DeepVF

DeepVF predicts virulence factors in bacterial genomes using a hybrid framework that combines classical machine learning and deep learning to improve genome-wide VF identification from protein-coding gene sequences.

Key Features:

• Hybrid Framework: Integrates classical machine learning algorithms and deep learning models to leverage diverse computational techniques for VF prediction.
• Stacking Strategy: Combines multiple baseline models into a robust meta model via a stacking approach.
• Comprehensive Feature Exploration: Systematically examines heterogeneous features extracted from bacterial genome sequences.
• Diverse Algorithmic Approach: Employs random forest, support vector machines, extreme gradient boosting, multilayer perceptron, convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and deep neural networks (DNNs) to train 62 baseline models.
• Up-to-Date Dataset: Trained on an enlarged, contemporary dataset to reflect evolving characteristics of virulence factors.
• Benchmarking Performance: Demonstrates more accurate and stable performance than individual baseline models and outperforms existing VF predictors on independent test datasets.

Scientific Applications:

• Genome-wide VF prediction: Identifies potential virulence factors from protein-coding gene sequences in bacterial genomes.
• Pathogen genomic analysis: Supports investigation of pathogenic mechanisms and discovery of VFs relevant to emerging antibiotic resistance to inform therapeutic research.

Methodology:

Feature extraction from bacterial genome data; training of 62 baseline models using random forest, support vector machines, extreme gradient boosting, multilayer perceptron, CNN, LSTM, and DNN; and construction of a final meta model via a stacking strategy.