The challenge that motivated the work is rooted in the physics of nanopore sequencing. As an RNA or DNA molecule threads through a protein pore, the sequencer records characteristic disruptions in ionic current, and basecalling algorithms translate those disruptions into sequence. Homopolymer runs, long stretches of a single base such as the adenines in a poly(A) tail, produce nearly identical current signals base after base, making it hard to count exactly how many adenines are present. On top of that, the 3′ ends of transcripts are architecturally complex: the boundary between the end of the transcript proper and the start of the tail can be fuzzy, alternative polyadenylation sites can generate multiple transcript isoforms from a single gene, and repetitive elements, including remnants of endogenous retroviruses, can complicate the identification of where a transcript truly terminates.

PolyAnalysis tackles these problems at the signal level rather than working from basecalled sequence alone. The framework combines a convolutional neural network with a bidirectional long short-term memory network, a pairing commonly abbreviated as CNN–BiLSTM, to classify regions of the raw signal as poly(A)-proximal or otherwise. It then applies a multi-task connectionist temporal classification decoding scheme, adapted from speech recognition, to infer the boundaries of the poly(A)-proximal region and the tail length itself. Crucially, the authors built adenine-aware training and decoding into the pipeline: the loss function used during training weights adenine-related errors differently, and the decoding step is biased toward A-rich interpretations, reflecting the biological expectation that the region of interest is dominated by adenines.

Benchmarking was carried out against three established tools: Nanopolish, tailfindr, and Dorado. The evaluation used controlled datasets generated with both RNA002 and RNA004 nanopore chemistries, allowing the team to assess performance across sequencing generations. According to the study, PolyAnalysis achieved tail-length accuracy that was competitive with the existing methods while delivering high callability, meaning it produced usable estimates for a larger fraction of reads. The authors stress that tail-length estimation is the most extensively validated output of the framework, and they are careful not to overstate the maturity of the other analyses the tool offers.

To understand which components of the model actually mattered, the researchers performed component-wise ablation experiments, systematically removing or altering parts of the pipeline and measuring the effect on performance. They also ran bias-control analyses. The results indicated that the region classification module, the multi-task optimization objective, the adenine-weighted loss, and the A-rich decoding strategy each contributed to the overall performance, suggesting that the design choices were not arbitrary but jointly responsible for the accuracy gains. This kind of ablation work is increasingly expected in machine learning applied to genomics, where it is easy to attribute improvements to a single flashy component when in fact the benefit comes from an interplay of design decisions.

Beyond tail length, PolyAnalysis links its per-read estimates to analyses of polyadenylation sites, alternative polyadenylation, and repetitive-element-associated 3′ ends. When applied to human cell-line datasets and vertebrate transcriptome datasets, the framework identified variation in tail-length distributions at the dataset level and heterogeneity at the gene level. It also found only weak associations between tail length and steady-state transcript abundance, a finding that adds nuance to the widely held expectation that longer tails should reliably predict more abundant or more translated transcripts. The relationship between tail length and expression, the data suggest, is more context-dependent than simple models imply.

One of the more distinctive features of the framework is its confidence-based filtering of polyadenylation sites. Rather than presenting all detected sites as equally reliable, PolyAnalysis separates them into database-matched high-confidence sites, novel candidates flagged with high confidence, and low-confidence calls. This tiered output acknowledges a persistent problem in 3′ end profiling: novel polyadenylation sites predicted from sequencing data can reflect genuine biology or merely artifacts of alignment, basecalling, or annotation gaps. By making confidence explicit, the tool gives downstream users a principled way to decide which calls to trust for follow-up experiments.

The treatment of endogenous retroviruses illustrates the same caution. Retroviral remnants scattered through the genome can supply polyadenylation signals to nearby genes, and distinguishing true locus-level evidence of such events from family-level patterns, where many similar retroviral sequences produce ambiguous signals, is genuinely difficult. PolyAnalysis applies ambiguity-aware analysis to separate these two levels of evidence, and the authors are explicit that the sequence-composition, polyadenylation-site, alternative-polyadenylation, and endogenous-retrovirus-related results all remain dependent on model assumptions, sequencing chemistry, and annotation quality, and require further orthogonal validation before being treated as established biology.

