henhaobio
Sign in

bioRxivGenomicsPreprint

A technology-agnostic long-read analysis pipeline for transcriptome discovery and quantification

Dana WymaniD, Gabriela Balderrama-GutierreziD, Fairlie ReeseiD, Shan JiangiD, Sorena RahmanianiD, Stefânia ForneriD, Dina P. MatheosiD, Weihua ZengiD, Brian A. WilliamsiD, Diane E. TroutiD, Whitney EnglandiD, Shu‐Hui ChuiD, Robert C. SpitaleiD, Andrea J. TenneriD, Barbara J. WoldiD, Ali MortazaviiD

PDFDOIData
In-depth full-text digest available · Jump to digest ↓

Abstract

Alternative splicing is widely acknowledged to be a crucial regulator of gene expression and is a key contributor to both normal developmental processes and disease states. While cost-effective and accurate for quantification, short-read RNA-seq lacks the ability to resolve full-length transcript isoforms despite increasingly sophisticated computational methods. Long-read sequencing platforms such as Pacific Biosciences (PacBio) and Oxford Nanopore (ONT) bypass the transcript reconstruction challenges of short reads. Here we introduce TALON, the ENCODE4 pipeline for platform-independent analysis of long-read transcriptomes. We apply TALON to the GM12878 cell line and show that while both PacBio and ONT technologies perform well at full-transcript discovery and quantification, each displayed distinct technical artifacts. We further apply TALON to mouse hippocampus and cortex transcriptomes and find that 422 genes found in these regions have more reads associated with novel isoforms than with annotated ones. We demonstrate that TALON is a capable of tracking both known and novel transcript models as well as their expression levels across datasets for both simple studies and in larger projects. These properties will enable TALON users to move beyond the limitations of short-read data to perform isoform discovery and quantification in a uniform manner on existing and future long-read platforms.

In-depth summary

Claude Opus 5.5 · xhigh effort · 17 Oct 2026 · from the full text of version 2 (24 Mar 2020) · CC BY-ND 4.0

TALON annotates and counts full-length reads from PacBio or Nanopore in one framework; in mouse brain, 422 genes showed more novel-isoform reads than annotated ones.

Short reads cannot say which exon combination was in a given molecule, and the long-read tools that can were built per platform, cluster reads before deciding what is present, and mostly rely on PacBio-specific upstream software. TALON instead assigns each read to known or previously seen models by its splice junctions and ends, so the same pipeline tracks isoforms and their counts across PacBio, direct-RNA Nanopore and successive datasets within ENCODE4.

Key results

  1. ShownLong reads alone give reproducible gene and transcript quantification in GM12878, agreeing with short-read RNA-seq above a low expression floor.

    PacBio Sequel II, two replicates: 15,727 known GENCODE genes and 26,841 known transcripts detected; gene expression Pearson r = 0.97, Spearman rho = 0.92 between replicates; Spearman rho 0.78 against Illumina across 14,947 shared genes, with upwards of 70% of genes above 2 TPM in short reads reproducibly detected, and no significant platform difference for 87% of GENCODE transcripts. · Performance of TALON on human ENCODE Tier 1…

  2. ShownThe two platforms fail differently: conversion to cDNA for PacBio apparently generates spurious antisense transcripts, while direct-RNA Nanopore accumulates truncated incomplete-splice-match (ISM) models.

    Of 414 antisense transcripts called across platforms, 276 were PacBio-only against 26 ONT-only, and antisense models had scant CAGE support; ISMs were overrepresented in ONT, attributed to degradation or pores stopping mid-molecule. 88 genes with TPM > 10 in both were over 10-fold higher in ONT, among them mitochondrial genes (MT-RNR1, MT-CO2, MT-ATP6, MT-CO3, MT-CYB). · Performance of TALON on Oxford Nanopore data and…; Assessing completeness of TALON transcript…

  3. ShownOrthogonal end data separate trustworthy novelty from artifact: NIC and NNC models match annotated transcripts for 5' and 3' support, while ISM classes are supported at one end only.

    PacBio GM12878 transcripts with CAGE support: 76% of known, 94% of NIC, 87% of NNC, 97% of prefix ISMs but 34% of suffix ISMs. A poly(A) motif in the last 35 bases was present for 64% of known, 84% of NIC, 86% of NNC and 86% of suffix ISMs, against 43% of prefix ISMs. RNA-PET support was strong for NIC and NNC in both platforms, and extremely high for antisense transcripts. · Assessing completeness of TALON transcript…

  4. ShownIn mouse cortex and hippocampus, hundreds of genes have more reads on novel (NIC/NNC) isoforms than on annotated ones, differences the authors say would be difficult to recover from short reads alone.

