← Latest papers
💻 bioinformatics

RawMod: Modification-Agnostic Detection of DNA Base Modifications from Nanopore Signal Pileups

RawMod is a novel, control-free, transformer-based method that achieves modification-agnostic detection of diverse DNA base modifications from nanopore signals by learning a unified representation of modified sites, outperforming existing models in accuracy and enabling applications like label-free cell type classification.

Original authors: Srinivasan, B., Nazar, S., Gandhi, V., Zhang, E., Bachvaroff, T., Kolmogorov, M., Place, A., Firtina, C.

Published 2026-11-11
📖 6 min read🧠 Deep dive

Original authors: Srinivasan, B., Nazar, S., Gandhi, V., Zhang, E., Bachvaroff, T., Kolmogorov, M., Place, A., Firtina, C.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Life is written in a code of four letters: A, C, G, and T. These letters form the DNA that guides every living thing, from the smallest bacteria to the largest whale. For decades, scientists have focused on reading this sequence, treating the letters as the sole source of biological instruction. However, the cell often adds chemical tags to these letters, like sticky notes on a page of text. These tags, known as base modifications, do not change the spelling of the genetic code, but they alter how the cell reads it. They can turn genes on or off, determine whether a cell becomes part of a liver or a brain, and help bacteria defend themselves against viruses. Understanding these tags is crucial for unlocking the secrets of cancer, development, and evolution.

To read these tags, scientists use a technology called nanopore sequencing. Imagine a tiny tunnel through which a single strand of DNA is pulled. As the strand moves through, it disrupts an electrical current flowing inside the tunnel. The pattern of this disruption is unique to the specific letters passing through, allowing a computer to translate the electrical signals back into the DNA sequence. But the story does not end there. When a chemical tag is attached to a letter, it changes the shape of the DNA slightly, which in turn changes the electrical signal in a subtle but detectable way. The challenge has always been that these changes are tiny, and the signals are noisy. To figure out where a tag exists, researchers have traditionally needed two things: a specific computer model trained to recognize that exact tag, and a "control" sample of the same DNA that has been stripped of all tags for comparison. This requirement makes it incredibly difficult to study rare or unknown tags, especially in organisms where creating a tag-free control is impossible.

A team of researchers has now developed a new approach that breaks these barriers. They created a tool called RawMod, which acts as a universal detector for these chemical tags. Unlike previous methods that act like specialized flashlights, only illuminating the specific tag they were built to find, RawMod is designed to spot any modification, regardless of what it is. It does this by comparing the raw electrical signals from many copies of the same DNA strand against the expected signal of a standard, untagged sequence. Instead of needing a control sample from the same organism, the tool learns what a "normal" signal looks like by studying a vast array of different DNA types and modification patterns. It treats every unknown chemical change as a single category of "modified," allowing it to flag a site as altered without ever needing to know the specific name of the tag causing the change.

The researchers tested this system on five different types of modifications found across bacteria, viruses, plants, and humans. In every case, RawMod proved more accurate than the best existing tools, even when those tools were combined. It achieved an AUC of 0.964 in identifying where modifications occurred, compared to roughly 80 percent for the next best method. Remarkably, the tool is also incredibly efficient, using a fraction of the computing power required by its competitors. Perhaps most significantly, the system demonstrated a form of learning that allows it to generalize. When the researchers trained the model on four types of tags and then tested it on a fifth type it had never seen before, it still performed well, provided that the new tag shared a chemical relationship with one of the tags it had studied. For instance, if the model learned to recognize changes on a specific type of base, it could often detect a new, unseen change on that same base.

This capability opened the door to mapping modifications in places where no one had looked before. The team applied RawMod to the genome of a microscopic marine organism called Amphidinium carterae. This organism is known to replace a large portion of its standard DNA letters with a rare tag called 5hmU, but scientists had no way to map exactly where these replacements occurred without a difficult-to-produce control sample. Using RawMod, the researchers generated a detailed map of these tags across a large stretch of the organism's genome. The map confirmed two patterns that had been suggested by previous, less precise studies: the tags are found more often in non-coding regions of the DNA and in younger genetic elements. More importantly, the new map revealed a specific pattern that had never been observed before, suggesting that these tags cluster at specific junctions where gene arrays point in opposite directions. This discovery provides a new hypothesis for how these organisms organize their genetic material.

Beyond mapping, the tool offers a new way to understand the origin of DNA samples. Because the pattern of tags can vary between different cell types or between healthy and diseased tissue, the researchers showed that RawMod could distinguish the source of a DNA strand without needing to know the specific tags or having a labeled dataset. By analyzing the subtle electrical signatures, the system could tell the difference between DNA from a tumor and DNA from normal tissue, or between DNA from a virus and a laboratory copy of that virus. This suggests that the tool can act as a fingerprint for the biological history of a sample, revealing its story even when the specific details of its chemical tags are unknown.

The work represents a shift in how scientists approach the hidden layer of genetic information. By moving away from the need for specific controls and specialized models, RawMod provides a flexible, powerful lens for observing the chemical landscape of DNA. It allows researchers to ask questions about rare and unknown modifications that were previously unanswerable, offering a clearer view of the complex mechanisms that drive life. While the tool does not yet identify the exact name of every tag it finds, its ability to flag their presence with high accuracy across diverse organisms marks a significant step forward in the study of the epigenome.

Drowning in papers in your field?

Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.

Try Digest →