MM-PBSA Screening of Natural Product Antivirals

ISEF Category: Chemistry

Subcategory: Computational Chemistry  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A weak binder can still look strong on a docking score. That is why antiviral screens often need a second pass. You can test which plant compounds stay attached more convincingly when the protein and ligand are treated as a moving system, not a frozen puzzle piece. That gives you a much better estimate of real binding strength.

What Is It?

This project asks a simple question with a complicated answer, which plant-derived molecules seem most likely to bind SARS-CoV-2 nsp13 helicase. Docking gives you a quick first guess. It places each molecule into the protein pocket and scores the fit, like trying keys in a lock. MM-PBSA then goes a step farther by estimating how much energy favors the bound state after accounting for the surrounding water and the protein's movement.

Think of docking as a snapshot and MM-PBSA as a short video clip. The snapshot can miss flexibility, and some molecules only look good because they fit one pose well. MM-PBSA, which stands for molecular mechanics, Poisson-Boltzmann surface area, tries to estimate the balance of forces more realistically. OpenMM is the engine that can run parts of this workflow, while decomposition lets you ask which parts of the interaction matter most, such as hydrophobic contact, electrostatics, or specific residues in the binding site.

Why This Is a Good Topic

This is a strong science fair topic because you can compare two prediction methods on the same set of compounds and ask which one ranks candidates more consistently. The work connects to antiviral discovery, especially the search for lower-cost leads from natural products. You can collect public structures and compound data, build a clear scoring pipeline, and generate results that are measurable, comparable, and easy to graph.

Research Questions

• How does MM-PBSA ranking compare with docking score ranking for plant-derived natural products against SARS-CoV-2 nsp13 helicase?
• What is the effect of using different natural product classes on predicted binding free energy for nsp13?
• Does residue-level energy decomposition identify the same key binding residues across multiple top-ranked ligands?
• To what extent do docking poses predict MM-PBSA stability after short molecular dynamics refinement?
• Which compound properties, such as logP, hydrogen bond count, or aromatic ring count, best correlate with predicted binding strength?
• How does the choice of solvation model change the final rank order of candidate ligands?

Basic Materials

• Computer with a modern processor and at least 16 GB RAM.
• Internet access for downloading protein structures and compound data.
• OpenMM installed on a local machine or university computer.
• Python with NumPy, pandas, matplotlib, and RDKit.
• Molecular visualization software such as UCSF ChimeraX or PyMOL.
• Public SARS-CoV-2 nsp13 structure from the Protein Data Bank.
• Public natural product compound list from PubChem or a similar database.
• Spreadsheet software for tracking compounds, scores, and rankings.

Advanced Materials

• Access to a Linux workstation or server for longer simulations.
• GPU-enabled computer for faster molecular dynamics runs.
• OpenMM with MM-PBSA analysis scripts.
• AmberTools or compatible force-field preparation tools.
• Protein structure preparation tools for protonation and cleanup.
• High-quality compound library of plant natural products.
• Statistical analysis software for rank correlation, clustering, and regression.
• Optional access to molecular docking software for baseline comparison.

Software & Tools

• OpenMM: Runs molecular dynamics and supports the simulation workflow for binding analysis.
• Python: Organizes the pipeline, scores compounds, and makes plots.
• RDKit: Filters natural products and calculates basic molecular descriptors.
• UCSF ChimeraX: Lets you inspect protein-ligand poses and prepare structures.
• ImageJ: Not needed here, so skip it unless you also analyze figure panels or screenshots.

Experiment Steps

1. Define a small, chemically diverse set of plant-derived compounds and decide how you will justify that set.
2. Prepare one consistent protein structure and one clean ligand-processing workflow so every compound starts from the same rules.
3. Run a docking baseline first, then decide how many top poses deserve molecular dynamics refinement.
4. Build a comparison plan for MM-PBSA results, including how you will rank compounds and handle tied scores.
5. Choose which energy terms or residue contributions you will track so you can explain why one ligand beats another.
6. Design a validation check against compound properties, literature hints, or repeated runs so your rankings are not just a one-off result.

Common Pitfalls

• Mixing docking poses and MM-PBSA poses from different preparation rules, which makes the ranking unfair.
• Comparing ligands with different protonation states, which can flip the apparent binding trend.
• Trusting a single simulation run, which can make noisy energy values look meaningful.
• Ignoring protein flexibility around the pocket, which can hide why a ligand appears to score well.
• Treating a raw MM-PBSA number as a final truth, which can lead you to miss uncertainty and overlap between candidates.

What Makes This Competitive

A competitive version of this project goes beyond a simple top-10 list. You would compare methods, test repeatability, and show why a ranking changes when the model gets more realistic. Strong projects also explain residue-level interactions and connect the computation to a clear biological question. If you add uncertainty estimates, rank correlation, and a careful discussion of model limits, your work starts to look like real screening research.

Project Variations

• Compare plant alkaloids, flavonoids, and terpenoids as separate natural product classes against nsp13.
• Test whether a second target protein from SARS-CoV-2 changes the docking versus MM-PBSA rank order.
• Add a property-based analysis that asks whether molecular weight, polarity, or ring count predicts better binding scores.

Learn More

• PubChem: Use the compound database to find structures, properties, and identifiers for natural products.
• RCSB Protein Data Bank: Download SARS-CoV-2 protein structures and read the associated structure notes.
• OpenMM Documentation: Learn how to set up molecular dynamics and analysis workflows.
• NCBI PubMed: Search for review articles on SARS-CoV-2 helicase, MM-PBSA, and natural product screening.
• MIT OpenCourseWare: Look for molecular modeling and computational chemistry course materials that explain force fields and free energy methods.