FIRE maps the genome’s two copies and finds a hidden layer of difference

A cell can hold two stretches of DNA with identical letters in different regulatory states. One is accessible to proteins that help control genes; its counterpart is less accessible. A new analysis makes those mismatches visible, molecule by molecule.[1]

The Cell study, announced on 15 October, uses a machine-learning method called FIRE. About 40% of the regulatory elements showing a difference between inherited copies lacked a DNA-sequence difference at that location. The denominator is these differently regulated elements—not the entire genome.[1]

Two copies, one misleading average

Consider an illustrative example. A pooled measurement says a regulatory element is accessible on half the DNA molecules examined. One explanation is that the copy inherited from one parent is usually open and the other copy usually closed. Another is that both copies alternate between states, each accessible half the time. The average is the same. The organisation beneath it is entirely different. Separating the inherited copies turns one blurry observation into two much more informative ones.

FIRE works with Fiber-seq, a technique that reads long individual DNA molecules. Nearby sequence differences can identify which inherited copy a molecule belongs to, even when the regulatory patch itself matches perfectly. That surrounding context is what makes it possible to investigate apparently identical patches separately.[3]

A molecular stencil and an AI filter

Fiber-seq, introduced in 2020, preserves an unusual pairing of information. An enzyme adds experimental chemical marks to accessible DNA, creating a record of the molecular packaging along each strand. Long-read sequencing then retrieves both the DNA letters and that accessibility pattern. DNA wound around nucleosomes—the protein spools that package it—and regions occupied by other proteins leave characteristic interruptions in the pattern.[4]

FIRE tackles the interpretation problem. An exposed gap can contain a regulatory element, but it can also be ordinary space between nucleosomes. The system evaluates patches and their surroundings to estimate which are regulatory. The authors’ methodology note describes a two-stage process: classify candidate patches on individual reads, then combine the observations into genomic regions with estimated false-discovery rates.[2]

Its training uses existing biochemical maps of accessible DNA and protein binding as starting labels, then iteratively refines the positive examples. This matters because a region identified as active in a pooled experiment need not be active on every individual molecule. The AI is learning to distinguish those individual states, where the bulk label alone would be misleading.[2]

The result is a change in the question researchers can ask. Instead of asking only whether a stretch of DNA looks accessible across a sample, they can ask which copy is accessible, on which molecules, and how that pattern relates to the sequence carried alongside it.

A difference with a memory

The team’s earlier public manuscript describes comparisons of immune and melanoma cell lines from the same individual, along with analyses across tissues and families. These comparisons help separate different explanations for a regulatory mismatch: an inherited sequence variant, a consistent parent-of-origin effect, or a change acquired during the history of a cell lineage. Those possibilities can look deceptively similar in a single measurement.[3]

Some regulatory differences persisted through cell division. That is a form of cellular memory: a state can be maintained as cells multiply.[1]

Persistence within a cell lineage is different from transmission between human generations. And matching sequence inside an element does not rule out every influence from elsewhere in the genome. The important finding is that local DNA letters alone leave part of the observed regulatory variation unexplained.[3]

The same approach can sharpen an explanation when a letter does matter. At a regulatory element near an immune gene, the researchers identified a celiac-associated variant affecting protein attachment. Changing that letter experimentally changed the element’s accessibility, according to the university’s account of the new study.[1]

A disease-associated DNA change becomes more informative when researchers can see what it does to regulation. A regulatory difference becomes more interesting when the expected sequence explanation is absent. FIRE gives both inquiries a clearer starting point: the physical state of an identified DNA molecule, rather than an average assembled from copies whose behaviour may diverge.

Report an error or suggest a correction