What does gene-level summarization mean?

Microarray data provides probe-level expression measurements and RNA-seq data provides exon-level or transcript-level (i.e. different isoforms of the same gene) expression measurements. However, current functional annotations are mainly assigned at the gene or protein level. Therefore, when multiple probes or transcripts are mapped to the same gene, they need to be summarized into a single value for that gene. At the Gene Annotation step, users can choose to use the averages or medians of multiple probe intensities (microarray), or sums of counts from multiple transcripts (RNA-seq) to perform gene-level summarization.

What is the rationale behind filtering gene expression data?

The purpose of filtering is to increase the statistical power of differential expression analysis be removing any genes that are less likely to be informative. Please refer to the paper Independent filtering increases detection power for high-throughput experiments for detailed discussion and benchmark tests

Low variance filter: genes whose expression values do not change across different samples, and thus have very low variance. Genes are ranked by their variance from low to high, and you can exclude a certain percentile of genes with the lowest variance by adjusting the "Variance filter" slider. The above referenced study has suggested that up to 50% genes can be removed based on their variance with improved results

Low abundance filter: genes with very low abundance are not measured relaibly and amy not be biologically important. You can exclude genes below a certain threshold by adjusting the "Low abundance" slider. The above referenced study has suggested 10% genes can be removed based on their abundance with improved results

How should I choose a suitable normalization procedure?

Normalizing the data accounts for systematic technical sources of variation so that biologically-driven changes in gene expression can be better detected between samples. A gene expression normalization method should be chosen unless the data has already been normalized, in which case the user should select "None".

All of the normalization methods available on NetworkAnalyst are well-established and have been used in many previous studies. They are based on slightly different assumptions about the underlaying distributions, but should produce relatively similar results. If you are concerned about significant differences between normalization methods, you can try out more than one and visualize the results using the provided plots.

Tip: if you are not sure whether the data is already log transformed or not, you can easily figure this out by visualizing the data (i.e. boxplot). For microarray data, log transformed data values are usually less than 16. For RNA-seq data with 1 million reads, log2(1,000,000) is less than 20. Therefore if all data values are all below 20, it is reasonable to assume that the data has already been log transformed.

How to detect and deal with potential outliers?

Potential outlier samples can be identified from PCA plots. The potential outlier will distinguish itself as the one located far away from the major clusters formed by the remaining samples. To deal with outliers, the first thing is to check if the sample was measured properly. In many cases, outliers are the result of operational errors during the analytical process. If those values cannot be corrected, the sample should be removed from the input data and the analysis re-started.

Which statistical method to use for differential analysis?

Limma is a popular method for differential analysis that was first developed for microarray differential analysis. It addresses the problem of low sample sizes typical to whole-transcriptome studies by using the whole expression profile to make more stable estimates of gene expression variance. EdgeR and DESeq2 were both developed to analyze RNAseq data. All three methods are well-established and should give similar results. Please note:

• EdgeR and DESeq2 are only designed for RNAseq data and will be disabled for microarray data.
• Due to high computational resources required, DESeq2 will be disabled when your dataset contain over 50 samples.

My data contains multiple metadata, how should I choose a proper method for differential analysis?

In differential expression analysis, you should first determine whether any of the metadata encode blocking factors, then decide on how to classify individual samples into groups, and finally decide which groups of samples should be compared to each other using statistical tests. Let's assume that none of your metadata are blocking factors (more on that later) and try to understand how selecting primary and secondary factors creates different groups of samples. Consider the "Estrogen" example data, generated in a study that measured gene expression at multiple time points in breast cancer cells in which the estrogen receptor (ER) was either present or absent. Here, the metadata are "ER" and "TIME". As the figure below shows, selecting "ER" as the primary factor divides the data into two groups because "ER" has two different levels ('present' and 'absent'). Selecting "TIME" as the secondary factor results in four groups because the two primary groups are split based on the two time points. If there were three time points, each primary group would be split into three groups, resulting in six groups overall.

The defined groups can now be compared to find genes that are differentially expressed between them (more details on this in later sections). In some experimental designs, we aren't interested in finding the genes that are differentially expressed between the groups defined by the secondary factor because it is a blocking factor. Examples of blocking factors are subject IDs when multiple samples were taken from the same subject (e.g. paired samples, multiple tissue types), or batches of samples that were measured at different times or in different locations. If you indicate that your secondary factor is a blocking factor, NetworkAnalyst will conduct comparisons within the groups that it defines, which typically improves the accuracy of the overall result.

I received the error message "no residual degrees of freedom", what should I do?

This means you do not have enough samples to perform the analysis you specified, usually when combining two metadata in an independent two-factor analysis (no blocking factors). In this case, the total number of groups will be the product of the number of levels in each metadata factor (i.e. if the primary metadata contains 3 levels, and the secondary metadata contains 4, the total number of groups will be 3 * 4 = 12). We recommend a minimum of 3 samples per group, therefore at least 36 samples are required in order to perform a 3 x 4 two-factor analysis.

