Mitochondria are best known for producing energy for cells, but they also contain their own DNA and produce their own RNA molecules. These mitochondrial RNAs must be carefully processed before they can function properly. Errors in this processing have been associated with mitochondrial dysfunction and human disease.

A research team from Northwestern Polytechnical University, China investigated how RNA sequencing data can be used to more accurately identify the locations where mitochondrial RNA molecules are cut during processing.

Detecting mitochondrial RNA cleavage

Human mitochondrial DNA produces long RNA molecules containing multiple genes. These transcripts are subsequently cut at specific locations to release individual RNAs, including messenger RNAs, ribosomal RNAs, and transfer RNAs.

Strand-specific RNA sequencing can capture evidence of these cleavage events. One potential source of information comes from soft-clipped reads. These occur when most of an RNA sequencing read aligns to the mitochondrial genome, but a small sequence at one end does not.

Although soft-clipping is common in sequencing data, researchers have lacked a systematic understanding of how these patterns relate to mitochondrial RNA processing.

The researchers analyzed strand-specific RNA sequencing data from 54 samples representing two different library preparation methods. They examined how frequently soft-clipping occurred, where it appeared within reads, and characteristics such as sequence length and GC content.

Sequencing methods influence soft-clipping

The analysis revealed that most soft-clipped sequences were short, generally between one and six nucleotides. They were also rich in guanine and appeared much more frequently at the 3′ ends of sequencing reads.

Importantly, the way RNA sequencing libraries were prepared had a major influence on these patterns. Random priming produced much higher levels of 3′ soft-clipping than non-random priming.

The researchers also found that allowing soft-clipping during sequence alignment improved sequencing depth and precision. These findings demonstrate that choices made during both library preparation and data analysis can influence the ability to detect mitochondrial RNA processing events.

Using machine learning to identify cleavage sites

To distinguish meaningful cleavage signals from background sequencing effects, the researchers developed a machine learning framework called MitoClipSplice.

The system uses a random forest model that evaluates multiple characteristics of soft-clipped reads to predict high-confidence mitochondrial RNA cleavage sites. The model achieved an F1 score above 0.85 and an area under the curve greater than 0.90, indicating strong predictive performance.

Interestingly, very short soft-clipped sequences containing only one or two nucleotides provided the strongest signal relative to background noise.

The researchers applied the model to RNA sequencing data from 20 hepatocellular carcinoma samples, demonstrating how the approach can be used to investigate mitochondrial RNA processing in biological samples.

A new way to examine mitochondrial RNA processing

Mitochondrial RNA processing plays an important role in maintaining normal mitochondrial function. Being able to precisely identify RNA cleavage sites can help researchers investigate how these processes are regulated and how they may change in disease.

MitoClipSplice provides a computational approach for extracting information that is already present within strand-specific RNA sequencing data but can be overlooked during conventional analysis. Combining soft-clipped sequencing reads with machine learning could provide researchers with a more accessible way to investigate mitochondrial RNA processing and post-transcriptional regulation.

Availability – The source code is available at https://github.com/Xing-laboratory/MitoClipSplice.