintains comprehensive protein and gene databases that researchers can use to verify identifier accuracy and retrieve additional annotation information before constructing networks. Cytoscape also accepts attribute files that contain node-level information such as quantitative values, functional annotations, or experimental group assignments. These attributes become the basis for visual styling decisions, allowing researchers to encode fold changes, significance values, or protein abundance directly into node appearance. ### Filtering Interactions Before Visualization Raw interaction data almost always requires filtering before it produces a readable figure. Protein interaction databases contain millions of records, and importing all interactions for a set of hundreds or thousands of quantified proteins creates an unreadable hairball. The filtering strategy depends on the biological question and the confidence metrics available in the source data. Confidence scores provide the first filtering criterion. Many interaction databases report scores that reflect the likelihood of a genuine biological interaction based on experimental evidence and computational predictions. Setting a minimum confidence threshold reduces false-positive interactions and focuses the network on the most reliable relationships. The appropriate threshold varies by database and research question, so researchers should document their choice and justify it in the methods section. For quantitative proteomics data, additional filtering based on detection frequency and abundance can reduce noise. Proteins identified in only a small fraction of samples or with high missingness across replicates may not warrant inclusion in the network. The high-dimensional nature of proteomic datasets, including missingness and batch effects, requires careful preprocessing before network construction, as noted in recent reviews of machine learning approaches in proteomics-driven research. ### Handling Quantitative Data for Network Encoding Quantitative proteomics data can be encoded in network figures through node color, node size, or both. A common approach uses a color gradient to represent log2 fold change between experimental conditions, with red indicating upregulation and blue indicating downregulation. Node size can represent absolute abundance, statistical significance, or connectivity degree, depending on the message the figure should convey. The choice of quantitative encoding should follow the principles of effective data visualization. Color gradients work well for continuous variables with a natural midpoint, such as log2 fold change centered at zero. Sequential color schemes suit variables where only magnitude matters, such as abundance or confidence. Researchers should avoid rainbow color schemes that introduce perceptual artifacts and should ensure that color choices remain distinguishable for readers with color vision deficiencies. ## Core Principles for Publication-Ready Network Figures ### Clarity Through Node and Edge Styling The visual properties of nodes and edges determine whether a network figure communicates effectively or confuses the reader. Node size, color, shape, and border properties each carry meaning and should be assigned deliberately instead of left at default values. Node size should reflect a meaningful quantitative property when possible. In proteomics networks, node size often encodes protein abundance, connectivity degree, or statistical significance. The range of sizes should be large enough to be perceptible but not so extreme that small nodes become invisible or large nodes obscure their neighbors. A minimum node size of approximately 10 pixels and a maximum of approximately 60 pixels provides a reasonable range for most figure dimensions. Node color can encode categorical information such as functional class, experimental group, or subcellular localization. When color encodes a categorical variable, the palette should use perceptually distinct hues with sufficient contrast. When color encodes a continuous variable, the gradient should progress smoothly from a light to a dark shade or through a diverging scheme with a neutral midpoint. Edge styling communicates interaction properties. Edge thickness can represent interaction confidence or quantitative strength, while edge color can indicate interaction type or direction. Dashed edges commonly represent predicted or low-confidence interactions, while solid edges represent experimentally validated interactions. The default gray edges in Cytoscape often appear too faint in printed figures, so increasing edge width and adjusting opacity improves visibility. ### Layout Selection and Network Topology The layout algorithm determines the spatial arrangement of nodes and edges, which profoundly affects the readability of the network. No single layout works for all networks, and researchers should test multiple options before selecting the final arrangement. Force-directed layouts, such as the Prefuse Force Directed Layout and the yFiles Organic Layout, position nodes based on simulated physical forces where connected nodes attract each other and unconnected nodes repel. These layouts often reveal community structure and work well for networks with moderate node counts. However, force-directed layouts can produce unstable results across runs, so researchers should fix the random seed to ensure reproducibility. Circular layouts place nodes around a circle and work well for networks with a clear hub-and-spoke structure or for comparing node properties without emphasizing topology. Grid layouts provide uniform spacing and suit very large networks where individual edge paths matter less than overall density patterns. The network topology itself should inform layout choice. Networks with strong community structure benefit from layouts that separate clusters visually. The