The genome's tangled DNA is tripping up AI

A new Quanta Magazine piece drills into why AI keeps stumbling on the human genome — and it's not because the models are dumb, but because they're looking at the wrong thing.

Most AI models treat DNA as a linear string of letters — the A, C, T, G sequence — the way they treat words in a sentence. But the genome isn't a string. It's a three-dimensional structure, folded and twisted inside the cell nucleus. Genes that sit far apart on the linear sequence can be right next to each other in physical space. That folding matters because it determines how genes interact, switch on, and respond to signals.

When AI ignores the 3D geometry, it misses the mechanics. A model might spot a pattern in the base-pair sequence and call it a finding — when the real action is happening in the folded structure, not the flat string. The fix is getting models to learn spatial relationships, not just sequence patterns. It's a small shift in how AI approaches the problem, but it's the difference between guessing and understanding.

Why this matters for us: The same pattern-recognition models powering everything from medical diagnostics to ancestry tools will be more reliable when they stop treating DNA like a sentence and start treating it like the tangled mess it actually is.