A team led by researchers at the Robert Koch Institute has developed a powerful new tool to make RNA sequencing using nanopore technology faster, more accurate, and more affordable. The tool, called WarpDemuX, solves a common problem in direct RNA sequencing (dRNA-seq): how to efficiently sort different RNA samples that have been combined in a single sequencing run.

In RNA sequencing, researchers often use “barcodes”—short sequences attached to each sample—to tell them apart. But when working with nanopore sequencing, identifying these barcodes quickly and accurately has been a challenge, especially when using Oxford Nanopore’s SQK-RNA002 and SQK-RNA004 chemistries. WarpDemuX tackles this by analyzing the raw electrical signals from the nanopore directly, rather than waiting until after the RNA has been turned into base sequences. It also uses a lightweight machine-learning algorithm and specially optimized barcode designs.

WarpDemuX model

a Generation of training data using dual-barcoded RNA molecules containing an in-line RNA barcode positioned at the 3’ end of the transcript, serving as a ground truth label, and an RTA barcode, introduced during standard dRNA-seq library preparation, which served as the target for classification. b The direct RNA sequencing raw signal denotes a time series of electric current measurement. The signal (blue) consists of three main components: the DNA adapter signal, the RNA poly(A) tail, and the RNA body. The last part of the adapter signal contains the RTA barcode to be classified (red). c The adapter signal is segmented into events, with each segment represented by its average signal. The final 25 segments of the adapter are designated as the RTA barcode-specific fingerprint. Dynamic Time Warping Distance (DTWD) is used to quantify the similarity between these fingerprints, with higher similarity (lower DTWD) observed within the same barcode class compared to different barcode classes. d Model training and classification of barcode fingerprints into barcode classes using the DTWD kernel function. Training of the Support Vector Machine (SVM) classifier with the DTWD kernel function on barcode fingerprints from sequencing reads captures the nuances of the barcode signals and promotes robust, noise-resilient classification. e, f Performance of models. Values reported reflect the recall (confusion matrix, normalized over the actual class; true label), on both test sets combined, evaluated at a confidence cutoff of 0.5. (e) DeePlexiCon (DPC) CNN model. f WarpDemuX SVM-DTWD. Source data are provided as a Source Data file.The team demonstrated WarpDemuX’s power by sequencing RNA from different SARS-CoV-2 virus samples over time on a single flowcell. They could detect changes in RNA abundance and poly(A) tail lengths—both important markers of viral behavior during infection. Even more impressively, WarpDemuX enabled real-time “adaptive sampling,” meaning it could enrich for hard-to-detect RNA targets during the sequencing run itself.

Overall, WarpDemuX could make RNA sequencing more accessible and informative for a wide range of biological and medical research, including studies on viruses, gene regulation, and transcriptome dynamics.