GAN-ML

Generative based data augmentation for ACPs

#GAN-ML: Generative based data augmentation for anticancer peptide prediction using inter- pretable machine learning

Overview

This project aims to [develop a generative based data augmentation method as a nascent approach to Sampling based and feature based augmentation].

Dataset

• Description: [Main Dataset] Dataset The dataset used in this project is sourced from [AntiCP2.0] and consists of [861] POSITVE samples and [861] NEGATIVE samples in fasta format. Additionally, this project consist of alterante dataset sourced from AntiCP 2.0 which consists of 970 positive sample and 970 negative samples in fasta format.
• Features: The dataset includes the following features:
  • Main Dataset: Positive samples : ACP (Anticancer peptides without antimicrobial activity) and Negative Samples : AMP(Antimicrobial peptides without Anticancer Activity (ACPs).
  • Alternate Dataset: Positive samples : ACP (Anticancer peptides (with or without antimicrobial activity) and Negative Samples :Non-Secretory Protein samples from UniProt.
  • ...
• Data Preprocessing: Prior to modeling, the dataset underwent preprocessing steps, including cleaning, normalization/scaling, and feature engineering specially MRMD based method, SFS Method, Boruta Method, and Shapely Method for Feature Selection .

Feature Encoding

• Feature Encoding Techniques: Peptide were encoded using AAC, AAE, DPC, AAI, GTPC. CKSAAGP, PAAC, CTD, PSAAC, ASDC, QSO, Comprehensive feature .
• Feature Scaling: Numerical features were scaled using Min-Max Scaling and Normalization method to ensure uniformity across features.

GAN Augmentation

• GAN Architecture: The Generative Adversarial Network (GAN) architecture used for data augmentation consisted of DC-GAN architecture where peptides were encoded using a PC6 (Physiochemcial Encoding method).
• Data Augmentation Process: The GAN was trained on the dataset to generate synthetic data samples specially Anticancer peptides, which were then combined with the original dataset for training.
• Benefits of GAN Augmentation: GAN augmentation helped in increasing the diversity of the dataset and improving model generalization.

Comparison

• Evaluation Metrics: Model performance was evaluated using balanced and imbalanced case metrics, including accuracy, precision, sensitivity, specificity, MCC, G-mean, AUC-ROC and F1-score.
• Experimental Setup: Models were trained using DC-GAN Architecture. The dataset was split into training/validation/test sets in the ratio 80:20.
• Discussion: Analysis of results revealed Generative based data augmentation surpasses Oversampling based approach and information based data augmentation, highlighting the strengths GAN-ML over other state of art method in ACPs prediction.

Shapely Analysis Result

• **Contributing Feature Analysis : Summary Plot for the General Prediction - Model Intrepretability

• Further Experimental result will be made available upon request by researcher working in this field. Please contact me at sadik90@jbnu.ac.kr for inquiries.

Usage

• Installation: To install dependencies, run pip install -r requirements.txt.
• Usage Instructions: Encode peptides, run fea_extract.py. generates a peptide encoding in Data Frame format.
• Example Code: Below is an example code snippet for generatingh encoding:

# from fea_extract import read_fasta,insert_AAC,insert_DPC,insert_CKSAAGP,insert_CTD,insert_PAAC,insert_AAI,insert_GTPC,insert_QSO,insert_AAE,insert_PSAAC,insert_word2int,insert_ASDC
#This script is used to extract the feature for the peptide sequences