In this case, you should focus on a single primary metadata and leave the seconday metadata as "Not available", and perform differential analysis with regard to individual metadata. You can then choose the other metadata as the primary metadata and perform the analysis again. If there are no or very few significant genes identified, it is most likely that incorporating the secondary metadata into the analysis will not affect the result.

What are the differences between pair-wise comparisons and time-series comparisons?

A pair-wise comparison tests for genes that are differentially expressed between any pair of groups. For example, take three groups A, B, and C. The "all pairwise" comparison will contrast A-B, A-C, and B-C. A time-series comparison will only contrast consecutive pairs of groups, so in our example only A-B and B-C. Time-series are commonly used when gene expression was measured at multiple time points, or after treatments with varying concentrations/durations.

What is a nested comparison?

A nested comparison allows you to determine which genes respond differently to a treatment condition, respective to some other metadata. For example, consider the experimental design described in the section on multiple metadata where cells with and without an ER were measured at 10hrs and at 48 hrs. To find the genes that respond differently over time in the ER vs. noER cells, you would perform a nested comparison. First, compare ER10-ER48 to find the genes that are differentially expressed in cells with an ER (ERgenes). Next, compare noER10-noER48 to find the genes differentially expressed in cells with no ER (noERgenes). Finally, to find the genes that respond differently over time in ER vs. noER cells, compare ERgenes-noERgenes.

Selecting "Interaction only" will return significant results from only the ERgenes-noERgenes contrast. Otherwise the full model is returned, which is the combination of significant genes from the ER10-ER48, noER10-noER48, and the ERgenes-noERgenes contrasts.

What is the difference between PCA and t-SNE?

PCA (principal component analysis) and t-SNE (t-distributed stochastic neighbor embedding) are both popular dimension reduction techniques. In PCA, each principal component is the linear combination of predictor variables that explains the greatest amount of variability in the outcome variable, after accounting for previously computed principal components. Unlike PCA, t-SNE utilizes random walks to estimate non-linear relationships between predictor variables for each sample. This means that each iteration of t-SNE will generate slightly different results.

How to use and interpret 3D PCA and t-SNE visualization?

The interactive PCA visualization summarizes all the data into the the first three principal components (PCs). Each data point in the Scores Plot represents a sample. Samples that are close together are more similar to each other. The colors of these data points are based on the factor labels. Users can change the colors according to any of the two factor labels.

Each data point in the Loadings Plot represents a feature. When scores and loadings plots are viewed from the identical perspective, the direction of separation on the scores plot can be explained by the corresponding features on the same directions - i.e. features on the two ends of the direction contribute more to the pattern of separation.

Which gene set libraries are available for enrichment analysis?

NetworkAnalyst supports enrichment analysis with gene sets from the Gene Ontology, PANTHER, KEGG, Reactome, and MSigDB databases. Note - not all gene set libraries are available for all species.

The GO:BP, GO:MF, and GO:CC gene sets include the complete set of Gene Ontology terms (> 45 000) for the biological process, molecular function, and cellular component categories. The PANTHER:BP, PANTHER:MF, and PANTHER:CC are reduced sets of GO terms ("GO slims") that have been manually chosen based on the PANTHER protein classification system. Briefly, the PANTHER project has created > 15 000 phylogenetic trees that encode the evolutionary relationships within protein families. Subsets of GO terms were chosen that best reflect the function gain or loss along the branches of the PANTHER trees for each of the BP, MF, and CC categories. In general, GO slims can simplify the interpretation of enrichment analysis results because they reduce the number of highly similar GO terms.

The KEGG and Reactome gene sets are networks of molecular interactions that represent biological pathways and processes. Reactome pathways are created through a process similar to scientific peer review, where different experts create and review the pathway organization, and all interactions contain references to the primary literature. KEGG pathways are also based on molecular interactions in the primary literature, but are accompanied by an extensive ortholog mapping that allows KEGG pathways to be rapidly extended to additional species based on genome sequence homology.

The Motif gene sets are based on shared upstream regulatory motifs (short nucleotide or amino acid pattern) that can function as potential transcription factor binding sites (source: MSigDB, set C3:TFT).

What is overrepresentation analysis (ORA)?

ORA is a statistical technique to identify gene sets or pathways that have a significant overlap with the selected genes of interest. In NetworkAnalyst, Hypergeometric tests are used to compute the p-values. The gene sets are described in the above FAQ on gene set libraries.

What is Gene Set Enrichment Analysis (GSEA)?