NORMA web tool demonstrates how annotation-based group separation can improve network readability, allowing users to adjust network views using standard layout algorithms or slight modifications that enhance group separation for publication figures. ### Annotation and Labeling Strategies Labels transform a network from an abstract diagram into an interpretable biological figure. However, labeling every node creates clutter that obscures the network structure. The labeling strategy should balance information content with visual clarity. For most publication figures, labeling only the most important nodes provides the best balance. Hub proteins, differentially abundant proteins, and proteins central to the biological narrative deserve labels. Other nodes can remain unlabeled, with their identities recoverable from supplementary tables or interactive versions of the figure. Label placement and font size require attention. Labels should not overlap nodes or edges, and font sizes should remain legible at the final figure dimensions. Cytoscape offers label positioning options that place text adjacent to nodes, and researchers should adjust label offsets to minimize overlap. Functional annotations enrich network interpretation. Gene Ontology terms, pathway memberships, and protein domain information can be encoded through node color, shape, or border style. The NORMA tool demonstrates how precalculated annotations such as Gene Ontology terms, pathway enrichment results, and community detection outputs can be visualized in networks as colored pie-chart nodes or color-filled areas, enabling the simultaneous encoding of network structure, community membership, and node categories or expression values. ## Practical Workflow for Creating Network Figures in Cytoscape ### Step 1: Prepare and Import Data Begin by assembling the edge list and node attribute files. The edge list should contain at least source and target identifiers, with optional columns for interaction confidence, evidence type, or quantitative strength. The node attribute file should contain node identifiers matching the edge list, along with any quantitative values, functional annotations, or group assignments to be visualized. Import the edge list into Cytoscape using the Import Network from File option. Verify that the identifier columns are correctly mapped and that the network contains the expected number of nodes and edges. Import the node attribute file using the Import Table from File option, ensuring that the key column matches the node identifiers in the network. ### Step 2: Apply Initial Styling Open the Style panel and define the visual mappings for nodes and edges. Set the default node size, color, and label properties before applying data-driven mappings. For quantitative proteomics data, create a continuous mapping for node color based on log2 fold change and a continuous mapping for node size based on abundance or significance. Apply edge width mapping based on interaction confidence or quantitative strength. Set the default edge color to a neutral gray and adjust opacity to approximately 70 percent so that overlapping edges remain distinguishable. ### Step 3: Select and Apply a Layout Test several layout algorithms to identify the arrangement that best reveals the network structure. Start with a force-directed layout and evaluate whether communities separate clearly. If the network contains known functional modules, check whether these modules appear as distinct clusters. For networks with more than 200 nodes, consider filtering to a subnetwork of interest before layout. The full network may be too dense for meaningful visualization, and a focused subnetwork often communicates the biological message more effectively. ### Step 4: Refine Labels and Annotations Select the nodes that require labels based on their biological importance. Apply label mappings that display gene symbols or protein names for these selected nodes while leaving other nodes unlabeled. Adjust label font size and position to minimize overlap. Add annotations that provide biological context. This may include coloring nodes by functional category, adding borders to indicate membership in specific complexes, or using node shape to distinguish protein types. ### Step 5: Export at Publication Resolution Export the final figure at a resolution appropriate for the target journal. Most journals require a minimum of 300 dots per inch for color figures and 600 dots per inch for line art. Cytoscape supports PDF and SVG export, which provide vector graphics that scale without loss of quality. Set the figure dimensions to match the journal column width. A single-column figure typically measures 85 to 90 millimeters wide, while a double-column figure measures 170 to 180 millimeters. Adjust node sizes, label fonts, and edge widths so that all elements remain legible at the final printed dimensions. ## At a Glance: Network Visualization Decision Table | Decision Point | Recommended Approach | Common Alternative | When to Choose Alternative | | --- | --- | --- | --- | | Node identifier system | UniProt accession numbers | Gene symbols | Gene symbols when readers need immediate biological recognition | | Interaction filtering | Confidence score threshold | Top N interactions by score | Top N approach when confidence scores are unavailable or unreliable | | Quantitative encoding | Node color for fold change, node size for abundance | Node color for abundance, node size for significance | Swap when abundance differences are the primary finding | | Layout algorithm | Force-directed for community structure | Circular for hub-and-spoke topology | Circular layout when the network has a clear central hub | | Labeling strategy | Label hubs and key differentially abundant proteins | Label all nodes | Label all nodes only for small networks under 50 nodes | | Export format | Vector PDF or SVG | High-resolution PNG or TIFF | Raster format when the journal requires specific raster specifications | ## Layout Algorithms and Their Tradeoffs ### Force-Directed Layouts Force-directed layouts simulate physical systems where nodes repel each other and edges act as springs pulling connected nodes together. These layouts excel at revealing community structure and are the default choice for most biological networks. The Prefuse Force Directed Layout in Cytoscape provides a good balance of speed and visual quality for networks up to several thousand nodes. The main limitation of force-directed layouts is their sensitivity to initial conditions and random seeds. Running the same layout twice can produce different arrangements, which complicates reproducibility. Researchers should document the layout algorithm and any random seed settings in their methods. Some layout implementations allow fixing the random seed to ensure identical results across runs. ### Circular and Grid Layouts Circular layouts place all nodes on the circumference of a circle, with edges drawn as chords across the interior. This layout works well for networks with a central hub connected to many peripheral nodes, as it clearly shows the hub-and-spoke structure. Circular layouts also facilitate comparisons of node properties because all nodes receive equal visual prominence. Grid layouts arrange nodes in a regular grid pattern, which suits very large networks where individual edge paths are less important than overall density and connectivity patterns. Grid layouts do not reveal community structure and should be reserved for specific use cases such as adjacency matrix visualizations. ### Hierarchical Layouts Hierarchical layouts arrange nodes in layers based on their position in a directed graph, making them suitable for signaling pathways and other networks with directional information. These layouts emphasize the flow of information from upstream to downstream nodes. Cytoscape supports hierarchical layouts through plugins and built-in options, though they require directed edges to produce meaningful arrangements. ### Layout Selection Criteria The choice of layout should follow from the biological question and the network properties. Networks with strong modular structure benefit from force-directed layouts that separate communities. Networks with a clear hierarchical organization benefit from hierarchical layouts. Networks with a dominant hub benefit from circular layouts that emphasize the hub-and-spoke pattern. Researchers should evaluate layout quality by examining whether known functional relationships appear as spatial proximity. If proteins from the same pathway or complex scatter across the figure, the layout may not be revealing the underlying structure effectively. The NORMA tool offers automated community detection when no preexisting annotations exist, and it can modify standard layouts to achieve better group separation for publication figures. ## Node and Edge Styling for Quantitative Proteomics Data ### Encoding Quantitative Values Through Node Color Color mapping provides the most flexible encoding for quantitative proteomics data. A diverging color scheme with a neutral midpoint works well for log2 fold change values, where the midpoint represents no change. Common choices include red-blue, red-green, or orange-purple diverging palettes. Red-green schemes should be avoided because they are indistinguishable for readers with red-green color blindness. The color range should be set to the data range or to a biologically meaningful threshold. Setting the range to the full data distribution can compress differences among the majority of nodes if extreme outliers exist. Trimming the color range to the 5th and 95th percentiles of the data often produces more informative figures, with nodes beyond the range shown at the extreme colors. ### Encoding Quantitative Values Through Node Size Node size provides a second channel for quantitative encoding that works well in combination with color. Size is particularly effective for encoding abundance, connectivity degree, or statistical significance because larger nodes naturally draw attention. The size mapping should use a continuous scale with a minimum and maximum that remain visually distinct. A common approach maps the square root of the quantitative value to node size, which prevents highly connected hubs from dominating the figure. The relationship between the mapped value and node size should be documented in the figure legend. ### Edge Styling for Interaction Properties Edge width and color can encode interaction confidence, quantitative strength, or interaction type. Thicker edges draw more attention and should represent stronger or more important interactions. Edge color can distinguish interaction types, such as physical interactions versus genetic interactions, or can encode a continuous confidence score. Edge opacity affects the readability of dense networks. When many edges overlap, reducing opacity to approximately 50 to 70 percent allows overlapping edges to remain visible while preventing the figure from becoming a solid mass. The default edge color in Cytoscape is often too light for print, so increasing edge darkness improves figure quality. ### Visual Hierarchy and Emphasis Effective network figures establish a visual hierarchy that guides the reader to the most important information. Nodes with high connectivity, significant quantitative changes, or central roles in the biological narrative should receive visual emphasis through larger size, stronger colors, or thicker borders. Less important nodes should recede visually through smaller size and muted colors. The visual hierarchy should align with the biological message. If the figure aims to highlight differentially abundant proteins, these proteins should stand out clearly from