🧬 biology

Integrated Machine Learning, Molecular Docking, and Molecular Dynamics Simulations Design Novel Peptide Inhibitors of Rabies Virus Glycoprotein

This study employs an integrated computational approach combining machine learning, molecular docking, and molecular dynamics simulations to design and validate novel non-toxic peptide inhibitors derived from α-bungarotoxin, identifying the P40 candidate as a promising therapeutic agent targeting the Rabies virus glycoprotein.

Original authors: Harshada Sawant, Ashish Asrondkar, Vrushali Patil

Published 2026-08-03

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Harshada Sawant, Ashish Asrondkar, Vrushali Patil

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine the human body as a bustling city, and viruses as sneaky burglars trying to break in. To get inside, these burglars need a master key to unlock the front door. In the case of the rabies virus, that "front door" is a specific receptor on our nerve cells, and the "master key" is a protein on the virus's surface called the Glycoprotein (or RABV-G). This key is so effective that it mimics the shape of a snake venom toxin, tricking the body's defenses and allowing the virus to slip right into the nervous system. Once inside, the virus causes a disease that is almost always fatal, making the race to find a way to jam that lock a matter of life and death. Scientists have long known that if they could design a "fake key" or a "plug" that fits perfectly into the virus's lock but doesn't open the door, they could stop the infection before it starts. This is where the world of computer science meets biology: using powerful algorithms to design tiny molecular pieces, called peptides, that act as these perfect plugs.

In this study, a team of researchers from the Haffkine Institute in Mumbai decided to play the ultimate game of molecular Tetris. They started with a blueprint from a very different source: a toxin found in the venom of the krait snake, known as α-bungarotoxin. This snake toxin is famous for its ability to lock onto the same nerve receptors that the rabies virus uses. The researchers realized that the virus and the snake toxin are essentially competing for the same spot. So, they asked a clever question: "What if we could take the part of the snake toxin that grabs the lock, but strip away the deadly poison, and turn it into a harmless shield?"

Using a mix of advanced computer tools, the team designed a library of these "harmless shields." First, they used a hybrid Machine Learning and Deep Learning framework—think of it as a super-smart digital filter—to screen thousands of potential designs. This filter was trained to spot and discard any design that might still be toxic or harmful to humans, ensuring only the safe candidates moved forward. They then used molecular docking, which is like a high-speed virtual simulation where they tried to fit each peptide into the virus's lock to see which ones clicked together the tightest.

The results of this digital screening were exciting. The researchers identified a standout candidate, a peptide they named P40 (with the sequence MKMKCHTTATMKIV). In their computer simulations, P40 fit into the rabies virus's Glycoprotein better than any other design they tested. It formed a strong, stable bond, effectively "jamming" the lock. To be absolutely sure this wasn't just a fluke of the simulation, they ran a 100-nanosecond molecular dynamics simulation. Imagine this as a high-speed movie showing the peptide and the virus interacting over time; the movie showed that P40 held its ground, staying firmly attached to the virus without wobbling or falling off, even as the molecules jiggled and moved.

The study suggests that this engineered peptide, derived from a snake toxin but stripped of its danger, could be a powerful new weapon against rabies. While the paper emphasizes that these findings are currently based on computer models and simulations, the results point to P40 as a highly promising lead. It's a hopeful step toward a future where we might have a peptide-based drug that can stop the rabies virus in its tracks, turning a deadly lock into a harmless puzzle that the virus can no longer solve.

Technical Summary: Integrated Machine Learning, Molecular Docking, and Molecular Dynamics Simulations Design Novel Peptide Inhibitors of Rabies Virus Glycoprotein

Problem Statement
Rabies remains one of the most lethal viral zoonotic diseases globally, causing approximately 55,000 deaths annually, with the majority occurring in Asia and Africa. The disease is caused by the Rabies Virus (RABV), a neurotropic RNA virus where the Glycoprotein (RABV-G) serves as the primary determinant for host cell recognition and viral entry. RABV-G interacts with host receptors, including the nicotinic acetylcholine receptor (nAChR), a mechanism shared with the neurotoxin $\alpha$ -bungarotoxin. While RABV-G is a validated target for antiviral development and vaccine design, there is a critical need for novel therapeutic agents that can specifically disrupt the virus-receptor interaction without the neurotoxicity associated with natural toxins. The study addresses the challenge of designing safe, non-toxic peptide inhibitors that mimic the receptor-binding motifs of $\alpha$ -bungarotoxin but lack its toxic properties.

