Nanopore technology enables the sequencing of RNA at single-molecule resolution. To enable the translocation of RNA through the pore, standard library preparation protocols for nanopore direct RNA sequencing (DRS) ligate an adaptor and “motor” protein to the 3’ end of transcripts that are natively or synthetically 3’-polyadenylated (poly(A)+) RNA. The resultant libraries therefore typically contain a medley of RNA classes including messenger RNA (mRNA) and long non-coding RNAs (lncRNAs), amongst others.
Such a mixture can be detrimental to studies where only a specific RNA class is of interest, since unwanted RNAs consume the available working time of the nanopores and thus limit sequencing capacity. It is therefore imperative that for each nanopore the available sequencing time is concentrated on the RNA class of interest to maximize the sequencing depth of relevant transcripts. Thus far, efforts to target RNA sequencing have been limited to biochemical approaches, which require time-consuming and expensive specialized experimental protocols. They are also restricted in their applicability to a pre-determined set of specific transcripts.
Here we present RISER, the first biochemical-free technology for the enrichment or depletion of RNA classes in real-time during DRS. RISER identifies RNA classes directly from the start of DRS signals with a convolutional neural network, without the need for basecalling or a reference, and communicates with the sequencing hardware in real-time to enact biochemical-free targeted RNA sequencing. We illustrate RISER for the enrichment and depletion of coding and non-coding RNA, demonstrating a 3.4-3.6x enrichment and 6.2-6.7x depletion of non-coding RNA in live sequencing experiments of multiple cell lines. RISER’s modular design and retrainable deep learning model facilitates easy adaptation to other RNA classes. Finally, RISER is freely available to use through a simple and intuitive command-line tool to empower RNA researchers with biochemical-free, real-time targeted RNA sequencing.