AlloFusion: Allosteric Site Prediction Based on Language Models and Multi-Feature Fusion

The AlloFusion program is a method for protein allosteric site prediction.

Requirement

• python 3.11

• numpy 1.26.4

• pandas 2.2.3

• joblib 1.4.0

• ProDy 2.4.1

• torch 2.5.0+cu118

• tensorflow 2.12.0

• transformers 4.42.4

---

To run the AlloFusion, you need to install the bioinformatics tools and download the corresponding databases.

（1）Download the prot_t5_xl_uniref50 model from the following link: https://huggingface.co/Rostlab/prot_t5_xl_uniref50/tree/main

（2）Install blast+ for extracting PSSM(position-specific scoring matrix) profiles

​ To install blast-2.15.0+ and download NR database (ftp://ftp.ncbi.nlm.nih.gov/blast/db/) for psiblast, please refer to BLAST(https://www.ncbi.nlm.nih.gov/books/NBK52640/).

---

🔍 Model Download Instructions

Due to the large size of the trained model file (exceeding GitHub's file size limit), it has been uploaded to the Releases section of this repository.

Please visit the Releases page to download the full model file:

👉 Click here to download the model from the Releases page

After downloading, place the model file in the project root directory or the designated model folder to use it or reproduce the experimental results.

---

📁 Dataset Download Instructions

Due to the large size of the dataset (exceeding GitHub's file size limit), it has also been uploaded to the Releases section of this repository.

Please visit the Releases page to download the full dataset:

👉 Click here to download the dataset from the Releases page

After downloading, extract the dataset to the project root directory or the designated data folder to proceed with training or evaluation.

---

How to run

Step 1: Extract protein sequence based on input pdbid and chain

Step 2: Residue feature extraction

Step 3: Combined residue characterization

Step 4: Loading the model for predicting AFRs

---

Example

An allosteric protein with PDB ID "4ZSI" is used as an example to show the process. This PDB file is 4ZSI.pdb. Only the protein functional chain is preserved.

python AlloFusionMain.py --PDBID [pdbid] --CHAIN [chain]

for example:

python AlloFusionMain.py --PDBID 4ZSI --CHAIN B

The parameter [pdbid] is the PDB file name of the allosteric protein.

The parameter [chain] is the functional chain of the target protein.

Then, AlloFusion program will perform the feature extraction and prediction process, which will take some time.

The final prediction result is a file containing the residue IDs of AFRs residues:

4ZSI_allosteric_residues.txt

AlloFusion Allosteric Site Forming Residues:
Residues: ( Chain B and resid 92, 95, 96, 97, 98, 99, 109, 111, 140, 142, 144, 152, 154, 156, 174, 175, 177, 193, 219, 221, 223, 232, 233, 234, 236, 238, 247)

and the script for viewing the allosteric sites composed of AFRs in PyMol: 4ZSI_allosteric_sites.pml.