Comparative Analysis of Protein Structure Prediction Tools

This repository contains code and data related to the preprint "Balancing Speed and Precision in Protein Folding: A Comparison of AlphaFold2, ESMFold, and OmegaFold" that has been adapted from of Anna Hyskova's thesis at Masaryk University in Brno.

The folder dataset_creation contains two Jupyter notebooks for creating the dataset:

• creating_dataset.ipynb describes the process of selecting suitable protein structures from PDB
• protein_loading.ipynb contains code for downloading files necessary for further work and other preprocessing steps

The complete dataset is available on Hugging Face.

Notebooks for the prediction of protein structures are in the directory protein_structure_prediction, which also contains an example script alphafold_prediction_example_script.sh used for the protein structure prediction by AlphaFold2 on Metacentrum OnDemand.

Materials related to the evaluation of predicted structures are in the folder evaluation:

• removing_complexes.ipynb removes redundant protein chains labelled as "complex", chains selected for evaluation are in the file chains_filtered.csv
• evaluation.ipynb describes the process of counting selected metrics (average pLDDT, RMSD, TM-score) for all predictions, the results are saved as chains_evaluation.csv
• badly_predicted_proteins.csv visualizes selected structures and explores the question of whether there is an overlap between protein chains with low-quality predictions generated by different tools
• statistics.Rmd is an R Markdown file for statistical analysis of the data and generation of graphs

The last part of the analysis focused on exploring the relationship between the prediction quality and various characteristics of the proteins. The code and data are in the folder evaluation_families:

• annotations.ipynb is a Jupyter notebook that obtains the annotations for protein chains and protein families
• protein_families.Rmd tests the significance of the relationships and produces visualisations
• data is a folder containing all the outputted results

If you find this work useful, you can cite us:

@article{hyskova2025balancing,
  title={Balancing Speed and Precision in Protein Folding: A Comparison of AlphaFold2, ESMFold, and OmegaFold},
  author={Hyskova, Anna and Marsalkova, Eva and Simecek, Petr},
  journal={bioRxiv},
  pages={2025--06},
  year={2025},
  publisher={Cold Spring Harbor Laboratory}
}