GSEA is a statistical method that determines whether a predefined gene set (GO, KEGG, etc) demonstrates statistically significant difference between two groups. Taking as input a list of ranked genes and a gene set, it looks at whether the genes from the gene set are randomly distributed in the ranked list or significantly enriched in the top and bottom extremes of the ranked list. In the following schema, the gene set A is significantly enriched, while gene set B represents a case where the genes are more randomly distributed. In contrast to ORA, GSEA can detect weakly coordinated changes of gene expression in sets of functionally related genes because it is not limited by the issue of losing information when setting a threshold.

How should I choose the ranking metric used before performing GSEA?

Ranking the list of genes to be analyzed by GSEA is a critical step that can greatly influence the result. Many ranking metrics are present in the literature and there is no consensus on which is best to use. NetworkAnalyst offers four different methods. Rank based on DE method used and Fold change are the most intuitive, ranking genes according to their p-values and fold changes with respect to the primary metadata factor. Moderated Welch's t-test (MWT) and signal-to-noise ratio (S2N) are two other metrics that have been found to perform well with a low computational load. MWT is a version of the t-test that allows for unequal variance between groups, and S2N is the difference between the mean expression divided by the sum of the expression standard deviation for two phenotype groups.

Please note the above gene ranking methods are not applicable to meta-analysis. Instead, the genes are ranked based on the summary statistic obtained from the previous meta-analysis (combine p-value, effect-size or direct merging). Results obtained from vote count can not be used to perform GSEA.

A recent publication compared different ranking metrics using 28 benchmark datasets and scored each one based on their sensitivity and false positive rate, summarized in the table below for the four metrics supported by NetworkAnalyst. While all metrics are widely accepted, you should choose based on how important the sensitivity/false positive rate is to your analysis. For more details on how the sensitivity and false positive rate were determined, refer to the original publication.

What is the enrichment score in GSEA?

The enrichment score is the main output of GSEA. It represents the number of genes in the gene set that are over-represented at the extremes on the ranked list (most up or down regulated). It is the maximum deviation from zero encountered during the random walk that goes through the ranked list.

Where are GSEA and ORA used in NetworkAnalyst?

GSEA requires an entire profile of gene expression values, and so it is only available after data processing and differential analysis of uploaded gene expression table(s) in the GSEA Enrichment Network and GSEA Heatmap Clustering tools. ORA is more flexible since it only requires a list of genes of interest. In addition to the stand-alone ORA Enrichment Network and ORA Heatmap Clustering tools, ORA can be performed on subsets of genes identified in volcano plots, network modules, sections of Venn diagrams and chord diagrams, and the focus view of any heatmap. The GSEA and ORA enrichment networks are described in more detail in the following FAQ section.

Is it possible to visualize the data from a single enriched gene set?

Yes, after you have performed functional enrichment analysis, the significant gene sets will be displayed in a table. By double clicking on a gene set name, all members will be displayed on the focus view (heatmap analysis), as highlighted node(s) within the current network (network analysis/enrichment network), as highlighted points in the volcano plot, or as highlighted chords in the chord diagram.

When are enrichment networks most appropriate?

Enrichment networks are a good way of visualizing the output from enrichment analysis when there are many significant results. Enriched gene sets are displayed in network form, where gene sets with overlapping genes are connected by edges. This groups functionally similar gene sets together, which can be easier to interpret than a list of enriched gene sets in tabular form. Enrichment networks are particularly useful for nested gene sets, such as in the Gene Ontology.

How do I interpret an enrichment network?

Gene sets are represented by the nodes that are automatically generated in the default view. The nodes are coloured according to their enrichment score (GSEA) or p-value (ORA) from the results table. The size of the node corresponds to the number of genes from that gene set that are on the analyzed gene list. The smaller nodes correspond to individual genes, and they are coloured according to their fold change. More details on how to manipulate the appearance of the network can be found in the "Network Visualization" section.

What are meta-nodes?

Gene set nodes are considered "meta-nodes" because double-clicking them reveals smaller nodes that correspond to the individual genes belonging to that gene set from the analyzed gene list. There will be an edge between the individual gene and any enriched gene set that they are a part of, so you can easily see the which genes are shared between sets. The hierarchical organization of meta-nodes allows users to customize the level of detail represented by an enrichment network.

What is a bipartite network?

A bipartite network displays nodes for all gene sets and individual genes. The same network could be generated from the default enrichment network view by double-clicking each gene set node. Bipartite networks are appropriate when there are a smaller number of enriched gene sets.

How are connections between gene sets determined?

There are two options for determining whether an edge is drawn between two gene sets. The overlap coefficient (OC) is calculated as the overlap of two gene sets divided by the size of the smaller set. The Jaccard index (JI) is the overlap of two gene sets divided by the size of their union. The JI is more applicable when gene sets have a relatively similar size, such as KEGG pathways or PANTHER GO slims. The OC is better at detecting parent-child relationships within hierarchically organized gene sets, such as the full Gene Ontology.

