xTB Screening of Psychedelic Analog Selectivity

ISEF Category: Translational Medical Science

Subcategory: Drug Identification and Testing  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A tiny change in a molecule can flip its effect from useful to unsafe. That is why drug teams obsess over selectivity, the idea that a compound should hit one receptor, not a different one that causes side effects. You can model that pattern with quantum-chemistry tools, even before a lab makes the molecules. This project teaches you how chemists turn structure into predictions.

What Is It?

This project uses low-cost computational chemistry to compare closely related molecules and estimate which ones may fit one receptor better than another. Think of each receptor like a lock, and each molecule like a key with small notches and bends. You are not measuring real binding in a patient or a wet lab. You are building a proxy score from molecular shape, charge, and energy features that can help rank candidates for follow-up work.

xTB and GFN2 are semiempirical quantum methods. That means they are faster than full quantum calculations, but still detailed enough to estimate geometry, energy, and electronic properties. In a project like this, you can compare analogs of psilocin or DMT and ask whether certain structural changes make a molecule look more like a 5-HT2A-friendly candidate and less like a 5-HT2B risk signal. The goal is a structure-affinity ladder, a ranked map that links small chemical changes to predicted changes in behavior.

Why This Is a Good Topic

This is a strong science fair topic because you can test many related molecules with the same method and compare clear numeric outputs. The project connects to drug safety, antidepressant discovery, and why receptor selectivity matters in real medicine. You can learn molecular modeling, dataset cleaning, feature selection, and basic statistics without needing a wet lab. A careful student can build a serious computational study from public structures and free cloud tools.

Research Questions

• How does adding or removing a hydroxyl group change the predicted 5-HT2A versus 5-HT2B selectivity proxy in psilocin-like analogs?
• What is the effect of ring substitution pattern on the structure-affinity ladder for DMT-derived compounds?
• Does side-chain length change the predicted geometry score more than aromatic substitution does?
• To what extent do xTB-derived descriptors separate candidate molecules into higher and lower selectivity groups?
• Which molecular features best predict the proxy ratio between 5-HT2A and 5-HT2B for a small analog library?
• How does protonation state change the ranking of psychedelic-derived analogs in a Colab workflow?

Basic Materials

• Laptop or desktop computer with reliable internet access.
• Google account for Colab.
• PubChem for downloading 2D structures.
• NIH PubChem PUG-REST pages or structure files.
• Spreadsheet software such as Google Sheets or Excel.
• Citation manager such as Zotero.
• Basic calculator for checking rankings and ratios.
• Optional, external mouse for easier molecule review.

Advanced Materials

• Access to Linux command line or a remote server for batch jobs.
• Python environment in Colab or locally with RDKit, pandas, NumPy, and matplotlib.
• xTB installed through a package manager or cloud notebook.
• GFN2-xTB input and output files for geometry optimization and descriptor extraction.
• Conformer search tools such as RDKit or CREST if available.
• Statistical analysis tools such as scipy, statsmodels, or R.
• ImageJ or a plotting tool for figure cleanup.
• Access to published receptor structure summaries from peer-reviewed papers.

Software & Tools

• Google Colab: Runs Python notebooks in the browser and lets you prototype xTB workflows without local setup.
• PubChem: Provides compound structures, identifiers, and linked chemical data for your analog set.
• RDKit: Builds, edits, and filters molecular structures before you send them into calculations.
• Python: Organizes your pipeline, stores descriptors, and makes plots and summary tables.
• Zotero: Keeps your papers, notes, and source links organized while you write.

Experiment Steps

1. Define the exact analog family you will compare and set rules for which structures count as close relatives.
2. Decide which computed descriptors will stand in for selectivity, then justify each one from prior literature.
3. Build a clean molecule table from public sources and check that each structure has the same naming logic.
4. Plan a consistent modeling workflow so each analog gets the same geometry optimization and scoring steps.
5. Design controls that test whether your ranking changes when you alter protonation, conformer choice, or descriptor set.
6. Predefine the statistics you will use to compare groups and rank candidate compounds.

Common Pitfalls

• Mixing compounds with different protonation states, which makes the same molecule look like several different candidates.
• Comparing raw xTB outputs without standardizing geometry, which turns conformer noise into fake trends.
• Building too many analogs at once, which makes the project hard to debug and hard to explain.
• Treating a proxy score like a real binding result, which overstates what the model can actually claim.
• Using inconsistent naming or structure files, which breaks your dataset and makes your ranking unreliable.

What Makes This Competitive

A stronger project goes beyond a simple ranked list. You can compare multiple proxy definitions, test whether the ranking survives a second conformer method, and check how sensitive your result is to structure cleanup choices. You can also relate your computed ladder to known medicinal chemistry patterns from the literature. That kind of careful validation shows that you understand both the chemistry and the limits of the model.

Project Variations

• Compare only oxygenated versus non-oxygenated analogs to see whether a hydroxyl group shifts the proxy ranking.
• Swap the target family to try tryptamine analogs with different N-substitutions and compare the ladder shape.
• Add a data analysis angle by training a simple classifier on your computed descriptors and testing which features matter most.

Learn More

• NIH PubChem: Search compound pages and 2D structure records for tryptamine and indole derivatives.
• NCBI PubMed: Search review articles on 5-HT2A, 5-HT2B, and psychedelic drug discovery.
• MIT OpenCourseWare: Look for physical chemistry and computational chemistry lecture notes that explain molecular energy and geometry.
• xTB Documentation: Read the method overview and input examples on the official xTB project pages.
• RDKit Documentation: Use the official tutorials to learn how to read, edit, and filter molecular structures.
• Google Colab Help: Find notebook guidance for running Python code in the browser.