Midterm Route M3: Drug Lookalikes

RouteID: M003
Wall: Small Molecule Representations (W04)
Grade: 5.10d (Midterm — Extra Credit)
Routesetter: Adrian
Date: 02/05/2026

The Setup

A medicinal chemist friend calls you in a panic:

"I'm working on a project and I need to find drugs that are structurally similar to a few molecules I'm interested in. I don't have access to expensive software — can you help me search a drug database using the fingerprint stuff you learned in class?"

You remember from R010 that molecular fingerprints encode structural features, and Tanimoto similarity can compare them. Time to scale up from a handful of steroids to a real drug database.

Your Mission

Load a database of FDA-approved drugs
Compute fingerprints for all molecules
Given query molecules, find the most similar drugs by Tanimoto similarity
Investigate whether structurally similar drugs have similar biological functions

Prerequisites

R009 (From SMILES to Molecular Properties) — SMILES parsing, RDKit basics
R010 (Morgan the Finger Printer) — fingerprints, Tanimoto similarity

Data Files

File	Description	Link
`FDA_Approved_structures.csv`	~2,500 FDA-approved drugs with names and SMILES	Download

Dataset credit: This dataset comes from Kaggle. Every serious student of AI/ML for biochemistry should get comfortable browsing Kaggle — it's a goldmine of curated datasets for practice and exploration.

Your Query Molecules

Find drugs similar to these well-known compounds:

Name	What it does	SMILES
Aspirin	Pain reliever, anti-inflammatory	`CC(=O)OC1=CC=CC=C1C(=O)O`
Caffeine	Stimulant	`CN1C=NC2=C1C(=O)N(C(=O)N2C)C`
Metformin	Diabetes medication	`CN(C)C(=N)NC(=N)N`
Ibuprofen	Pain reliever, anti-inflammatory	`CC(C)CC1=CC=C(C=C1)C(C)C(=O)O`

Exercise 1: Load the Drug Database

Goal: Load the approved drugs dataset and convert SMILES to RDKit molecules.

Load FDA_Approved_structures.csv into a DataFrame
Convert all SMILES to RDKit Mol objects
Handle any invalid SMILES gracefully (drop or flag them)
How many valid molecules do you have?

Success check:

You have ~2,500 valid Mol objects
No crashes from bad SMILES

Exercise 2: Compute Fingerprints

Goal: Generate fingerprints for all drugs in the database.

Choose a fingerprint type (Morgan recommended — you know it well from R010)
Pick a radius (r=2 is standard for drug-likeness)
Compute fingerprints for all molecules
Store them in a way you can use for similarity calculations

Hints:

This is the same process as R010, just more molecules
Consider storing fingerprints in a list or adding to your DataFrame

Success check:

Every valid molecule has a fingerprint
You can retrieve any molecule's fingerprint by index or name

Exercise 3: Build Query Fingerprints

Goal: Create Mol objects and fingerprints for your query molecules.

Parse the SMILES for Aspirin, Caffeine, Metformin, and Ibuprofen
Compute fingerprints using the same settings as Exercise 2
Verify they parsed correctly by drawing them

Success check:

All 4 query molecules have valid Mol objects and fingerprints

Exercise 4: Similarity Search

Goal: For each query molecule, find the most similar drugs in the database.

For each query, compute Tanimoto similarity to ALL drugs in the database
Rank drugs by similarity
Extract the top 10 most similar drugs for each query
Store results in a table: query name, drug name, similarity score

Hints:

DataStructs.TanimotoSimilarity(fp1, fp2) computes similarity between two fingerprints
You can loop, or be clever with vectorization

Success check:

You have top 10 hits for each of the 4 queries
Similarity scores are between 0 and 1
The query molecule itself (if in the database) should have similarity = 1.0

Exercise 5: Investigate Your Hits

Goal: Explore whether structurally similar drugs have similar functions.

For each query molecule:

Look at the top 5 hits (by name)
Look up what those drugs do (Google, Wikipedia, DrugBank — whatever works)
Do they share a therapeutic use with the query? Same drug class? Same target?

Questions to answer in your notebook:

For Aspirin's top hits — are they also pain relievers or anti-inflammatories?
For Caffeine's top hits — are they also stimulants or related?
Do any hits surprise you? Similar structure but very different use?

Write 3-4 sentences per query about what you found.

Exercise 6: Reflection

Goal: Connect structure to function.

Answer in your notebook (2-3 sentences each):

Based on your results, how well does "similar structure = similar function" hold up?
When might two drugs have similar fingerprints but different biological effects?

Deliverables

Submit your completed notebook (.ipynb) with:

All code cells executed
Top 10 similar drugs for each query
Your investigation of the hits
Reflection answers
A Logbook section at the end with [LOGBOOK] entries

Reminder: We've switched to including [LOGBOOK] entries directly in each route's notebook rather than in a separate file. Add your logbook entries as markdown cells at the end of this notebook.

Submission

Submit your notebook here