How to change the size of gene set nodes?

To increase the size of a gene set node, double-click on the name of the gene set in the results table on the right hand side. Each time the gene set name is double-clicked, the size will increase. Since the appearance of labels depends on the size of the node, this is a way too add labels to specific nodes in the network.

How to focus the enrichment network on specific gene sets?

If there are a subset of enriched gene sets that are of particular interest, you can visualize them separately from the rest of the network by extracting them. Select the gene sets from the "Result Table" panel and click the "Extract" button at the top left corner. If you want to see the detailed connections between a few gene sets (shared individual genes), this can be an effective way of simplifying the network so that these details are easier to visualize and interpret.

What is the goal of network construction?

The goal of network construction is to generate a clear visualization of the biological context of the genes of interest. This includes capturing relevant biological pathways and molecules (TFs, drugs, chemicals) that interact with the gene list, as well as important connections between them. For a small gene list (< 100), this is accomplished by adding a substantial number of interacting nodes from the underlaying databases. For a large gene list (> 1000), there are likely enough biological interactions within the uploaded list and so network construction is focused on pruning nodes and edges to reduce complexity to more effectively interpret the most critical connections.

How are networks generated from my data?

The networks are generated by first mapping the significant genes/proteins to the selected underlying database. A search algorithm is then performed to identify proteins that directly interact with the uploaded genes/proteins ("seeds"). The seeds and their interaction partners are returned to build the subnetworks.

This approach will typically return one giant subnetwork ("continent") with multiple smaller ones ("islands"). Most subsequent analysis is performed on the continent. Note, networks with less than 3 nodes will be excluded.

How do I integrate multiple databases from different interaction types?

To perform integration, select more than databases in "Network Selection" page. There exists two integration options

1. Union: create multi-modal networks by creating a network that is the union of first-order networks of selected databases.
2. Intersection: identify the share portion of multiple first-order networks. A useful use case is to integrate tissue specific coexpression with generic PPI.

How many nodes can be visualized (network size limit)?

The visualization is actually limited by the performance of users' computers and screen resolutions. Too many nodes will make the network too dense to visualize and the computer slow to respond. We recommend limiting the total number of nodes to between 200 ~ 2000 for the best experience. For very large networks, please make sure you have a decent computer equipped with a modern browser (we recommend the latest Google Chrome).

What if the input gene list is very small (< 100)?

When there are too few seed genes, the resulting network will be too simplified to identify themes in the biological context of you genes of interest. There are two main solutions here:

1. Expand your search of the underlaying database by using a Second-order Network;
2. Increase the input genes by using smaller fold change and/or larger p value cutoffs.

What if the input gene list is very large (> 1000)?

When there are a large number of significant genes or seed proteins, the resulting networks will be too large and complex for effective visualization and interpretation. There are four possible solutions here:

1. Reduce the networks using direct connections between seed proteins Zero-order Network;
2. Trim the networks to keep only seeds and their connecting nodes using Minimum Network or Steiner Forest Network;
3. Filter the networks using the Degree Filter or Betweenness Filter;
4. Reduce the input genes by using larger fold change and/or smaller p value cutoffs.

The above approaches aim to reduce the network size and complexity, and to retain the most relevant information for downstream functional analysis.

What are zero-order, first-order, and second-order networks?

The "order" refers to the type of relationships that will be used to extract nodes from the underlaying database. The default is a first-order network, which returns all seed genes and all nodes directly connected to them in the database. A second-order network increases the size because it returns all seed genes and all nodes that are within two connections in the database. Drawing a comparison to social networks, first-order networks return seed genes and their "friends", while second-order networks return the seed genes, their "friends", and the "friends of their friends".

A zero-order network can reduce the number of seed genes because it retains only genes that are connected to each other within the underlaying database. This can help simplify your gene list to highlight the biological theme of the database of interest (i.e. protein-protein interactions, TF-gene interactions etc).

How do I simplify a dense network?

Both the Minimum Network and Steiner Forest Network tools aim to construct a minimally connected network that contains all of the seed genes. This means that the only added nodes are ones that connect previously disjointed networks of seed genes. The difference between the minimum network and the Steiner forest network is the way in which the approximate solution is computed. For the minimum network, NetworkAnalyst implements an approximate approach based on shortest paths: we compute pair-wise shortest paths between all seed nodes, and remove the nodes that are not on the shortest paths. For the Steiner forest network, NetworkAnalyst implements a fast heuristic prize-collecting Steiner forest algorithm.

When should I use the degree or betweenness filters?

The degree and betweenness filters allow you to reduce the size of the network based on its connectivity alone (see later FAQ sections for explanations of "degree" and "betweenness"). The key takeaway is that the degree filter tends to retain hub genes (genes with many connections to other genes), and the betweenness filter tends to retain genes that connect dense clusters of genes.

