Concepts That Matter
Each concept is one screen, 5-8 minutes. Learn the fundamentals that drive modern computational drug discovery.
Molecular fingerprints are binary or count vectors that encode the presence of specific structural features in a molecule. Think of them as a unique 'barcode' for each chemical compound.
If two molecules have 90% fingerprint similarity, does that guarantee similar biological activity?
Graph Neural Networks represent molecules as graphs where atoms are nodes and bonds are edges. This preserves the natural structure of molecules, unlike fixed-length fingerprints that lose spatial information.
What chemical properties might GNNs capture that fingerprints miss?
Quantitative Structure-Activity Relationship models predict biological activity from molecular structure. However, high predictive power doesn't mean we understand the mechanism — correlation isn't causation.
If your QSAR model achieves 95% accuracy on test data but fails on new scaffolds, what went wrong?
A molecule can bind perfectly to its target but still fail as a drug. ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) determines whether a compound can actually work in the human body.
If your model predicts potency but fails ADMET, is it useful? Why or why not?