Pattern recognition applied to life sciences.
WHAT IT IS USED FOR
Predicting protein structure Classifying cell types Predicting variant effects Image analysis of tissue and cells Drug candidate screening
WHAT STRUCTURE PREDICTION CHANGED
Predicting protein folding accurately, which was a decades-old open problem.
WHY THAT MATTERED
Structure determines function, and experimental determination is slow and expensive.
WHAT TO BE CAREFUL WITH
Data leakage, where related sequences appear in both training and test sets Class imbalance, common in biological data Models learning batch effects rather than biology Interpreting correlation as mechanism
WHY THE FIRST POINT IS SO COMMON HERE
Biological sequences are related by evolution, so random splitting leaves near-identical examples on both sides.
WHAT TO DO ABOUT IT
Split by similarity, not randomly.
WHAT INTERPRETABILITY MATTERS FOR
Anything informing a clinical or experimental decision.
WHY
A prediction nobody can explain cannot be acted on responsibly.
WHAT TO VALIDATE AGAINST
Independent data, ideally from a different source.
WHAT TO NEVER CLAIM
Clinical utility without clinical validation.