Can I manually exclude specific nodes from the network?

Yes, there are two main ways to exclude specific nodes from the network. A list of nodes can be uploaded using the Batch Exclusion tool and the network will be re-computed without these nodes. Alternatively, you can delete nodes manually using the Delete button at the top of the Node Table. See the Network Visualization FAQ section for more details on deleting nodes manually.

Can I undo network construction steps?

Clicking the Network Reset button will return the network to the default first-order network.

What are protein-protein interaction networks?

Protein-protein interactions (PPI) include many types of relationship between proteins, including physical associations as parts of molecular complexes, information-transfer associations in signaling pathways, and computationally predicted functional associations based on shared membership in densely connected network modules. A PPI network summarizes these types of interactions in graphical form.

What are the differences between the generic PPI databases?

The IMEx Interactome PPI data come from InnateDB, a database aimed at facilitating systems-level analysis of the mammalian innate immune system by annotating the relationships between biological pathways and molecules related to the innate immune system. All interactions are manually curated from the literature according to the International Molecular Exchange Consortium (IMEx) standards.

The STRING Interactome integrate PPI interaction data from many sources, including using direct (physical associations from experimental data) and indirect (functional associations based on computational predictions) evidence, for over 2000 species. The key distinguishing factor of the STRING project is that they assign a confidence score to each interaction, with interactions with more evidence scoring higher. The "Confidence score cut-off" can be adjusted to restrict addition of PPIs below the specified value from being added to your network. Checking the "Require experimental evidence" box will exclude PPIs that are supported by computational predictions only. The data were downloaded from the STRING database (version 10).

The Rolland Interactome PPI data are a collection of human binary PPIs from the literature in 7 public databases. Binary PPIs refer to direct physical interactions between proteins. To produce the Rolland Interactome, 33 000 binary human PPIs were collected from the literature. Of these, the 11 045 with multiple supporting studies were retained.

What are the advantages of tissue-specific PPI networks?

Using tissue-specific PPI networks gives the option of focusing on tissue-specific processes and phenotypes. The tissue-specific PPI data is from DifferentialNet and was produced by integrating experimental binary PPI data with RNA-sequencing profiles from different tissues, collected by the Genotype-Tissue Expression consortium. Each PPI was given a score for each tissue that indicates whether the corresponding genes were similarly expressed across many tissues, or significantly dysregulated only in the tissue of interest.

How to use the filter for tissue-specific PPI networks?

The filter can be adjusted to change how unique the PPI should be to the tissue of interest. A lower score will filter out more PPIs, so the resulting PPIs will be highly unique to the selected tissue. See the latest DifferentialNet publication for more details on the scoring metric.

What are the differences between the types of gene regulatory networks?

The Gene-miRNA Interactions rely on the TarBase database, which is a collection of experimentally supported miRNA targets. This means that miRNAs are returned that interact with the uploaded seed genes. The TF-gene Interactions have three different database options (see next FAQ for more details), all of which return genes that function as transcription factors for the uploaded genes of interest. Finally, the TF-miRNA Coregulatory Network draws from the RegNetwork, which contains TF-TF, TF-gene, TF-miRNA, miRNA-TF, miRNA-gene binding interactions for human and mouse.

For all three of these network types, the returned nodes (miRNAs or TFs) only have connections to the uploaded seed genes, not to each other. This gives these networks a characteristic appearance where the seed genes have connections to many regulatory elements, while the regulatory molecules are only connected to a few seed genes. miRNA nodes are represented as squares instead of circles.

What are the differences between the TF-gene interactions databases?

The ENCODE TF-gene interactions are inferred from ENCODE ChIP-seq data using the BETA algorithm. BETA integrates factor binding and differential expression analysis to predict whether a TF has an activating or a repressing effect, to infer the gene targets, and to identify the binding motif. JASPAR uses a collection of position frequency matrices to predict transcription factor binding sites on the DNA. ChEA collected ChIP-X (includes ChIP-chip, ChIP-seq, ChIP-PET, and DamID) data from the literature to describe the binding of TFs to target genes in mammalian species.

Where do the "Diseases, drugs, and chemicals" interactions come from?

The Protein-drug Interactions come from DrugBank, a database that combines bioinformatics and cheminformatics data on drugs and drug targets. The Protein-chemical Interactions come from the Comparative Toxicogenomics Database, which contains curated interactions between chemicals and genes from the literature. The Gene-disease Associations come from DisGeNET, which integrates data from expert curated repositories, GWAS studies, multiple species, and the literature. As in the gene regulatory networks, the added nodes are connected to seed genes but not to each other, and are represented as squares instead of as circles.

What are gene co-expression networks and how are they different from PPI networks?