the background network. If the figure aims to show pathway organization, the pathway members should be visually grouped and distinguished from nonmembers. ## Annotation Strategies for Biological Context ### Functional Annotation Encoding Functional annotations add biological meaning to network figures. Gene Ontology terms, pathway memberships, protein complexes, and subcellular localizations can all be encoded through node visual properties. The choice of annotation depends on the biological question and the message the figure should convey. Categorical annotations work well with node color or shape. Assigning distinct colors to functional categories allows readers to see at a glance whether proteins from the same pathway cluster together in the network. Node shape can encode a second categorical variable, such as protein class or experimental group. The NORMA web tool demonstrates advanced annotation visualization approaches, including colored pie-chart nodes that show the proportion of annotations for each node and color-filled areas that show annotation enrichment across network regions. These approaches allow the simultaneous encoding of network structure, community membership, and node categories or expression values, which is valuable for complex proteomics datasets. ### Community Detection and Cluster Annotation When functional annotations are unavailable or incomplete, community detection algorithms can identify groups of densely connected nodes that may represent functional modules. Cytoscape includes community detection through plugins such as clusterMaker, which implements multiple algorithms including Markov clustering and community clustering. The results of community detection can be encoded through node color, with each community assigned a distinct color. This approach reveals the modular organization of the network without requiring external functional annotations. The NORMA tool offers automated community detection when no preexisting annotations exist, making it a useful complement to Cytoscape for researchers who need annotation-based visualization. ### Labeling for Interpretability Labels provide the most direct form of annotation but require careful management to avoid clutter. The labeling strategy should prioritize nodes that contribute most to the biological message. Hub proteins, differentially abundant proteins, and proteins mentioned in the text deserve labels. Other nodes can remain unlabeled, with identities available in supplementary materials. Label font size should match the figure scale. Labels that are too small become illegible when the figure is printed at column width, while labels that are too large obscure the network structure. A font size between 6 and 10 points at final figure dimensions works well for most publications. ## Common Failure Patterns in Network Figure Creation ### The Hairball Problem The most common failure in network visualization is the hairball, a dense mass of nodes and edges that conveys no meaningful information. Hairballs result from importing too many interactions without adequate filtering or from visualizing networks that are too large for the figure dimensions. Preventing hairballs requires aggressive filtering before visualization. Set confidence thresholds that retain only high-quality interactions, filter to a subnetwork of biological interest, and consider whether the full network needs to appear in the main figure or can be relegated to supplementary materials. The high-dimensional and heterogeneous nature of proteomic datasets, including missingness and batch effects, requires careful preprocessing before network construction. ### Misleading Quantitative Encoding Quantitative encoding can mislead readers when the mapping is poorly chosen or undocumented. Color gradients that compress most of the data into a narrow range obscure differences among the majority of nodes. Node size mappings that use a linear scale allow highly connected hubs to dominate the figure and obscure smaller nodes. Avoiding misleading encoding requires attention to the data distribution and the message the figure should convey. Check the distribution of quantitative values before setting color and size ranges. Consider whether a logarithmic transformation would produce a more informative mapping. Document all mappings in the figure legend so readers can interpret the visual encoding correctly. ### Overlapping Labels and Nodes Label overlap is a common problem that reduces figure readability. When many nodes cluster together, their labels overlap and become unreadable. Cytoscape offers some automatic label placement options, but manual adjustment is often necessary for publication figures. Strategies for reducing label overlap include labeling only a subset of nodes, increasing label offset from nodes, and manually repositioning labels in the final figure. Some researchers prefer to add labels in a vector graphics editor after exporting the network from Cytoscape, which provides complete control over label placement. ### Inconsistent Identifier Usage Inconsistent protein identifiers create confusion and undermine reproducibility. Mixing UniProt accessions, gene symbols, and RefSeq identifiers in the same figure makes it difficult for readers to match nodes to the proteins they represent. The NCBI maintains comprehensive databases that researchers can use to verify identifier accuracy and convert between identifier systems. Standardize on a single identifier system for the entire figure. UniProt accession numbers provide the most stable and unambiguous identifiers, while gene symbols offer readability. If gene symbols are used, verify that they are current and unambiguous, as some gene symbols have changed or been retired. ## Reproducibility and Documentation Standards ### Recording Visualization Parameters Reproducible network figures require