EnDL-HemoLyt

EnDL-HemoLyt predicts low-hemolytic therapeutic peptides using an ensemble deep-learning framework that combines handcrafted biochemical features and deep-learning-derived features to improve hemolytic activity prediction.

Key Features:

• Ensemble Learning Approach: Integrates decisions from multiple deep learning algorithms, including bidirectional long short-term memory (BiLSTM), bidirectional temporal convolutional network (BiaTConvNet), and 1-dimensional convolutional neural network (1D-CNN).
• Comprehensive Feature Utilization: Combines handcrafted features (HCF) with deep learning-based features (DLF) to form a comprehensive feature vector capturing biochemical properties and data-driven patterns.
• Ablation Studies: Performs ablation studies demonstrating the contribution of the ensemble algorithm, HCF, and DLF to overall performance.
• Performance Metrics: Reports mean test values of approximately 87 for A_cc, 85 for S_n, 86 for P_r, 86 for F_s, 88 for S_p, 87 for B_a, and 73 for Mcc.
• N/C Terminal Modifications: Provides predictions that include peptides with N/C terminal modifications.
• Recent Dataset: Uses a peptide dataset assembled from data generated over the past eight years.

Scientific Applications:

• Preliminary peptide selection: Enables in-silico identification of candidate therapeutic peptides with low hemolytic activity prior to experimental testing.
• Reduction of RBC assays: Reduces reliance on mammalian red blood cell hemolysis assays by prioritizing low-hemolytic candidates.
• Modified-peptide evaluation: Assesses peptides with N/C terminal modifications to inform downstream in-vitro validation.

Methodology:

Ensemble deep learning integrating BiLSTM, BiaTConvNet, and 1D-CNN models; integration of handcrafted features (HCF) and deep-learning features (DLF); trained and evaluated on a recent peptide dataset spanning the past eight years; ablation studies used to evaluate component importance.