Computational Drug Screening

ISEF Category: Biochemistry

Subcategory: Other  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A compound can look like a winner in docking and still fail once the model gets stricter. That gap is where your project lives. You are not just picking a molecule, you are testing whether a notebook can help students screen compounds in a repeatable way.

What Is It?

This project builds a browser-based workflow that takes a new compound and runs it through three checks. Docking asks, “How well might this molecule fit the target?” Molecular dynamics, or MD, checks whether that fit stays steady when the atoms move. ADMET looks at how a body might absorb, distribute, metabolize, excrete, and tolerate the compound.

Think of it like a sports tryout. Docking is the first drill. MD is the scrimmage. ADMET is the health and eligibility check. Your real project is the full notebook, because you are testing whether other students can run the same workflow and get the same kind of result on new compounds.

Why This Is a Good Topic

This makes a strong science fair topic because you can test a clear claim, not just build a demo. You can compare known compounds, measure how the pipeline ranks them, and see whether each step improves the result. The work connects to drug discovery, toxicology, and reproducibility, which gives it real-world weight. You can learn how computational screening works without needing a wet lab.

Research Questions

• How does changing the docking program affect the rank order of known active compounds?
• What is the effect of adding molecular dynamics snapshots before rescoring on hit ranking?
• Does an ADMET filter remove compounds that later look strong in docking and MD?
• To what extent do different protein structures change the top-ranked compounds?
• Which compound descriptors best predict whether a molecule passes your pipeline's cutoff rules?
• How does repeated running with different random seeds change the final ranking?

Basic Materials

• Laptop with a modern web browser
• Free Google account for Colab
• Stable internet connection
• Spreadsheet software such as Google Sheets or Excel Online
• PubChem compound records
• RCSB Protein Data Bank structures
• Plain-text editor for notes
• Reference set of known active and inactive compounds
• Headphones or a quiet workspace for long runs.

Advanced Materials

• Linux workstation with a CUDA-capable GPU
• Access to a campus or cloud compute queue
• AutoDock Vina or a similar docking package
• OpenMM or GROMACS for molecular dynamics
• ChimeraX or PyMOL for structure prep and inspection
• RDKit for compound handling and descriptor checks
• Curated bioactivity set from ChEMBL or literature
• Assay data for outside validation of the notebook.

Software & Tools

• Google Colab: Runs the notebook in a browser without local installs.
• RDKit: Prepares molecules, filters compounds, and computes basic descriptors.
• AutoDock Vina: Scores how well a compound fits the target pocket.
• OpenMM: Runs molecular dynamics on prepared systems.
• PubChem: Supplies compound records and structure data for your inputs.

Experiment Steps

1. Define the exact claim your notebook will test, such as rank order, filter agreement, or repeatability.
2. Choose one protein target and a reference set with known active and inactive compounds.
3. Decide how you will prepare ligands, protein structures, and repeat runs so the workflow stays reproducible.
4. Build a scoring ladder that combines docking, MD-based stability checks, and ADMET filters.
5. Plan a validation test that compares your notebook against the reference set and records where it fails.
6. Decide how you will report uncertainty, runtime, and edge cases so another student can repeat the workflow.

Common Pitfalls

• Mixing protein structures from different sources without aligning residue numbering, which breaks comparisons.
• Comparing docking scores across runs without fixing the seed or repeating the setup, which makes ranking noisy.
• Treating one best score as truth, which hides compounds that only look good before MD or ADMET filtering.
• Feeding unprepared ligands or uncertain protonation states into the notebook, which changes both docking and property estimates.
• Claiming biological success from in-silico scores alone, which overstates what the pipeline can prove.

What Makes This Competitive

A stronger project compares the whole pipeline against a reference set with known outcomes, not just one docking score. The best version checks whether MD rescoring or ADMET screens improve the ranking, then reports where they do not. If you add repeat runs, error ranges, and a clear audit trail, your work starts to look like a reproducibility study, not a class demo. That kind of analysis stands out because it asks how well the method works, not only whether the notebook runs.

Project Variations

• Use a different target protein from a disease area and compare how the pipeline changes.
• Swap in natural products versus drug-like synthetic compounds and see which passes the filters.
• Compare docking-only ranking with docking-plus-MD ranking on a small validated compound set.

Learn More

• PubChem: Search compound records, structures, and bioassay summaries on the NIH NCBI site.
• RCSB Protein Data Bank: Find protein structures and annotation on the RCSB site.
• NCBI Bookshelf: Read free chapters on pharmacology, binding, and ADMET basics.
• ChEMBL: Explore bioactivity records and reference compounds through the EMBL-EBI database.
• AutoDock Vina documentation: Review docking setup and scoring details in the official project docs.
• OpenMM documentation: Read the free guides for molecular dynamics setup and analysis.