Structure Comparison

This repository contains the source code for "An interpretable alphabet for local protein structure search based on amino acid neighborhoods."

We suggest getting started with our example folder, 3Dn_example. This folder supplies the script 3Dn_example.ipynb, which demonstrates the process of generating a 3Dn alignment from two PDB files.

This repository contains the following folders:

• 3Dn_example, which contains the example script
• benchmarking, which contains infrastructure to benchmark a structure character alphabet and combine it with other alphabets (e.g. 3Di, 3Dn, AA/BLOSUM)
• blosum_construction, which details the construction of a 3Dn BLOSUM-like matrix using training alignments
• blurry_vector_construction, which analyzes local structure information of proteins to construct their blurry neighborhoods
• compare_and_interpret_alphabets, which shows comparison of 3Dn and 3Di alphabets
• data_processing, which details the handling and type conversion of data necessary for the training and testing processes
• dihedral_blosum, which shows the construction and training of a dihedral BLOSUM
• graph_clustering_alphabets, which shows the graph clustering processes used to cluster dihedral and blurry neighborhood alphabets
• nbhd_visualization, which allows for visualizing blurry neighborhoods
• vqvae_on_nhots, which shows the process of training blurry neighborhoods from n-hot vectors using a VQ-VAE

If you use this code, you are welcome to cite us.

@article {zerefa2025,
	author = {Zerefa, Saba and Cool, Jesse and Singh, Pramesh and Petti, Samantha},
	title = {An interpretable alphabet for local protein structure search based on amino acid neighborhoods},
	year = {2025},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2025/04/24/2025.04.21.649886},
	journal = {bioRxiv}
}