An evaluation of clustering and assembly strategies from Iso-Seq data in the absence of reference genomes in non-model animals

An evaluation of clustering and assembly strategies from Iso-Seq data in the absence of reference genomes in non-model animals | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results An evaluation of clustering and assembly strategies from Iso-Seq data in the absence of reference genomes in non-model animals View ORCID Profile Klara Eleftheriadi , Marçal Vázquez-Valls , View ORCID Profile Rosa Fernández doi: https://doi.org/10.1101/2025.09.18.677004 Klara Eleftheriadi 1 Metazoa Phylogenomics and Genome Evolution Lab, Institute of Evolutionary Biology (CSIC-UPF) , Barcelona ( Spain ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Klara Eleftheriadi For correspondence: klara.eleftheriadi{at}ibe.upf-csic.es rosa.fernandez{at}ibe.upf-csic.es Marçal Vázquez-Valls 1 Metazoa Phylogenomics and Genome Evolution Lab, Institute of Evolutionary Biology (CSIC-UPF) , Barcelona ( Spain ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site Rosa Fernández 1 Metazoa Phylogenomics and Genome Evolution Lab, Institute of Evolutionary Biology (CSIC-UPF) , Barcelona ( Spain ) Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Rosa Fernández For correspondence: klara.eleftheriadi{at}ibe.upf-csic.es rosa.fernandez{at}ibe.upf-csic.es Abstract Full Text Info/History Metrics Preview PDF Abstract Transcriptome assembly enables the recovery of expressed genes and isoforms, but the optimal strategy for reconstructing transcriptomes from long-read sequencing remains unresolved. In particular, establishing best practices for generating accurate gene models and selecting representative isoforms is essential for comparative genomics, as for orthology inference typically only the longest isoform per gene model is included. Here, we systematically compare clustering and de novo assembly methods using PacBio Iso-Seq data from 17 animal lineages spanning seven phyla, most of them non-model species, with the goal of investigating which methodology is more adequate to select one isoform per gene model, in the absence of specific pipelines to do so. We evaluate four approaches: isoseq cluster, CD-HIT, RNA-Bloom2 and isONform. We benchmark them with short-reads using Trinity, assessing assembly quality with BUSCO completeness, short-read mapping rates, coding sequence recovery, and longest isoform prediction. Our results show that CD-HIT clustering at high similarity thresholds (≥99%) yields the most complete and coding-rich long-read transcriptomes, rivaling Trinity while avoiding its high redundancy. Consensus-based methods such as isoseq cluster and isONform recover fewer single-copy orthologs (mirrored in a lower BUSCO score) and achieve lower mapping rates, while RNA-Bloom2 provide intermediate performance with reduced duplication. Together, these findings establish, to date, CD-HIT as a robust and practical strategy for transcriptome reconstruction from long-read data when genomic references are unavailable. By benchmarking de novo methods across a taxonomically broad dataset, this work defines the realistic capabilities of long-read transcriptome reconstruction in the absence of a reference genome and provides practical guidance for deriving high-quality gene models and selecting representative isoforms for orthology inference in non-model species. Introduction Transcriptome assembly is the computational reconstruction of all RNA transcripts present in a cell, tissue, or organism. These assemblies provide insight into gene expression, alternative splicing and regulatory networks that cannot be captured by genomic information alone. Thus, transcriptome assembly has become an essential tool in fields ranging from developmental biology and evolutionary genomics to agriculture and medicine, where it enables the discovery of novel genes, non-coding RNAs, and biomarkers relevant to health and disease ( Wang et al. 2009 ; Conesa et al. 2016 ). Broadly, there are two main categories for reconstructing a transcriptome assembly from raw reads: genome-guided and de novo. Genome-guided methods leverage an available reference genome to align the reads and assemble the transcripts, which typically improves accuracy, but their utility is limited in species with incomplete or poorly annotated references ( Haas et al. 2013 ). In contrast, de novo assemblers reconstruct transcripts directly from the sequencing data, making it indispensable for non-model organisms and for identifying transcriptomic features absent from reference genomes. However, de novo approaches are computationally more demanding and may be prone to artifacts such as fragmented or chimeric transcripts ( Smith-Unna et al. 2016 ). Both approaches thus involve trade-offs that must be carefully considered depending on the organism and resources available. Short-read sequencing platforms such as Illumina have dominated the field for more than a decade because of their affordability and high throughput, and