    Two PacBio replicates per region at a minimum of 6 million raw reads each, averaging 17,000 known genes and 26,000 known transcripts per tissue: 422 genes had more reads on novel than annotated isoforms (352 shared, plus 29 cortex-only and 41 hippocampus-only); 694 transcripts from 612 genes differed between regions (log fold-change > 1, adjusted p < 0.01), of which 607 known and 87 novel. · Comparison of PacBio transcriptomes of mouse…; Abstract

  5. ShownRun on the same reads, TALON recovered more annotated genes and transcripts than FLAIR on both platforms, while replicate reproducibility was comparable.

    On GM12878 PacBio, TALON detected 2,525 more GENCODE genes than FLAIR among genes also expressed in the matched short-read data; FLAIR likewise detected fewer known genes and transcripts in direct-RNA ONT, with the gap widest at low expression but still present above 50 TPM. FLAIR replicate correlations were Pearson r = 0.96 and Spearman rho = 0.94 for known genes. Precision or false discovery was not compared. · Comparison of TALON and FLAIR on GM12878 Pacbio…

Methods, data and reproducibility

Reads are aligned with Minimap2, corrected by TranscriptClean, labelled for internal-priming risk from the fraction of As after the alignment, then matched one by one to known and previously seen models on splice junctions plus start and end, taking SQANTI-style novelty classes; novel models need at least 5 reads in two replicates, and expression is read counts as TPM.

  • GM12878 PacBio Sequel II long reads, two biological replicates, one SMRT cell each · human · n = at least 6 million raw reads per replicate
  • GM12878 direct-RNA Oxford Nanopore reads, two biological replicates, R9.4 flowcells · human · n = at least 2 million basecalled reads per replicate
  • PacBio Sequel II long reads from dissected cortex and hippocampus of 8-month male C57BL/6 mice, two replicates per region · mouse · n = minimum 6 million raw reads per replicate
  • Matched Illumina NextSeq500 short-read libraries for the mouse brain samples · mouse · n = minimum 50 million paired-75bp reads per sample
  • GM12878 Illumina short-read RNA-seq from the ENCODE portal, quantified with Kallisto · human · ENCSR000AEH
  • RNA-PET clusters for GM12878 from the ENCODE portal, lifted hg19 to hg38 · human · ENCFF001TIL
  • Robust human CAGE peaks from FANTOM5 used to check transcription start sites · human
  • SIRV and ERCC (Set 3) spike-ins with known splice patterns, used to test the transcript filter · n = 69 expected known isoforms from 7 SIRV genes
Methods & tools
TALON, TranscriptClean, Minimap2, FLAIR, Kallisto, EdgeR, STAR, Bedtools, ONT Guppy, Isoseq3 (CCS, Lima, Refine), GENCODE, UCSC LiftOver
Reagents & environment
TALON version 5.0, TranscriptClean v2.0.2, Minimap2 2.17, EdgeR v3.28.1, CCS v4.0.0, Lima v1.10.0, Isoseq3 v3.2.2 from SMRTanalysis 8.0, ONT Guppy 3.2.1+334123b; annotations GENCODE v29 (human, GRCh38) and vM21 (mouse, mm10). Key command-line parameters are given for each step.
Reproducibility gaps
Code is identified only by repository paths (ENCODE-DCC/long-read-rna-pipeline, mortazavilab/TALON) and 'available on GitHub' statements for TranscriptClean, TALON and the custom scripts; no URL survives in this text, and no accession is given for the new PacBio or ONT reads.
Limitations (6)
  • Very long transcripts are under-detected without size selection: only 3 reads fully matched known XIST isoforms in GM12878, and the longest of them (3,539 nt) still missed several kb at both ends. · stated by the authors
  • Genuine ISM isoforms cannot be told from truncation artifacts without independent CAGE or 3' evidence, so the authors advise ignoring ISMs for discovery until then, and expect the problem to worsen in lower-quality post-mortem RNA. · stated by the authors
  • As cDNA sequencing depth rises, PCR duplicates may become common and would be hard to spot without unique molecular identifiers, an issue direct-RNA sequencing avoids. · stated by the authors
  • Cost at the time of the study was roughly an order of magnitude above the short-read equivalent, which the authors call a limiting factor. · stated by the authors
  • The platform comparison is confounded by depth: the ONT arm was sequenced to a lower read count than PacBio, so some of the gap in detected genes and transcripts need not reflect the chemistry. · noted in this summary
  • Scope is two biological replicates of one cell line and two brain regions from 8-month male C57BL/6 mice, and the FLAIR comparison uses the authors' own data and parameter choices, with sensitivity but not false-positive rate measured. · noted in this summary

Read this if you run or plan long-read RNA-seq and need to know how PacBio and direct-RNA Nanopore differ in artifacts, which novel isoform calls to trust, and how a junction-matching pipeline compares with FLAIR.

Citations
162

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISTardáguila '18Loving '25Paniagua '26Marchand '25Hu '21Kuo '20
69% semantic matchBioinformatics
Manuel Tardáguila, Lorena de la Fuente, Cristina Martí+16 more·Genome Research·Mar 1, 2018· 458 citations