Designing novel peptides with amyloid-β binding and clearance potential using BiLSTM and molecular dynamics

Area of Science:

• Biochemistry
• Computational Biology
• Artificial Intelligence

Background:

• Generative AI is advancing biomolecular design, but creating functional, target-specific peptides is challenging.
• Amyloid-beta (Aβ) peptides are central to Alzheimer's disease pathology.
• Existing methods struggle to generate peptides with reliable binding affinity and stability.

Purpose of the Study:

• To introduce and validate a novel two-stage Bidirectional Long Short-Term Memory (BiLSTM) framework for de novo peptide design.
• To generate functional peptides targeting Aβ42, a key player in Alzheimer's disease.
• To establish a generalizable AI-driven pipeline for therapeutic peptide discovery.

Main Methods:

• Trained a BiLSTM AI pipeline on proteins annotated with Gene Ontology terms for Aβ interaction.
• Fine-tuned the model on validated peptide fragments for motif capture.
• Applied biophysical filters, sequence similarity analysis, and structural modeling (AlphaFold2) for candidate selection.
• Validated binding affinity and stability using molecular docking (pyDockWEB) and molecular dynamics simulations.

Main Results:

• Generated 1,000 peptide sequences, shortlisted 11 AI-Designed Novel Peptides (ADNP1-ADNP11).
• Identified ADNP7 as the top candidate with favorable docking score (-63.33 kcal/mol) and predicted binding to Aβ aggregation regions.
• Molecular dynamics simulations confirmed complex stability with strong binding free energy (-50.6 kcal/mol) via MM/PBSA analysis.

Conclusions:

• The fine-tuned BiLSTM architecture successfully generates novel, stable peptides with high predicted binding affinity for therapeutic targets.
• The AI-driven pipeline demonstrates generalizability for functional peptide design in drug discovery and synthetic biology.
• Computational validation focused on binding; experimental testing is needed to confirm Aβ clearance potential.