ARCTIC-3D

Automatic Retrieval and ClusTering of Interfaces in Complexes from 3D structural information

WEB SERVER

ARCTIC-3D is available at this webserver https://wenmr.science.uu.nl/arctic3d/

ARCTIC-3D: all you want to know about protein-specific interfaces

ARCTIC-3D is a software for data-mining and clustering of protein interface information. It allows you to retrieve all the existing interface information for your desired protein from the PDBE graph database (https://www.ebi.ac.uk/pdbe/pdbe-kb/), grouping similar interfaces in interacting surfaces.

The software first checks your input (a uniprot ID, a FASTA file, or a PDB file), and then retrieves the existing interaction data from the graph API. Such interfaces are projected on a selected PDB structure and their dissimilarity is calculated, thus allowing for the application of a hierarchical clustering algorithm.

In output you will see how your favourite protein can display different binding surfaces, each one characterised by few residues that are always present (hotspots) and other amino acids which are at the interface only from time to time.

Developing

Check CONTRIBUTING.md for more information.

Installation

With conda

Clone the repository on your computer and navigate to it

git clone git@github.com:haddocking/arctic3d.git
cd arctic3d

Here you can create the arctic3d environment:

conda create -n arctic3d python=3.10
conda activate arctic3d
pip install .
arctic3d -h

To run BLAST locally

bash install_blast_deps.sh

And put blastp in your $PATH by adding the following line to your .bashrc or .bash_profile file:

export PATH="PATH_TO_YOUR_ARCTIC3D_INSTALLATION/src/ncbi-blast-2.15.0+/bin:$PATH"

Example usage

Please refer to the examples documentation page.

Detailed documentation

In order to generate a detailed html documentation please execute these commands

pip install myst_parser
pip install chardet
conda install sphinx
sphinx-build -E docs ./arctic3d-docs

Then you can open the file arctic3d-docs/index.html, which contains all the necessary documentation.

Result interpretation

After running ARCTIC-3D, results are stored in the output directory (default: arctic3d-{uniprot_id}/). Below is an explanation of each output file and how to interpret them.

Output files

File	Description
arctic3d.log	Log file with execution details and warnings
input_data/	Directory containing copies of input files
{pdb_id}_updated.cif	Structure file downloaded from PDBe (mmCIF format)
{pdb_id}-{chain}.pdb	Cleaned PDB structure used for analysis (renumbered to UniProt numbering)
retrieved_interfaces.out	All interfaces retrieved from PDBe, listing partner IDs and their residue lists
interface_matrix.txt	Pairwise dissimilarity values between all interfaces (used for clustering)
dendrogram_{linkage}.png	Hierarchical clustering dendrogram (e.g., dendrogram_average.png)
clustered_interfaces.out	Interfaces grouped into clusters (binding surfaces)
clustered_residues.out	Residues belonging to each cluster
clustered_residues_probs.out	Residues ranked by probability within each cluster
{pdb_id}-{chain}_cl{N}.pdb	PDB structure for cluster N with probabilities encoded in B-factor column
sequence_probability.html	Interactive bar plot of per-residue probabilities
sequence_probability.json	JSON data for the interactive plot

Understanding the clustering

ARCTIC-3D groups similar interfaces into binding surfaces (clusters). Two interfaces are considered similar when they overlap spatially on the protein surface. The dissimilarity is measured using the squared sine of the angle between interface vectors in a Hilbert space representation - values close to 0 indicate overlapping interfaces, while values close to 1 indicate completely distinct regions.

The interface_matrix.txt file contains the pairwise dissimilarity values in the format:

interface1 interface2 dissimilarity_value

Interpreting residue probabilities

The probability (or "contact probability score") represents the fraction of interfaces within a cluster where a residue is observed. It is calculated independently for each cluster:

probability = (number of interfaces containing the residue) / (total interfaces in cluster)

For each cluster, residues are assigned a probability value between 0 and 1:

• Probability = 1.0: The residue appears in every interface within the cluster (a "hotspot" residue)
• Probability = 0.5: The residue appears in half of the cluster's interfaces
• Probability close to 0: The residue rarely appears at this binding surface

Important: Probabilities do NOT sum to 1.0 across clusters for a given residue. A residue can have high probability in multiple clusters if it participates in different binding surfaces. For example, a residue with probability 0.8 in cluster 1 and 0.6 in cluster 2 means it appears in 80% of cluster 1's interfaces and 60% of cluster 2's interfaces.

The clustered_residues_probs.out file lists residues ranked by probability:

Cluster 1 : 15 residues
rank    resid   resname probability
1       42      ALA     1.000
2       45      GLU     0.875
...

Visualizing probabilities in PDB files

The output PDB files ({pdb_id}-{chain}_cl{N}.pdb) encode probabilities in the B-factor column:

• Cluster residues: B = 50 × (1 + probability), ranging from 50 (probability=0) to 100 (probability=1)
• Non-cluster residues: B = 0

This allows visualization in molecular viewers (PyMOL, ChimeraX, etc.) using a color spectrum where high B-factors (red) indicate hotspot residues and low values (blue) indicate residues not involved in that binding surface.

Interpreting the dendrogram

The dendrogram (dendrogram_average.png) shows the hierarchical relationship between all retrieved interfaces. The x-axis represents the dissimilarity between interfaces or groups. Interfaces that merge below the threshold (default: 0.866, corresponding to a 60° angle) form a single binding surface. The threshold can be adjusted with --threshold to obtain finer or coarser clustering.

Using the interactive plot

Open sequence_probability.html in a web browser to explore per-residue binding probabilities. Each cluster is shown as a separate colored bar series, allowing you to identify which residues are involved in which binding surface and compare hotspots across different clusters.

Citing us

If you used ARCTIC-3D in your work please cite the following publication:

Marco Giulini, Rodrigo V. Honorato, Jesús L. Rivera, and Alexandre MJJ Bonvin: "ARCTIC-3D: automatic retrieval and clustering of interfaces in complexes from 3D structural information." Communications Biology 7, no. 1 (2024): 49. (www.nature.com/articles/s42003-023-05718-w)