The significance of the work lies partly in its scope. Alternative polyadenylation is a major layer of gene regulation: by choosing a proximal or distal polyadenylation site, a cell can change the untranslated region of a transcript and thereby alter its stability, localization, and translation efficiency, with documented roles in development, immunity, and cancer. Tools that can profile these choices per read, on long-read platforms that capture full transcript molecules, offer a view that short-read sequencing cannot easily provide. By coupling tail-length estimation with polyadenylation-site analysis in a single signal-level framework, PolyAnalysis aims to make that view more accessible to laboratories that already run nanopore sequencers.

The study, a software contribution from researchers at Hunan University, Hunan University of Finance and Economics, and the University of Electronic Science and Technology of China, was supported by the National Natural Science Foundation of China and involved secondary analysis of publicly available datasets. The framework is published open access under a Creative Commons Attribution license, and the authors have released supplementary figures, tables, and numerical source data underlying the main results. For a field where poly(A) tail measurement has often been a bespoke, error-prone exercise, a benchmarked, confidence-aware pipeline that works across both RNA002 and RNA004 chemistries represents a practical step forward, even as the authors themselves caution that the most exploratory parts of their analysis, particularly the retrovirus-associated findings, should be treated as hypotheses awaiting independent confirmation rather than settled results.

Subject of Research: Deep learning-based estimation of poly(A) tail length and 3′-end regulation profiling from nanopore sequencing data

Article Title: PolyAnalysis: a deep learning–based framework for nanopore-based poly(A) tail estimation and 3′-end regulation profiling

Article References: Tian, Q., Song, B., Zou, Q., & Wang, Y. (2026). PolyAnalysis: a deep learning–based framework for nanopore-based poly(A) tail estimation and 3′-end regulation profiling. BMC Biology. https://doi.org/10.1186/s12915-026-02749-7

Image Credits: AI Generated

DOI: 10.1186/s12915-026-02749-7

Keywords: poly(A) tail, nanopore sequencing, deep learning, alternative polyadenylation, polyadenylation site, long-read transcriptomics, connectionist temporal classification, endogenous retrovirus, RNA sequencing, 3′ end regulation, Oxford Nanopore Technologies, BMC Biology

Nanopore sequencing has transformed genomics by allowing DNA and RNA molecules to be read in their entirety, producing long reads that span entire transcripts and reveal complex splicing patterns that short-read technologies often miss. Oxford Nanopore platforms thread individual molecules through protein nanopores and infer their sequence from disruptions in an electrical current. This approach delivers reads that can stretch tens of thousands of bases, capturing full-length transcript isoforms and preserving modifications that are lost when RNA is fragmented and amplified. Yet the technology has long been hampered by a higher raw error rate compared with Illumina short-read sequencing, and this limitation has been particularly consequential for SNP calling, where a single misread base can be mistaken for a genuine variant.

Conventional approaches to correcting these errors typically rely on consensus building, aligning many reads to a reference genome and counting the frequency of each base at every position. Such methods work reasonably well for DNA sequencing at high coverage, but transcriptome data presents unique complications. Expression levels vary enormously across genes, meaning some transcripts are covered by thousands of reads while others yield only a handful. Low-coverage transcripts offer too few observations for statistical consensus methods to separate signal from noise, and highly expressed genes can harbor natural RNA editing events that complicate variant interpretation further. The result has been a landscape in which SNP detection from nanopore transcriptome data has demanded heavy computational filtering and often produced inconsistent results across tools and datasets.

NanoTS takes a fundamentally different approach by framing SNP calling as a pattern recognition problem that deep learning is exceptionally well suited to solve. Rather than relying solely on base-level alignments, the tool learns from the rich features embedded in nanopore signal data, the raw electrical current measurements that underlie every base call. Neural networks can detect subtle signatures that distinguish systematic sequencing errors, which tend to follow repeatable patterns tied to sequence context and pore dynamics, from genuine genetic variation, which manifests consistently across independent molecules. This capacity to model complex, high-dimensional relationships in the data allows NanoTS to make confident genotype calls even in regions where traditional statistical methods falter.