Methodology
The study employed a multi-stage in silico pipeline integrating peptide engineering, machine learning (ML), deep learning (DL), and structural biology simulations:

Target Modeling: A 3D structural model of the RABV-G protein was generated using SWISS-MODEL (template 6lgx.1.A) and validated via Ramachandran plot analysis, confirming 92.5% of residues in favored regions.
Peptide Library Design: Starting from the amino acid sequence of $\alpha$ -bungarotoxin, a library of peptide analogues was created. The design strategy involved fragmenting the toxin sequence and applying systematic point mutagenesis within specific physicochemical groups (acidic, basic, hydrophobic, etc.) to retain binding motifs while eliminating neurotoxicity.
Toxicity Screening (ML-DL Framework): A hybrid framework combining supervised learning (Random Forest, SVM, KNN) and deep learning (Deep Neural Networks with autoencoders) was utilized. Peptides were screened based on physicochemical descriptors (hydrophobicity, charge, Boman index). This dual-model approach filtered out 163 potentially toxic candidates from an initial pool, retaining 40 non-toxic peptides.
Physicochemical and Antiviral Profiling: The 40 candidates were analyzed for stability, solubility, and predicted antiviral activity (IC50) using tools like ProtParam and AVP-IC50Pred.
Molecular Docking: High-ambiguity driven protein-peptide docking was performed using HADDOCK and HDOCK servers to identify binding modes and affinities with the RABV-G binding pocket. Key residues in the binding pocket (e.g., Glu272, Glu279, Asp283) were identified via DoGSiteScorer.
Molecular Dynamics Simulations (MDS): The top candidates were subjected to 100 ns MDS using the Desmond module. Stability was assessed through Root Mean Square Deviation (RMSD), Root Mean Square Fluctuation (RMSF), Hydrogen Bond analysis, Principal Component Analysis (PCA), Free Energy Landscape (FEL), and Dynamic Cross-Correlation Matrix (DCCM).

Key Results

Lead Candidate Identification: Among the evaluated candidates, P40 (MKMKCHTTATMKIV) emerged as the most promising inhibitor. It demonstrated the most favorable HADDOCK score of -90.0 ± 3.8 and a binding free energy ( $\Delta G$ ) of -10.3 kcal/mol, with a predicted dissociation constant ( $K_d$ ) of $2.7 \times 10^{-8}$ M. The abstract explicitly identifies P40 as the top candidate, followed by P3 (CLDLGYTIVCHTTATSPI) with a score of -86.3, P25 (MKIVCHTTMKSPMK) with a score of -86.0, and P21 (MKIVCHTTATSPMKV) with a score of -75.0.
Secondary Candidates: Peptides P3, P25, and P21 also showed competitive docking scores and binding affinities. Notably, P25 exhibited a superior HADDOCK score (-86.0) compared to P21 (-75.0), placing it as the second or third strongest binder depending on the specific metric (HADDOCK vs. binding free energy) considered, but consistently ranking higher than P21 in docking affinity.
Interaction Mechanism: Docking and interaction analysis revealed that the lead peptides bind to a specific pocket on RABV-G involving key residues: Glu272, Glu279, Asp283, Lys276, and Arg297. The binding is stabilized by a combination of electrostatic interactions (between basic peptide residues like Lys/His and acidic protein residues) and hydrophobic contacts.
Dynamic Stability: MDS results indicated that the P40-RABV-G complex achieved structural stability rapidly, maintaining low RMSD (2–4 Å) and RMSF values throughout the 100 ns simulation. The Free Energy Landscape analysis showed P40 formed a deep, well-defined energy basin, indicating a highly stable conformational state compared to the more flexible unbound protein or complexes with other peptides.
Structure-Activity Relationship (SAR): The study identified that peptides with intermediate to high isoelectric points, specific CHTT-containing motifs, and a length of 14–16 amino acids exhibited the strongest predicted antiviral potential.

Significance and Claims
The paper claims to successfully demonstrate an integrated computational strategy for the de novo design of peptide inhibitors against the Rabies Virus Glycoprotein. The significance of the work lies in:

Safety by Design: The successful elimination of neurotoxicity from $\alpha$ -bungarotoxin-derived sequences while retaining the ability to target the viral receptor.
Methodological Integration: The effective use of a hybrid ML-DL framework for toxicity screening, coupled with rigorous structural validation and dynamics simulations, provides a robust pipeline for antiviral peptide discovery.
Therapeutic Potential: The identification of P40 (MKMKCHTTATMKIV) as a stable, high-affinity binder suggests it is a viable lead candidate for the development of peptide-based anti-rabies therapeutics. The abstract notes that $\alpha$ -bungarotoxin derived peptides can target RABV-G and that P40 (corrected from the abstract's typo "P440") can be a promising lead candidate.
Future Direction: The authors state that these findings provide a "proper foundation" for future experimental validation, specifically citing the need for in vitro binding assays, antiviral activity studies, and in vivo investigations to confirm the computational predictions.

The study concludes that engineered $\alpha$ -bungarotoxin-derived peptides, specifically P40, possess significant potential to interfere with RABV-G-mediated host cell entry, offering a new avenue for rabies treatment.

More like this