RNA molecules do much more than carry genetic instructions from DNA to the cellular machinery that produces proteins. Many RNAs fold into complex shapes that influence how genes are regulated, how proteins are produced, and how cells respond to changes in their environment.

Determining these shapes, known as RNA secondary structures, is therefore an important part of understanding RNA biology. However, predicting how an RNA molecule folds can be challenging, especially when scientists want to examine the structures of individual RNA molecules rather than measure an average across thousands of molecules.

Researchers at Aalto University have introduced a computational method called segSHAPE that is designed to improve RNA secondary structure prediction using Nanopore direct RNA sequencing data.

Reading RNA molecules directly

Nanopore direct RNA sequencing works by passing individual RNA molecules through tiny protein pores. As each section of an RNA molecule moves through the pore, it produces a measurable electrical signal. Differences in this signal can provide information about the RNA sequence and chemical changes present on the molecule.

Scientists can combine this sequencing technology with chemical probing. Chemical probes react more strongly with RNA bases that are exposed and less strongly with bases that are paired or buried within a folded structure. These chemical modifications alter the Nanopore signal, giving researchers clues about the shape of the RNA.

In principle, this approach can reveal RNA structure at the level of individual molecules. In practice, accurately connecting each electrical signal to the correct position in the RNA sequence remains difficult.

Signal alignment is important

Nanopore sequencing produces a continuous electrical signal rather than a direct series of letters. Computational tools must divide that signal into smaller sections and determine which RNA bases produced each part.

Small alignment errors can cause a signal change to be assigned to the wrong nucleotide. This affects estimates of where chemical modifications occurred and can ultimately lead to incorrect predictions about RNA structure.

Individual sequencing reads may also have slightly different baseline signal levels. Without correcting for these shifts, normal signal variation can be mistaken for evidence of a chemical modification.

segSHAPE was designed to address these problems. It uses information from the initial basecalling process to improve signal alignment and corrects baseline signal shifts separately for each RNA molecule.

Learning what normal signals look like

The method also learns the expected electrical signals produced by short combinations of RNA bases, called k-mers. Because the surrounding sequence affects the signal generated by each nucleotide, learning position-specific k-mer patterns helps the software distinguish normal variation from changes caused by chemical probes.

After establishing the expected signal at each position, segSHAPE uses an unsupervised anomaly detection system. This type of analysis identifies unusual signals without requiring researchers to label every modified and unmodified nucleotide beforehand.

The resulting modification rates are then used to predict which RNA bases are paired and which are exposed, producing a model of the molecule’s secondary structure.

Supporting multiple probes and sequencing chemistries

An important feature of segSHAPE is that it is probe-agnostic. This means it is not restricted to one particular chemical probing method.

The researchers evaluated the method using public datasets created with several chemical probes, including AcIm, NAI-N3, and DEPC. They also tested it with both the RNA002 and newer RNA004 versions of Oxford Nanopore’s direct RNA sequencing chemistry.

Across three RNA002 datasets containing RNA molecules ranging from 421 to 1,552 nucleotides, segSHAPE produced the strongest overall structure predictions. It improved the Matthews correlation coefficient, a measurement of prediction quality, by between 3.4 and 5.8 percentage points compared with the strongest existing method.

On an RNA004 dataset, segSHAPE improved modification-rate detection by 17 ROC-AUC points and RNA structure prediction by 6.7 Matthews correlation coefficient points compared with the sm-PORE-cupine method.

Detecting RNA structural changes

RNA structures are not always fixed. Some RNA molecules change shape when they bind to specific compounds.

The researchers tested whether segSHAPE could detect a structural change in the thiamine pyrophosphate riboswitch. Riboswitches are sections of RNA that bind small molecules and change shape, allowing them to control gene activity.

Using Nanopore direct RNA sequencing data and the DEPC chemical probe, segSHAPE captured the structural change that occurred when the riboswitch interacted with its ligand. This demonstrates how the method could help researchers examine dynamic RNA structures and determine how they respond to biological signals.

Expanding single-molecule RNA analysis

Improved RNA structure prediction could help researchers investigate how RNA folding affects gene regulation, cellular signaling, and disease. It may also support the development of RNA-based medicines, where structure can influence the stability and biological activity of a therapeutic molecule.

By combining improved signal alignment, baseline correction, sequence-specific signal modeling, and automated modification detection, segSHAPE provides a unified approach for analyzing several types of chemical probing data.

The findings suggest that Nanopore direct RNA sequencing can provide more accurate structural information when raw electrical signals are carefully aligned and interpreted. As Nanopore technology and computational methods continue to improve, single-molecule RNA structure analysis may reveal forms and structural changes that are hidden by traditional measurements based on averaged samples.