Gene co-expression networks are constructed by measuring the similarity (i.e. correlation) in pairwise gene expression in profiles across many conditions. Two genes are connected to each other in the network if they tend to respond similarly (consistently up or down regulated together) to perturbations. Some PPI databases include gene co-expression data, along with other types of evidence, to define interactions. Since co-expression networks can be computed from expression data alone, it is easier to generate separate ones for many different tissues and even cell types compared to PPI networks.

How do I identify which nodes are important based on their position within the network?

A basic assumption is that changes in nodes that occupy key positions within a network will have a greater impact on the overall network structure than changes in relatively isolated positions. In graph theory, measures of centrality are used to identify the most important nodes. NetworkAnalyst provides two well-established node centrality measures - degree and betweenness. The degree of a node is the number of connections it has to other nodes. Nodes with a high degree act as hubs within the network. The betweenness of a node is the number of paths that pass through it when considering the pairwise shortest paths between all nodes in the network. A node that occurs between two dense clusters will have a high betweenness, even if it has a low degree. Note, you can sort the node table based on either degree or betweenness values by double clicking the corresponding column header.

What is a module and how are modules identified in NetworkAnalyst?

Modules are tightly clustered subnetworks with more internal connections than expected randomly in the whole network. They are considered as to be relatively independent components in a graph. Members within a module are likely to work collectively to perform a biological function. The biological functions of a module can be explored using enrichment analysis.

NetworkAnalyst currently offers three different approaches for module detection - the WalkTrap, InfoMap, and Label Propagation algorithms. The general idea behind the Walktrap Algorithm is that if you perform random walks on a graph, a higher number of walks are more likely to stay within a group of nodes that are highly connected to each other because there are only a few edges that lead outside of them. The Walktrap algorithm runs many short random walks and uses the results to detect small modules, and then merge separate smaller modules in a bottom-up manner. The InfoMap Algorithm is also based on random walks, which it uses to minimize the hierarchical map equation for different partitions of the network into modules. The Label Propagation Algorithm works by randomly assigning a unique label to every node. On each iteration, node labels are updated to match the one that the maximum of its neighbours has. The algorithm converges when each node has the same label as the majority of its neighbours.

NetworkAnalyst also integrates the gene expression values as edge weights during module searches. Weights are calculated as the square of the mean absolute log fold changes of the two adjacent nodes. Larger weights mean closer connections during random walks. To avoid zero-weight errors for non-seed proteins during program run, pseudo-expression values are given to non-seed proteins of 1/10 of the minimal absolute log fold changes of the seed proteins. By giving larger weights to seed proteins, the program encourages detecting modules containing more seed proteins (shorter distances).

How do I interpret the p-value of a module?

The p-value of a module is based solely on network connectivity, and gives some indication of how significant the connections within a defined module are. Let's call the edges within a module "internal" and the edges connecting the nodes of a module with the rest of the graph "external". The null hypothesis of the test is that there is no difference between the number of "internal" and "external" connections to a given node in the module. The p-value of a given module is calculated using a Wilcoxon rank-sum test of the "internal" and "external" degrees. Users should also consider whether the modules are 'active' under the experimental conditions, by taking into account the number of seed proteins, their average fold changes, as well as the enriched functions displayed in the Module Explorer table.

Can I perform enrichment analysis on my query genes alone?

Yes, you can test enriched gene sets or pathways for only your query genes. To do so, first select the check-box in the top left of the Node Explorer toolbar. This will highlight all of your seed genes. Next, go to the Function Explorer toolbar and change the query to "Highlighted nodes". Select the gene set library of interest and click "Submit".

Can I perform enrichment tests on only the nodes highlighted in the network?

Yes. Users can perform enrichment tests on currently highlighted nodes in the network.

• Module highlight (automatic): first perform module detection, then click on a module;
• Module highlight (manual): set Scope to "including dependents", double click a node in the network to highlight the node together with its direct neighbours, and repeat the process to select more nodes;
• Node highlight (automatic): use Hub Highlighting or Data Highlighting to select nodes based on degree or betweenness values;
• Node highlight (manual): select nodes from the node table on the left or by double clicking on a node (Single Mode).

After you have selected the nodes or modules, click the Perform Enrichment Analysis button. The result table will be displayed in the panel below. Note, enrichment analyses are performed on ALL currently highlighted nodes. To ensure only your current selections are being used, first Reset the network, then perform highlighting/selections before performing the enrichment analysis.

How is enrichment analysis performed in the network viewer?

The enrichment analysis tests whether there is a significant overlap between the selected genes/proteins and the user selected library of pre-defined gene sets/pathways (ORA). NetworkAnalyst's network viewer uses hypergeometric tests to compute the enrichment p-values.

How do I interpret the node colors and sizes in the default Topo View?

