WattmaMod enables high-resolution and extensible RNA modification profiling for nanopore direct RNA sequencing

Imagine your cell's genetic instructions (DNA) as a master blueprint, and the RNA molecules as the active workers carrying those instructions to the construction site. Sometimes, these RNA workers get tiny "sticky notes" or "highlighters" attached to them—these are called RNA modifications. These notes change how the worker behaves, telling the cell to speed up, slow down, or change its task.

For a long time, scientists have had a tool called Nanopore sequencing that acts like a high-speed conveyor belt. As the RNA strand zips through a tiny hole, it creates a unique electrical hum. Theoretically, if you listen closely to that hum, you can hear where the "sticky notes" are attached. However, in reality, the signal is messy. It's like trying to identify specific instruments in a chaotic jazz band where the volume keeps changing and the musicians switch instruments mid-song. Different types of "sticky notes" often sound the same, and the signal changes depending on which version of the conveyor belt (chemistry) you are using.

Enter WattmaMod, a new computer program designed to be the ultimate "sound engineer" for this messy jazz band.

Here is how it works, using simple analogies:

• The Smart Student (Self-Supervised Pretraining): Before WattmaMod tries to identify specific sticky notes, it spends a long time just listening to thousands of hours of raw RNA signals without any labels. It's like a music student listening to a million songs to learn the general "feel" of the music, rather than just memorizing the lyrics. This helps it understand the underlying patterns of the signal.
• The Focused Coach (Supervised Contrastive Fine-Tuning): Once it has the basics down, a human coach steps in to show it specific examples. The coach doesn't just say, "This is a sticky note." Instead, the coach says, "Look at how this sticky note sounds different from that one." This helps the program learn to tell very similar sounds apart, even when they are subtle.
• The Quick Learner (Low-Label Incremental Adaptation): Sometimes, scientists discover a new type of sticky note but only have a few examples of it. Most programs would get confused. WattmaMod, however, is like a student who can learn a new dance move after seeing it just a few times. It can adapt to new, rare modifications without needing a massive library of examples.
• The Multi-Lens Camera (Wavelet-guided Multi-scale Encoding): To make sure it doesn't miss anything, WattmaMod looks at the signal through different "lenses." One lens zooms in on the tiny, split-second electrical spikes (the fine details), while another lens zooms out to see the broader shape of the wave. It then uses a Dynamic Cross-Attention mechanism—think of it as a conductor in an orchestra—to decide which part of the signal is most important at any given moment, blending the fine details with the big picture.

What did they find?

The results show that WattmaMod is incredibly good at identifying a long list of different "sticky notes" (including m6A, m5C, m1A, and others) all at once. It doesn't just work on one type of RNA or one version of the sequencing machine; it works across different species and different experimental setups.

Perhaps most excitingly, WattmaMod doesn't just look at one sticky note in isolation. It can spot how different notes might be organizing themselves in groups, like seeing a pattern where three specific notes always appear together in a specific order.

In short, WattmaMod turns a noisy, confusing electrical signal into a clear, high-resolution map of RNA modifications, allowing scientists to see the "highlighted" instructions on the RNA workers with much greater clarity than before.