12 STRING (web)
12.1 
STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) is a resource for exploring protein-protein interaction (PPI) networks. It combines experimental data, predictions, and curated information to build networks that highlight functional relationships, helping to reveal shared pathways or biological processes within gene or protein lists.
12.1.1 Steps to Perform ORA in STRING:
- Select Multiple proteins tab.
- Input Gene List: Paste your prepared gene list directly into the input box on the STRING web page or upload a file containing your list.
- Select Organism: Choose the appropriate organism from the Organisms dropdown menu (e.g., Homo sapiens for human data). STRING would auto-detect the organism if ENSEMBL IDs provided.
- Modify Settings: Under Advanced Settings, you can modify Required score from low (0.15) to highest (0.9) confidence. Similarly FDR stringency and Network type can be selected.
NOTE: In cases where long list of features is provided, STRING may change some of its settings so that:
- the nodes will have a simplified (not 3D) design
- previews of protein structures are not shown
- the network edges show interaction confidence only
12.1.2 Browse the STRING ORA Results
STRING generates multiple tabs as output, shown here:
Figure 12.1: Results tabs in STRING
12.1.2.1 Viewers
Under the Viewers tab, various visualisation layouts are available, with the Network option being the most notable and widely used.
12.1.2.2 Legend
The Legend tab offers a guide to the colors of nodes and edges, along with annotations for each individual query in the input list.
12.1.2.3 Settings
In the Settings tab of the STRING results, users have the flexibility to adjust existing settings and apply new filters to customise their data view and analysis. This tab you to switch between network types, strengths, data sources, interaction scores and more.
12.1.2.4 Analysis
One of the most essential tabs is the Analysis tab, which offers comprehensive functional enrichment analysis from a range of databases. These include Gene Ontology (GO) for biological processes, molecular functions, and cellular components; Pathway enrichment from sources such as KEGG, Reactome, and WikiPathways; and other significant data sources such as Human Phenotype annotations and UniProt for protein function and structure.
Columns of the STRING enrichment table are explained as following:
- Count In Network: The first number indicates how many proteins in your network are annotated with a particular term. The second number indicates how many proteins in total (in your network and in the background) have this term assigned. You can click on the numbers to see the network view of the gene sets behind them.
- Strength: Log10(observed / expected). This measure describes how large the enrichment effect is. It’s the ratio between i) the number of proteins in your network that are annotated with a term and ii) the number of proteins that we expect to be annotated with this term in a random network of the same size.
- Signal: The signal is defined as a weighted harmonic mean between the observed/expected ratio and -log(FDR). FDR tends to emphasise larger terms due to their potential for achieving lower p-values, while the observed/expected ratio highlights smaller terms, which have a high foreground to background ratio but cannot achieve low FDR values due to their size. The signal measure seeks to balance both metrics for a more intuitive ordering of enriched terms.
- False Discovery Rate: This measure describes how significant the enrichment is. Shown are p-values corrected for multiple testing within each category using the Benjamini–Hochberg procedure.
STRING visualises terms within each category using a bubble plot, effectively showcasing the significance and size of enriched terms. Additionally, it renders groups of related terms based on a user-defined similarity level, allowing users to identify clusters of functionally related terms within the data. This helps in interpreting complex enrichment results and highlighting key biological processes or pathways that are closely associated.
Towards the bottom of the Analysis page, one can change the background including adding one of their own.
Finally the enriched terms can be downloaded at the end of the Analysis page, either individually per category or all enriched terms together.
12.1.2.5 Exports
The network data can be exported with the Exports tab. Also Network data can be directly sent to Cytoscape 
for further networking. It is expected to have Cytoscape installed before exporting to it.
12.1.2.6 Clusters
The Clusters tab essentially provides three different types of clustering algorithms:
k-means clustering: Initialises k centroids randomly, assigns each data point to the nearest centroid, recomputes the centroids as the mean of all points in a cluster until centroids do not change significantly.
MCL clustering (Markov clustering): is a graph-based algorithm that uses flow simulation to detect clusters in a network by modelling random walks.
DBSCAN clustering: is a density-based algorithm that groups points closely packed together while marking points in low-density regions as outliers or noise
Clusters can be downloaded in .tsv format.
12.1.3 Steps to Perform GSEA in STRING:
- Select Proteins with Values/Ranks.
- Input Gene List: Paste your gene list with a meaningful value for ranking (fold-change, log-pvalue, abundance, …) directly into the input box on the STRING web page or upload a file containing your list of features and their corresponding values.
- Select Organism: Same as above.
- Advanced Setting: FDR stringency and the initial sort order can be set up in advance and hit the Search.
12.1.4 Browse the STRING GSEA Results
The output differs from ORA. For each gene set, the results include the enrichment score, its direction within the ranked list, the number of overlapping features with the gene set, and the associated FDR.
When a user selects a gene set from the enriched table,
the associated genes are displayed within the ranking list. A table showing these genes along with their original ranking values is also provided.
Additionally, the locations of the corresponding proteins are highlighted in the proteome network:
A Functional enrichment visualisation (similar to that of ORA) is provided at below the enriched tables.
Modify Enrichment display settings tab before downloading the enriched tables. It is recommended to merge terms with a certain level of similarity to reduce redundancy, especially if there are many overlapping terms.
Here is an example output of GSEA on STRING from a previous run (the link will expire in future).