In the default network generated by NetworkAnalyst, the size of the nodes are based on their degree values, with a big size for large degree values. The color of nodes are proportional to their betweenness centrality values. When user switches to Expression View, the color will be based on their expression values (if available).

Can I view my queries in the network?

Yes, to view your query genes or proteins, use the color palette on the top-left corner of the network viewer to set a highlight color. From the "Display Options" on the top right panel, click the "Highlight". Select "Upregulated nodes" or "Downregulated nodes", then click Submit button. You may also want to increase their node sizes by using the Size function under Node Options. Nodes will be labeled automatically when their size increase above a certain level.

How can I create a 300 dpi high-resolution network for publication?

Please use the Download option and choose "SVG Format" to save the current network view (tested using Chrome or FireFox, known issue with Safari). SVG is a vector based graphic format and you can then export it into any resolution static image (i.e. png) using a suitable graphic tool, for example, Adobe Illustrator or the free tool InkScape. Note, it is best to save SVG in white background, as the default background color in InkScape is in white. If your SVG is saved in Black background, after opening the SVG in InkScape, set the Background color to black (hex code: #222222) using the Document Properties menu.

Can I change the background color of the network?

Yes. To switch background color, click the pull-down menu next to Background on the toolbar at the top of the screen. From the dropdown menu list, select either White, Black or Custom. Selecting custom will prompts a dialog in which you can choose the color you want.

Can I change node color, size, or shape?

You can change the color and size of a node. The shape cannot be changed in the current implementation. To change the node color, choose the color using the Color Palette and then double-click the node you want to change. The node color will be changed to your specification. You can also change the whole color spectrum of the network. Click the Node dropdown menu located on the top toolbar and click on the Color option. A pop-up dialog will appear in which you are free to choose among the selection of color spectrums. To change the node size, you can keep double-clicking it to increase its size. You can also use the Node Size functions to increase or decrease the node size.

Can I change the position of a node?

Yes. You can simply put your mouse cursor over the node. When its label shows up, left click and drag the node to a position. Release the mouse.

Can I change the position of a cluster of nodes?

Yes. First use the Scope option on the top menu bar and make sure that the option including dependents is selected. Then drag a central node to a new position, and all nodes connected to this one will be moved as well. If you also want to adjust the position of other nodes, switch the Scope to "Current node", and then drag these nodes individually to a new position.

How can I label the nodes in the network?

Nodes will be automatically labeled when their sizes reach a certain threshold. Therefore, you can simply increase node size to label any node. To label a single node, right click the node of interest and click on the "Add Label" option in the context menu. The size of the node will increase so that the label appears. To label all highlighted nodes, use the "Node" tab in the Display Options panel on the top right, select "Highlighted nodes" and "Increase ++", then keep clicking the "Submit" button to increase the size until labels show up. To label all nodes in the network, perform the same steps as above, but choose "All nodes" instead of "Highlighted nodes".

Can I hide all the node labels in the network?

Yes, it is possible to hide them. Click on the Nodes dropdown list located on the top menu and select label option. Click on the display tab and select the "Hide" option.

Can I delete nodes from the network?

Yes. You can delete nodes and their associated edges from the current network. First you need to select the nodes from the Node Table in the left pane. Then click the Delete button at the top of the node table. A confirmation dialog will appear asking if you really want to delete these nodes. Note, this action will trigger network re-arrangement, especially if hub nodes are removed. In addition, other nodes that are no longer connected to the larger subnetwork after node deletion will also be removed during re-arrangement.

How do I manually highlight any arbitrary sections of the network?

There are two basic steps in the network highlighting - setting the highlight color and selecting the nodes to highlight. Use the Color Palette to set the color for the next selection. You also need to choose the scope for node selection:

• Current node: highlight only the selected node;
• Including-dependents: highlight the selected node and its direct neighbours.

Now, double click on nodes to make your selections. Note, you can repeat the steps above to change colors and scope to make different effects.

Can I extract a module or a highlighted section from the network?

Yes. To do this, first select or highlight section of the network, then click the Extract icon on the left tool bar in the network view window. Note, the operation is computationally expensive, so you will have to wait for ~20 seconds for the extracted network to return. The returned network will be named as "moduleX" and is available in the "Network Explorer" panel on the top-left of the page for future reference.

How do I proceed to 3D and/or VR view of the current network?

To view the current network in 3D, click on 3D button located in the toolbar located at the top left corner of the network viewer. To view the network in VR, please make sure that you are in 3D view and click on VR button located in the same toolbar. Make sure that you have a VR device connected to your computer.

What is a meta-analysis?

Meta-analysis is a type of statistical technique used to integrate multiple independent datasets that have been collected under similar experimental conditions, in order to obtain more robust biomarkers. By combining multiple data sets, the approach can increase statistical power (more samples) and reduce potential bias.