The implications of accurate SNP calling from direct RNA sequencing extend across a wide range of biological and biomedical applications. Variant detection at the transcript level reveals which alleles are actively expressed, a dimension invisible to DNA-based genotyping. Allele-specific expression, in which one copy of a gene is transcribed more abundantly than the other, plays a critical role in imprinted genes, autoimmune disease, cancer biology and X-chromosome inactivation. Researchers studying these phenomena have historically needed to combine DNA genotyping with RNA expression analysis from separate experiments, introducing uncertainty about whether observed expression patterns truly reflect the same cells and conditions. A tool that can call genotypes directly from transcriptome data collapses these two measurements into a single, internally consistent dataset.

The deep learning architecture at the heart of NanoTS represents part of a broader movement in genomics toward learned models that outperform hand-crafted algorithms. Basecalling, the process of converting raw nanopore current signals into nucleotide sequences, was revolutionized years ago by neural networks that replaced earlier hidden Markov model approaches, dramatically improving read accuracy. Variant calling is now following a similar trajectory. DeepVariant, developed by Google, demonstrated that convolutional neural networks could outperform classical statistical variant callers on Illumina data by treating genomic alignment piles as images. NanoTS extends this paradigm to the specific and demanding context of long-read transcriptome sequencing, where error profiles, coverage distributions and biological variation patterns differ substantially from genomic DNA sequencing.

Training such a model requires data of exceptional quality, since the network must learn to associate signal and alignment patterns with known ground-truth genotypes. The developers of NanoTS tackled this challenge by curating training examples in which true variants could be established with high confidence, allowing the network to internalize the distinctions between authentic polymorphism and sequencing artifact. Once trained, the model generalizes across genes, transcripts and coverage regimes, producing variant calls that remain reliable even for transcripts sequenced at low depth, a scenario that has historically been the Achilles heel of long-read transcriptome analysis.

The clinical dimension of this work is particularly compelling. Personalized medicine increasingly depends on knowing both which variants a patient carries and how those variants are expressed in relevant tissues. RNA sequencing of patient samples, whether from tumor biopsies or blood cells, is already routine in many clinical and research settings. If those same RNA reads can yield accurate genotype information without a parallel DNA sequencing run, the cost and complexity of comprehensive molecular profiling drops considerably. Cancer genomics stands to benefit substantially, as tumor transcripts carry both expressed mutations and the allele-specific expression patterns that reveal loss of heterozygosity and other genome alterations relevant to treatment decisions.

Beyond clinical applications, NanoTS opens doors for basic research into population genetics and evolutionary biology at the transcript level. Understanding how genetic variation shapes transcript diversity across individuals and species requires tools that can reliably measure both. Long-read transcriptome sequencing captures the full complexity of isoform expression, and with accurate SNP calling layered on top, researchers can begin to map how specific variants influence splicing, expression levels and transcript structure in ways that short-read studies have only approximated. The combination of full-length isoform resolution and direct genotype calling provides a uniquely integrated view of the relationship between genome and transcriptome.

The publication of NanoTS in Nature Methods underscores the growing recognition that computational innovation is as essential to unlocking the potential of new sequencing technologies as the hardware itself. As nanopore sequencing continues its rapid improvements in accuracy and throughput, tools like NanoTS ensure that the analytical pipeline keeps pace, converting noisy raw signals into biologically meaningful and clinically actionable insights. For laboratories worldwide working with direct RNA sequencing, the arrival of a robust deep learning solution for SNP calling marks a significant step toward making long-read transcriptomics a complete, standalone platform for both discovery and diagnosis.

Subject of Research: A deep learning tool for accurate SNP calling and genotype detection in nanopore long-read transcriptome sequencing data.

Article Title: NanoTS: a deep learning tool for accurate SNP calling in nanopore long-read transcriptome data

Article References: NanoTS: a deep learning tool for accurate SNP calling in nanopore long-read transcriptome data. (n.d.). https://doi.org/10.1038/s41592-026-03225-4

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03225-4

Keywords: NanoTS, deep learning, SNP calling, nanopore sequencing, long-read transcriptomics, genotyping, direct RNA sequencing, variant detection, neural networks, Nature Methods, allele-specific expression, computational genomics