documentation of all visualization parameters. The methods section should describe the data sources, filtering thresholds, layout algorithm, and styling choices. This documentation allows other researchers to reproduce the figure and understand the decisions that shaped it. Cytoscape sessions save the network, styles, and layouts in a single file, which provides a convenient record of the visualization process. Saving the session file alongside the figure ensures that the visualization can be revisited and modified if needed. The session file should be deposited in a data repository or made available as supplementary material. ### Version Control for Figures Network figures often undergo multiple revisions during the publication process. Tracking these revisions requires a systematic approach to file naming and version control. Include the date and a version number in the figure filename, and maintain a record of changes made between versions. The reproducibility principles promoted by bioinformatics training resources apply to network visualization as well. The Carpentries lessons on version control with Git provide practical guidance for tracking changes to analysis files, including figure generation scripts. Even for researchers who do not use formal version control, maintaining a clear record of figure versions and the parameters used to generate each version supports scientific integrity. ### Integration with Reproducible Workflows Network visualization should be integrated into the broader reproducible analysis workflow. The Galaxy Training Network provides accessible workflow training that emphasizes reproducibility in bioinformatics analysis. Similarly, the nf-core documentation describes community standards for reproducible pipelines that can include network visualization steps. For proteomics data analysis, the workflow from raw mass spectrometry data to network visualization should be documented end to end. This includes the database search parameters, quantification methods, statistical analysis, and network construction steps. The Bioconductor project provides official documentation for reproducible genomic analysis workflows that can be adapted to include network visualization. ## Quality Control and Figure Validation ### Checking Figure Legibility at Final Dimensions A figure that looks clear on a computer monitor may become illegible when printed at column width. Before submitting a figure, print it at the final dimensions or view it at the final size on screen. Check that node sizes remain distinguishable, labels remain readable, and edge paths remain traceable. The minimum font size for publication figures is typically 6 points, with 7 to 8 points preferred for most labels. Node sizes should remain above approximately 10 pixels at final dimensions. Edge widths should be at least 1 pixel at final dimensions to remain visible in print. ### Verifying Data Integrity in the Figure The figure should accurately represent the underlying data. Verify that the quantitative values encoded in node color and size match the analysis results. Check that the network contains the expected number of nodes and edges and that the filtering criteria were applied correctly. For quantitative proteomics data, verify that the fold change values and significance thresholds used in the figure match the statistical analysis results. The CURTAIN tools provide interactive visualization of mass spectrometry proteomics data, allowing researchers to verify volcano plots and replicate data before creating network figures. These tools enable non-mass-spectrometry experts to peruse primary experimental data and export publication-ready figures. ### Seeking Feedback from Colleagues A fresh perspective often identifies readability problems that the figure creator has become blind to. Ask colleagues who are not familiar with the specific dataset to interpret the figure. If they cannot identify the main message or struggle to distinguish key nodes, the figure needs revision. Feedback should focus on the biological message instead of aesthetic preferences. Does the figure clearly show which proteins are differentially abundant? Does it reveal the expected pathway relationships? Does it support the claims made in the text? These questions guide effective figure revision. ## Limitations and Interpretation Boundaries ### Network Visualization Does Not Prove Interaction A network figure shows relationships that may be inferred from databases, predicted computationally, or measured experimentally. The figure itself does not prove that the interactions occur in the biological system under study. The distinction between experimentally validated interactions and predicted interactions should be clear in the figure through edge styling or annotation. Interaction databases contain both high-confidence experimentally derived interactions and lower-confidence computational predictions. The confidence scores associated with each interaction should inform both the filtering threshold and the visual encoding. Researchers should be transparent about the evidence basis for the interactions shown in their figures. ### Quantitative Values Reflect Relative Abundance Quantitative proteomics data typically measures relative abundance changes instead of absolute protein concentrations. The fold change values encoded in node color represent the ratio between conditions, not the absolute amount of protein present. This distinction should be clear in the figure legend and the accompanying text. The limitations of quantitative proteomics, including missingness, batch effects, and limited cohort size, apply to network visualization as well. A recent review of AI and machine learning approaches in proteomics-driven drug discovery emphasizes that proteomic datasets are high-dimensional, heterogeneous, and frequently affected by missingness and