Multi-Transformer Feature Fusion With MHSA-GRU for Toxicity Prediction

An advanced transformer-based system for toxicity prediction using feature extraction from ProtBERT, ESM2, ProtT5, and BioT5, followed by Differential Evolution fusion, MHSA-GRU classification, and SHAP-based feature selection.

Analyze Sequence

Enter sequences for AI-powered toxicity detection using advanced deep learning algorithms.

Number of Sequences

Default: 15 sequences (Min: 10)

Peptide Sequences

Enter one sequence per line or load sample sequences

Model Performance Metrics

96.42%

Accuracy

Overall Correctness

96.90%

Sensitivity

True Positive Rate

98.65%

AUC Score

Area Under Curve

95.88%

Precision

Positive Predictive Value

90.35%

Specificity

True Negative Rate

95.64%

F1-Score

Harmonic Mean

How It Works

Our multi-stage deep learning pipeline ensures accurate and reliable toxicity prediction

1. Input Sequence

Provide your amino acid sequence securely to the system

2. Feature Extraction

Multiple transformer models extract deep biological representations

3. AI Analysis

Fused features pass through MHSA-GRU for advanced toxicity prediction

4. Results

Receive a detailed prediction report along with confidence scores