PCBEP: A Point Cloud-based Deep Learning Framework for B-cell Epitope Prediction

The ability to identify B-cell epitopes is specifically important in the field of vaccine design, which is particularly relevant for rapid pandemic response. In the past, a range of computational tools have been developed for predicting B-cell pitopes based on the 3D structure of antigen. However, they have presented limited performance and generalization. Here, we proposed PCBEP, a novel B-cell epitope prediction framework based on the three-dimensional point clouds. It employs multiple Point Transfomer layers to learn the embedding describing the atomic properties, geometric features and evolutionary profile. The built-in self-attention mechanism is able to advance the informative components and message passing operator can model the interactions among neighboring atoms at different scales for stimulating the micro-environmental information for predicting epitopes.

Installation Guide

Install PCBEP from GitHub

git clone  https://github.com/yuyang-0825/PCBEP && cd PCBEP

Install dependency packages

1. Install PyTorch following the official guide.
2. Install torch-geometric following the official guide.
3. Install other dependencies:

pip install -r requirements.txt

The above steps take 20-25 mins to install all dependencies.

Evaluate our model

First unzip the file:'Data/data_feature_surface.7z'

If you want to evaluate the results of our data.

python pcbep_test.py

If you want to train our 5-fold cross-validation model.

python pcbep_5fold_train.py

Test your own data

If you want to use our model to test your own data. Please refer to the following steps:

1. Get the epitope label of the data.

Use 0 1 to encode epitope labels and organize them into fasta format, for example:

>6WIR-C
00000000000000000000000000000000010000000000111100000000000000000000000000000000000111111
>5W2B-A
0000001001100111111000100100000001011111000000000000000000000000000000000000000000000100000011001
>5TFW-O
0000000000000000000000000000000000000000000000000000010100000010000010000000000000000000000000000000000000011110011011001000000000000000000000000000000000

2. Get the surface label of the data.

Use 0 1 to encode surface labels and organize them into fasta format, for example:

>6WIR-C
011111111111111111111111111111111110011111111111111111111111111111111111111111111111111111
>5W2B-A
01111111111111111111110111111111111111111111111111111111111111111111111111111111011111111111011111
>5TFW-O
01111111111111111111101110011011110011101110111111111011111110010111111111100000101111101111111111101000010111111111011111111111110110111111101110111111111

3. Obtain the PSSM matrices of the antigens to be tested.

Last position-specific scoring matrix computed, weighted observed percentages rounded down, information per position, and relative weight of gapless real matches to pseudocounts
            A   R   N   D   C   Q   E   G   H   I   L   K   M   F   P   S   T   W   Y   V   A   R   N   D   C   Q   E   G   H   I   L   K   M   F   P   S   T   W   Y   V
    1 S     2  -1   0  -1  -1   0  -1   0  -1  -2  -2  -1  -1  -3  -1   4   2  -3  -2  -1   26   0   0   0   0   0   0   0   0   0   0   0   0   0   0  65   9   0   0   0  0.40 0.03
    2 D    -1  -2   1   5  -3   0   2  -1  -1  -3  -4  -1  -3  -4  -2   2   0  -4  -3  -3    0   0   0  68   0   0   8   0   0   0   0   0   0   0   0  23   0   0   0   0  0.71 0.04
    3 Y    -1  -2  -2  -3  -2  -2  -2  -3   0   0   0  -2   0   3  -3  -1   2   0   5   1    0   0   0   0   0   0   0   0   0   0   0   0   0  16   0   0  22   0  39  23  0.36 0.03  

4. Get the feature file for your own data

python Pretreatment/generate.py -l ../Data/label.txt -s ../Data/surface.txt -p ../Data/data -m ../Data/PSSM -o ../Data/data_feature_surface.txt

• -lor--label file path for epitope label. [default:'../Data/label.txt']
• -sor--surface file path for surface label. [default:'../Data/surface.txt']
• -por--pdb fold path for pdb files to be tested. [default:'../Data/data']
• -mor--pssm fold path for PSSM files of antigens. [default:'../Data/PSSM']
• -oor--output output file path. [default:'../Data/data_feature_surface.txt']

5. Get test result

python pcbep_test.py -i Data/data_feature_surface.txt -c checkpoint.pt -o result/result_pcbep.txt

• -ior--input file path for fature path. [default:'Data/data_feature_surface.txt']
• -cor--checkpoint file path for pre-trained model.. [default:'checkpoint.pt']
• -oor--ouptup output file path. [default:'result/result_pcbep.txt']