Rkmeans

Rkmeans is a small interactive R/Shiny application for structural clustering of molecular files (.cif or .pdb) based on RMSD and k-means clustering.

This app was born from the need to cluster AlphaFold 3 output structures, especially for R users. At the time, no simple and direct online tool allowed structural clustering of .cif files.
This tool is best suited for clustering a small number of predicted structures (e.g., 5–100 models), adjusting the number of clusters (k) accordingly.
For example:

• If you input 10 structures, trying k = 2–4 is sensible.
• For 30 structures, you might test k = 3–6.
• For 50+ structures, k = 10 may be appropriate.

---

Features

• Load multiple .pdb or .cif files from a folder.
• Automatically convert AlphaFold .cif files to .pdb (via Python and Biopython).
• Calculate pairwise RMSD using alpha carbons (Cα).
• Run multidimensional scaling (MDS) and k-means clustering.
• Visualize 3D structures with r3dmol directly in the app.
• Show cluster details (size, representative structure).
• Save cluster representatives.

---

Requirements

R packages

Install these packages if you don’t have them already:

install.packages(c(
  "shiny", "shinyFiles", "bio3d", "r3dmol", "tools", 
  "plotly", "dplyr", "readr", "tibble", "ggplot2"
))

(Optional) Python + Biopython

If you want to use the 3D structure viewer with .cif files (especially those downloaded from AlphaFold), you'll need Python and Biopython installed. This is only required for visualization — not for clustering.

pip install biopython

Make sure Python is installed and accessible from R via:

Sys.which("python")

If you're only interested in the clustering results and don't need structure visualization, you can skip this step.

---

How to Use

1. Clone this repository or download the app files.
2. Place your .cif or .pdb files in a folder (e.g., structures/).
3. Launch the app in R:

shiny::runApp("path_to_app_folder")

4. In the app:
   • Select a folder containing .cif or .pdb files.
   • The app will align the structures, calculate the RMSD matrix, and perform k-means clustering.
   • You can choose the number of clusters (k) manually.
   • Once clustering is complete, you'll see:
     • A summary table showing cluster sizes, within-cluster RMSD, and medoid filenames.
     • A 3D scatter plot of the structures in reduced dimensions (via MDS).
     • Optional 3D viewer for inspecting any selected structure.

---

Just a Simple App

This is not a polished piece of software — just a tool I made for myself to cluster structural models. Sharing it here in case it helps others using R for protein analysis.

---

License

MIT License.

---

👤 Author

Ivan Sanchis, PhD
Laboratorio de Péptidos Bioactivos
Facultad de Bioquímica y Ciencias Biológicas
Universidad Nacional del Litoral
Santa Fe, Argentina
📧 sanchisivan@gmail.com / sanchisivan@fbcb.unl.edu.ar