ABOUT EpiSKope
EpiSKope is a next-generation AI-powered viral immunogenic peptide evaluation resource for comprehensive prediction, evaluation and prioritization of B-cell and T-cell epitopes across human-infecting viruses. EpiSKope enables systematic identification and ranking of candidate epitopes from both curated viral proteins and user-submitted protein sequences, eliminating the need for multiple independent prediction tools. The platform integrates transformer-based protein language models, sequence-derived physicochemical analysis, hierarchical ensemble machine learning, biological safety assessment and interactive structural visualization into a single automated workflow.
✦ Key Screening Features ✦
DOCUMENTATION
Official Technical Specification Manual for the EpiSKope Universal AI-Based Viral Epitope Prediction Server.
1. Data Architecture & ETL Pipeline
To prevent machine learning models from overfitting to redundant or internal viral fragments, EpiSKope utilizes an automated Extraction, Transformation, and Loading (ETL) pipeline governed by strict mathematical logic.
Biological Exclusion Sieve & Deduplication
Genomic data queried from UniProt and NCBI contains massive redundancy. The pipeline utilizes a linguistic sieve to isolate surface-exposed structural antigens while discarding unviable targets (polymerases, helicases, internal replication factors). Additionally, the system employs an O(1) Memory Hashing Deduplication Shield to purge identical sub-strains before database commitment.
Redundant sequences purged
Curated surface antigens
3D PDB coordinates indexed
Human-infecting species
2. Mathematical Pipeline & Feature Engineering
The core prediction engine abandons generic motif-matching in favor of mathematically optimized physicochemical profiling combined with deep Protein Language Models (PLMs).
Biological Safety Gating
Before a peptide is passed to the final ranking matrix, it is subjected to strict thresholding parameters to simulate clinical viability:
- Antigenicity Gate: Evaluated via alignment-free Autocross-Covariance (ACC). Peptides <0.4 are rejected.
- Toxicity Gate: Evaluated via SVM analyzing dipeptide composition matrices. Probabilities >0.5 are flagged as toxic.
- Allergenicity Gate: Cross-referenced against FAO/WHO criteria (>35% identity over 80 aa).
The Unified Scoring Formulation
To provide a singular, actionable metric, EpiSKope calculates a Unified Score (\(U_{score}\)) harmonizing ML probability with biophysical propensities, penalized by clinical risks.
Deep Transformer Hidden-State Extraction
To capture global structural context, sequences are tokenized into high-dimensional tensors via two pre-trained PLMs. The multi-source feature space (\(X_i\)) is concatenated for the ML classifiers.
3. Machine Learning Validation & Stacking Ensemble
The 2304-dimensional feature space is processed by a highly tuned Stacking Ensemble comprising 8 algorithms (XGBoost, LightGBM, CatBoost, Random Forest, Extra Trees, SVM-Linear, Logistic Regression, Deep MLP). To mathematically prove generalized predictive power, the architecture was evaluated using a strict 80/20 Stratified Holdout methodology on unseen validation sets.
| Pipeline Target | F1-Score | Matthews Correlation (MCC) | AUC-ROC |
|---|---|---|---|
| T-Cell Pipeline (Base Models) | 0.4477 - 0.6304 | 0.1925 - 0.2956 | — |
| 🏆 T-Cell Stacking Ensemble | 0.5334 | 0.3122 | 0.7154 |
| B-Cell Pipeline (Base Models) | 0.4465 - 0.6266 | 0.2687 - 0.3877 | — |
| 🏆 B-Cell Stacking Ensemble | 0.5303 | 0.3644 | 0.7765 |
4. Intellectual Property
- Claim 1: Unified Dual-Route Architecture enabling indexed relational tracking and zero-shot ML sequence analysis with dynamic NCBI fallback.
- Claim 2: Multi-Source Transformer Pipeline fusing ESM-2 and ProtBERT dimensions into an 8-algorithm stacking ensemble.
- Claim 3: Integrated Safety Sieve executing real-time gating for conservancy, toxicity, allergenicity, and ADE mitigation.
Meet the Team
Amity University Uttar Pradesh, Noida
Amity University Uttar Pradesh, Noida
Amity University Uttar Pradesh, Noida
Amity University Uttar Pradesh, Noida