batch effects. These limitations should be acknowledged when interpreting network figures derived from such data. ### Network Topology Depends on Data Quality The topology of the visualized network depends on the quality and completeness of the underlying interaction data. Missing interactions produce networks with artificially low connectivity, while false-positive interactions produce networks with misleading connections. The filtering thresholds applied before visualization shape the network topology and should be chosen carefully. The choice of interaction database also affects network topology. Different databases contain different sets of interactions, and the same protein set can produce substantially different networks depending on the database used. Researchers should document the database version and access date, as databases are updated regularly. ## Professional Escalation Criteria ### When to Seek Specialized Support Most network visualization tasks can be completed with standard Cytoscape workflows. However, certain situations warrant consultation with bioinformatics specialists or computational biologists. These include analyses involving very large networks exceeding the capacity of standard desktop computers, integration of multiple omics data types, and development of custom visualization approaches. The EMBL-EBI Training program offers learning pathways for bioinformatics data resources and practical analysis education. Researchers who need to develop advanced network analysis skills can benefit from these structured training resources. Similarly, the Galaxy Training Network provides accessible workflow training that covers reproducible analysis approaches applicable to network visualization. ### When to Consider Alternative Tools Cytoscape is not the only option for network visualization, and some situations warrant alternative tools. The NORMA web tool provides specialized annotation visualization capabilities, including pie-chart nodes and color-filled areas for annotation enrichment. The CURTAIN tools provide interactive visualization of mass spectrometry proteomics data with publication-ready export options. For researchers who need to share interactive network figures with collaborators or readers, web-based visualization platforms may be more appropriate than static Cytoscape exports. Interactive figures allow readers to explore the network, zoom into regions of interest, and access node-level details that cannot be shown in a static figure. ### When to Escalate Data Quality Concerns Network figures derived from problematic data will inherit and amplify data quality issues. If the underlying proteomics data shows severe batch effects, high missingness, or poor replicate correlation, these issues should be addressed before investing effort in network visualization. The CURTAIN tools provide correlation matrix and profile plot analysis for assessing overall experimental quality. Professional escalation is appropriate when data quality issues cannot be resolved through standard preprocessing approaches. This may involve consultation with biostatisticians, bioinformatics core facilities, or collaborators with specialized expertise in proteomics data analysis. ## Records and Documentation for Publication ### Methods Section Documentation The methods section should provide sufficient detail for another researcher to reproduce the network figure. This includes the interaction data source and version, the filtering criteria and thresholds, the layout algorithm and parameters, and the styling choices for nodes and edges. For proteomics data, the methods should describe the mass spectrometry workflow, database search parameters, quantification approach, and statistical analysis. The integration of proteomics data with interaction databases should be described clearly, including how protein identifiers were matched and how quantitative values were mapped to network nodes. ### Supplementary Materials Supplementary materials should include the full network data, the Cytoscape session file, and any custom scripts used for data processing or visualization. These materials allow readers to explore the network interactively and verify the figure content. The trend toward sharing proteomics data in accessible formats is growing. The CURTAIN tools advocate for reporting mass spectrometry data in volcano plot format with shareable weblinks, allowing readers to better analyze and exploit the data. Similar principles apply to network figures, where sharing the underlying network data and visualization session enhances transparency and reproducibility. ### Figure Legends Figure legends should describe the visual encoding used in the figure. This includes the meaning of node color, node size, edge width, and edge color. The legend should also state the filtering criteria, the number of nodes and edges shown, and any limitations of the visualization. A well-written figure legend allows readers to interpret the figure without referring to the main text. It should be self-contained and provide all information needed to understand the visual encoding. The legend should also indicate where the full network data and visualization session can be accessed. ## Frequently Asked Questions ### What is the ideal number of nodes for a publication-ready network figure? The ideal node count depends on the figure dimensions and the biological message. For a single-column figure, 50 to 150 nodes typically produces a readable network. For a double-column figure, 100 to 300 nodes can work well. Networks with more than 500 nodes usually require filtering to a subnetwork of interest or relegation to supplementary materials. The key consideration is whether individual nodes and edges remain distinguishable at the final printed dimensions. ### How should I choose between UniProt accessions and gene symbols for