Welcome to ZEPPI

Overview

ZEPPI (Z-score Evaluation of Protein-Protein Interfaces) is a method designed for proteome-scale sequence-based evaluation of protein-protein interfaces as defined by 3D atomic-level models of protein-protein interaction (PPI) direct binary complexes. This method was first introduced in this paper.

Structural models for the PPI complexes may be derived from a number of sources, including:

• Experimentally determined structure in The Protein Data Bank (PDB)
• Template-based modeling as in PrePPI
• Protein docking studies
• Deep learning algorithms such as those based on AlphaFold Multimer.

ZEPPI addresses the evaluation challenge by capitalizing on sequence co-evolution and conservation of the residues that come into interfacial contact. The hallmark of ZEPPI lies in its computation of the interface Z-score, the ZEPPI score, achieved through the comparison of interface-derived contact metrics against metrics derived from randomly selected surface residues. Thus, ZEPPI negates the need for factoring in indirect interactions, as the structural model delineates the interacting residues.

This repository contains code for ZEPPI and the needed input data for the tutorial examples. ZEPPI can be run either on a PC for a small number of protein-protein structural models or, for larger-scale explorations, as batched jobs on SGE- or SLURM-equipped clusters.

Setup

ZEPPI is implemented in Python 3 and requires the following modules: biopython, numpy, numba, scipy, pandas, and the below published packages:

• Surfv (github) to calculate solvent accessible surface area for protein structures
• HMMER to make multiple sequence alignments for sequence homologs of a query sequence
• HH-suite to search for protein sequences similar to a query sequence in protein sequence databases

The installation time varies but typically should not exceed 30 minutes. This program has been tested on MacOS 12.6.7 and 13.4.1, Springdale Linux 7.9 (Verona), and Rocky Linux 8.5 (Green Obsidian) with bash-4.4.20, python-3.9.7, numpy-1.23.4, biopython-1.79, numba-0.56.2, scipy-1.9.1, and pandas-1.5.0.

Run ZEPPI with examples

Step 1. Download ZEPPI

Download the zipped GitHub codebase and unzip it. To run the tutorial examples, make sure the above-mentioned Python modules are installed on your machine.

Step 2. Run ZEPPI

Go to the unzipped folder, and configure your directory path and python path in the Run_ZEPPI.sh file. Run the command with bash to see the needed arguments and available options.

bash Run_ZEPPI.sh

Usage: bash Run_ZEPPI.sh PPI_list.csv Output.csv -option
where:
    PPI_list  A csv file containing your query PPIs.
    -m  calculate ZEPPI on mutual information & conservation (recommended for heterodimers).
    -md calculate ZEPPI on mutual information & conservation and direct coupling analysis (recommended for homodimers).
Output files will be named after the input file with different endings.

The parameters are:

• PPI_list is the input csv file containing the PPI name, chain names, and UniProt identifiers.
• Output is output ZEPPI metrics calculated for the input PPI list.
• -m is the option to calculate ZEPPI based on mutual information and conservation (recommended for heterodimers).
• -md is the option to calculate ZEPPI based on mutual information, conservation, and direct coupling analysis (recommended for homodimers; memory-consuming).
• Output files will be named after the PPI_list file with different endings.

As a demonstration, the Demo folder provides the input files for the example PPI_list that contains ten bacterial PDB dimer structures. Try to test ZEPPI in the Scratch folder with the below command. The expected running time is ~2 min for the -m option or ~8 min for the -md option. The expected output files of both option -m or -md are provided in the Demo folder. Detailed score files are in the Demo/Metrics folder. To run ZEPPI for your own structure models, the required input files are described in the Run_ZEPPI.sh file and the needed codes are stored in the Methods folder.

bash Run_ZEPPI.sh $YOUR_PATH/Demo/bacteria_PDBdimer_demo.csv $YOUR_PATH/Scratch/bacteria_PDBdimer_demo_ZEPPI_m.csv -m

Step 3. Understand ZEPPI results

In the above example, the final output bacteria_PDBdimer_demo_ZEPPI.csv contains the ZEPPI score (last column). The detailed results are described in the below:

Column	Meaning
PPI	the name of PPI
N_MSA	the MSA depth
N_IFR	the number of interface residues
Zmean_MI	Z-score using the mean Mutual Information of all the interface contacts
Zmean_Con	Z-score using the mean Conservation score of all the interface contacts
Zmean_DCA	Z-score using the mean DCA score of all the interface contacts
Ztop_MI	Z-score using the top Mutual Information of all the interface contacts
Ztop_Con	Z-score using the top Conservation score of all the interface contacts
Ztop_DCA	Z-score using the mean DCA score of all the interface contacts
ZEPPI	the largest Z-score among all the above metrics

For more details, please read the ZEPPI paper.

Citation

Zhao, Haiqing, et al. "ZEPPI: Proteome-scale sequence-based evaluation of protein–protein interaction models." Proceedings of the National Academy of Sciences 121.21 (2024): e2400260121.

@article{zhao2024zeppi,
  title={ZEPPI: Proteome-scale sequence-based evaluation of protein--protein interaction models},
  author={Zhao, Haiqing and Petrey, Donald and Murray, Diana and Honig, Barry},
  journal={Proceedings of the National Academy of Sciences},
  volume={121},
  number={21},
  pages={e2400260121},
  year={2024},
  publisher={National Acad Sciences}
}

Contact

Haiqing Zhao (hz2592@cumc.columbia.edu) or Barry Honig (bh6@cumc.columbia.edu)