Skin, Decoded
← Contents

Lesson 02 · Where most variants act

The 98% that isn’t a gene

5 minute read

If you only knew about genes, most of the genome would look like filler. It is not. It is largely a control system.

Only around 2% of the human genome codes for protein. A large part of the remainder is regulatory: promoters that start a gene, enhancers that boost it, silencers that damp it down. These elements work by binding proteins that read the DNA sequence and decide what happens next.

The awkward part is distance. An enhancer can sit hundreds of thousands of letters away from the gene it controls, because the DNA folds into loops that bring the two together in three dimensions. Reading a variant in isolation — without knowing which gene it touches, in which cell — tells you very little.

Three places a variant can land

DNA GENE VARIANT

The letter change lands in the protein recipe

A variant inside a coding exon changes an amino acid, or stops the protein early. The effect is usually large and works the same way in every cell that uses that gene.

Example: a frameshift in FLG removes filaggrin protein from the epidermis.

Distance from gene0 letters from the gene
Cells affectedAll cells that use the gene
Where the letter change landsSelect a position

Why this was the hard problem

Variants in coding regions are comparatively easy: change a letter in a recipe, read the recipe. Variants in regulatory regions are where most of the difficulty sits, because the same letter change can:

The two ends of the spectrum

Most coding variants that matter are rare and have large effects. Most regulatory variants that matter are common and have small ones. Confusing the two is the single most common mistake in consumer genetics marketing.

This is the gap that a model like AlphaGenome was built for: given a DNA sequence and a cell type, predict what the regulatory readouts would look like, and how a variant would change them.

Check yourself

An enhancer is:

Why can one variant affect a single cell type?