A key concept in meta-analysis is that it is generally not advisable to directly combine different independent datasets (i.e. merge them into a single large table) and analyze them as a single unit. This is due to potential batch effects associated with each datasets, which can completely overwhelm the biological effects. This issue has been well-studied in microarray experiments generated from different platforms.

Instead, meta-analysis is usually computed based on summary statistics (p values, effect sizes, etc.) to identify robust biomarkers. The meta-analysis module in NetworkAnalyst was developed to support these approaches, at both the individual gene and the gene set levels. Heatmaps and other visualization tools can used to explore patterns across different studies.

Which data sets are appropriate for meta-analysis?

Here are some basic rules for data collection for meta-analysis in NetworkAnalyst:

1. The data sets should have been collected under comparable experimental conditions, and/or the underlying experiments share the same hypothesis or have the same mechanistic underpinnings;
2. Only two-group comparisons are supported at the moment (i.e. control vs. treatment);
3. These datasets must share the same type of IDs so that the majority of the features overlap;
4. It is best to keep all data on the same scale or range (i.e. both raw or normalized in the same way). It is generally preferred to compare datasets on a log scale.

What should I do if my data fails the integrity check?

A common reason for the integrity check to fail is that the meta-data classes and comparisons are not consistent across the different datasets. To check if this is the problem, click View under "Data Summary" for each dataset. Make sure that the "Set order of comparison" is the same, and that the spelling and capitalization are consistent. If there are small differences, these can be corrected in the "Annotation" section - click Annotate next to the dataset you want to modify and edit the "Group labels".

How do I minimize study specific, platform specific bias of my gene expression datasets?

To normalize the gene expression values across datasets, we recommend the use of ComBat algorithm. At the quality check page, after datasets upload, you can first visualize the PCA clustering of samples from different datasets. If obvious batch effects are observed, select the checkbox located below the summary table to perform ComBat.

How does ComBat work for batch effect adjustment?

The method uses an empirical Bayes approach for adjusting batch effects in microarray and RNA-seq expression data. The algorithm can be summarized in three main steps:

1. Genes are standardized to have similar overall mean and variance;
2. Information is pooled across genes from a batch to estimate batch effects (increased level of expression, high variability, etc.);
3. The estimated batch effects are used to normalize the data to make them more comparable to each other.

For more information, refer to the original publication: Adjusting batch effects in microarray expression data using empirical Bayes methods

How do I use a Q-Q plot to choose between REM and FEM?

Q-Q plots are graphical tools to determine whether the assumption that the data came from a specific parametric distribution is plausible. They are generated by calculating evenly spaced percentiles from the sample data, and from the distribution of interest, and plotting the theoretical and actual percentiles against each other in a scatter plot. If the points fall on the diagonal straight line, it is reasonable to assume that the data came from that type of distribution.

The FEM assumes a chi-squared distribution, and NetworkAnalyst supports a Q-Q plot to check the validity of this assumption. The data will rarely fall perfectly on the straight line, even when they are randomly sampled from a known distribution, so you should look for large deviations such as the Q-Q plot below. In this case, since there is a significant deviation, the REM may be more appropriate since it does not assume any parametric distribution.

What should I do if I have > 2000 genes but want to make a chord diagram?

The number of differentially expressed genes can be reduced by making the significant threshold more stringent. This can be done when statistics are combined for the actual meta-analysis itself, and back at the "Data upload" step. Here, the p-value cut-off can be adjusted for the differential analysis of each data set by clicking Analyze under the "DE Analysis" step.

Can I perform enrichment analysis on specific sections of chord and Venn diagrams?

Yes, it is possible to perform enrichment analysis (ORA) on parts of these diagrams. For Venn diagrams, portions of diagram will be highlighted after you click them. Multiple sections can be highlighted at once, and the union of the genes from the highlighted sections are listed under "Gene List View" on the bottom left. The "Enrichment Analysis" tool performs ORA on this list of genes. For chord diagrams, start by selecting a specific colour that corresponds to the significant results from either an individual dataset or the the meta-analysis. The options under "Enrichment Analysis" will change depending on the section of the chord diagram that's been selected. You can perform ORA on either the genes unique to the selected dataset, or on the genes belonging to the intersection of the selected dataset with another dataset.

What if I want to visualize the overlap of > 4 lists that have > 2000 genes?

Venn diagrams are limited to 4 or fewer gene lists, and while chord diagrams have no limitation on the number of lists, there must be fewer than 2000 genes. This limitation can be overcome by uploading the genes of interest from each dataset you are comparing as multiple gene lists in the "Gene List Input" from the NetworkAnalyst home page and visualizing them with the "Heatmap View". This heatmap colours genes according to the number of lists they are a part of. A gene is represented by a grey square if it is not a member of that list. There is currently no limit to the number of genes or lists that can be visualized in this way.