node labels? UniProt accession numbers provide stable and unambiguous identifiers that are ideal for data processing and database matching. Gene symbols offer immediate biological recognition for readers but can be ambiguous or outdated. A practical approach uses UniProt accessions for the underlying data and gene symbols for figure labels, with a mapping table provided in supplementary materials. Verify gene symbol currency against the NCBI databases before finalizing the figure. ### What confidence threshold should I use when filtering protein interactions? The appropriate confidence threshold depends on the interaction database and the research question. Higher thresholds produce smaller, more reliable networks, while lower thresholds produce larger, more comprehensive networks with more false positives. A common approach starts with a moderate threshold and adjusts based on the resulting network size and the biological plausibility of the interactions. Document the chosen threshold and justify it in the methods section. ### How can I make my network figure accessible to color-blind readers? Avoid red-green color schemes, which are indistinguishable for readers with red-green color blindness. Use color palettes that remain distinguishable under common color vision deficiencies, such as blue-orange or purple-green diverging schemes. Additionally, encode important information through multiple channels, such as node size or shape in addition to color, so that the figure remains interpretable even without color perception. ### Should I label every node in my network figure? Labeling every node creates clutter that obscures the network structure. A better approach labels only the most important nodes, such as hub proteins, differentially abundant proteins, and proteins central to the biological narrative. Other nodes remain unlabeled, with their identities available in supplementary tables or interactive versions of the figure. For small networks under 50 nodes, labeling all nodes may be feasible if labels are placed carefully to avoid overlap. ### What export format should I use for journal submission? Vector formats such as PDF or SVG provide the highest quality for journal submission because they scale without loss of resolution. Most journals accept vector figures and prefer them for line art and diagrams. If the journal requires raster formats, export at 300 dots per inch for color figures and 600 dots per inch for line art. Set the figure dimensions to match the journal column width before export. ### How do I document my visualization workflow for reproducibility? Save the Cytoscape session file, which contains the network, styles, and layout. Document the data sources, filtering thresholds, layout algorithm, and styling choices in the methods section. Deposit the session file and underlying data in a repository or provide them as supplementary materials. Record the software version and any plugins used, as these can affect the visualization output. ### When should I consider using tools other than Cytoscape? Consider alternative tools when you need specialized capabilities that Cytoscape does not provide. The NORMA web tool offers advanced annotation visualization with pie-chart nodes and color-filled areas for annotation enrichment. The CURTAIN tools provide interactive visualization of mass spectrometry proteomics data with publication-ready export options. Web-based platforms may be more appropriate when you need to share interactive figures with collaborators or readers. ## Related Bioinformatics Guides - [Volcano Plot Proteomics: How to Create and Interpret Them Effectively](/knowledge/bioinformatics/volcano-plot-proteomics-how-to-create-and-interpret-them-effectively) - [Genomic Data Visualization Tools: Choosing and Using Them Effectively](/knowledge/bioinformatics/genomic-data-visualization-tools-choosing-and-using-them-effectively) - [Proteomics Data Analysis in R: A Practical Workflow for Differential Expression and Visualization](/knowledge/bioinformatics/proteomics-data-analysis-in-r-a-practical-workflow-for-differential-expression-and-visualization) - [Radiomics Feature Selection: Methods and Best Practices](/knowledge/bioinformatics/radiomics-feature-selection-methods-and-best-practices) - [Digital Pathology Guidelines: A Reference for Implementation](/knowledge/bioinformatics/digital-pathology-guidelines-a-reference-for-implementation) ## References and Further Reading - [NCBI Data Resources](https://www.ncbi.nlm.nih.gov/). National Center for Biotechnology Information. - [EMBL-EBI Training](https://www.ebi.ac.uk/training). European Bioinformatics Institute. - [Bioconductor](https://bioconductor.org/). Bioconductor Project. - [Galaxy Training Network](https://training.galaxyproject.org/). Galaxy Project. - [nf-core Documentation](https://nf-co.re/docs). nf-core. - [The Carpentries Lessons](https://carpentries.org/lessons). The Carpentries. - [NORMA: The Network Makeup Artist - A Web Tool for Network Annotation Visualization.](https://pubmed.ncbi.nlm.nih.gov/34171457). Genomics, proteomics & bioinformatics, 2022. - [CURTAIN-A unique web-based tool for exploration and sharing of MS-based proteomics data.](https://pubmed.ncbi.nlm.nih.gov/38324566). Proceedings of the National Academy of Sciences of the United States of America, 2024. - [AI and Machine Learning for Proteomics-Driven Drug Discovery: Methods, Tools, and Best Practices.](https://doi.org/10.3390/cimb48050532). 2026. - [Proteomic profiling in personalized nutrition: a systematic review and methodological frameworks of randomized controlled trials.](https://doi.org/10.3389/fnut.2026.1826381). 2026. - [Integrative multi-omics and network-based machine learning for early diagnosis of Parkinson's disease.](https://doi.org/10.1371/journal.pone.0329980). 2026. > This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.