they remain widely used in transcriptomic research. Nevertheless, the short length of reads prevents the direct recovery of full-length transcripts and isoforms, limiting the completeness and accuracy of transcriptomes ( Byrne et al. 2019 ). Reliance on fragmentary data introduces particular challenges in reconstructing complex transcriptomes with extensive alternative splicing, often resulting in incomplete, fragmented, or chimeric assemblies ( Smith-Unna et al. 2016 ). The advent of third-generation sequencing technologies, particularly Pacific Biosciences’ (PacBio) single-molecule real-time (SMRT) Iso-Seq protocol, has marked a turning point in transcriptomics. PacBio high-fidelity (HiFi) sequencing now produces long, accurate reads (>99.9% median accuracy), frequently spanning tens of kilobases and preserving exon order and orientation ( Gonzalez-Garay 2016 ; Wijeratne et al. 2024 ; Wang et al. 2025 ). This improves isoform detection and reduces the complexity of transcriptome reconstruction. In eukaryotes, a single gene typically contains multiple exons separated by introns, and through processes such as alternative splicing, it can give rise to multiple transcript isoforms that together define the gene model. When a high-quality reference genome is available, gene models are inferred with confidence by mapping long-read transcripts to genomic coordinates using established pipelines (e.g. StringTie, Mandalorion) ( Kovaka et al. 2019 ; Volden et al. 2023 ). In the absence of a genome, gene-level inference must be approximated directly from the transcript sequences. In short-read RNA-seq, where the read-length does not span full transcripts, graph-based assembly approaches (e.g. Trinity) ( Grabherr et al. 2011 ) are typically used to first reconstruct isoforms before aggregate them into putative orthogroups or gene families, although these methods suffer from fragmentation and redundancy, as mentioned before. By contrast, long-read sequencing technologies capture full-length transcripts directly, eliminating the need for assembly ( Sahlin and Medvedev 2020 ). Nonetheless, grouping these transcript isoforms into clusters to construct gene models remains a major challenge ( Sahlin and Medvedev 2020 ). For a variety of downstream analyses in comparative studies, including orthology inference ( Li et al. 2003 ; Emms and Kelly 2019 ), phylogenomic analyses ( Fernández et al. 2014 ; Fernández and Gabaldón 2020 ; Balart-García et al. 2021 ), gene-level expression quantification ( Davidson and Oshlack 2014 ), it is necessary to collapse transcript isoforms into putative gene models, typically by retaining a single representative isoform per gene, most often the longest coding sequence. However, the optimal strategy for reconstructing de novo transcriptome assemblies with gene models from Iso-Seq data remains unclear. A key question persists: Is clustering alone sufficient to recover a complete transcriptome, or does a de novo assembly algorithmic step still offer advantages, even with long-read data? While many studies default to the PacBio Iso-Seq clustering pipeline, few have rigorously assessed its completeness or quality, especially in non-model species where benchmarking remains sparse ( Pootakham et al. 2020 ; Ali et al. 2021 ). In this study, we address this gap by systematically comparing clustering algorithms and de novo assembly methods for long-read transcriptome reconstruction using experimental data from 17 animal lineages. We evaluate the performance of five software packages, categorized as (i) clustering tools, including PacBio’s isoseq cluster module ( https://github.com/PacificBiosciences/pbbioconda ) and CD-HIT ( Fu et al. 2012 ), and (ii) graph-based de novo assemblers, including RNA-Bloom2 ( Nip et al. 2020 ) and isONform ( Petri and Sahlin 2023 ). Additionally, we use Trinity ( Grabherr et al. 2011 ) as a benchmark for completeness and comparison with long-read approaches. To assess transcriptome quality, we employ multiple metrics: BUSCO scores ( Simão et al. 2015 ), short-read alignment rates, open reading frame (ORF) recovery, and the total number of longest protein isoforms per assembly. Through this comprehensive comparison, we aim to identify the most robust and practical strategy, to date, for constructing high-quality transcriptomes from long-read Iso-Seq data, particularly in non-model organisms, and provide practical recommendations for the biodiversity genomics community. Methods Sampling and sequencing The dataset analyzed in this study comprises 17 species across 7 animal phyla ( Figure 1 ). The species include both model organisms such as Caenorhabditis elegans and Schmidtea mediterranea , and numerous non-model species from diverse ecological and evolutionary lineages. This broad taxonomic sampling provides a robust framework for evaluating transcriptome assembly methods across deep evolutionary divergence. For each species, both long-read Iso-Seq data and short-read RNA-Seq data were generated from whole specimens. Download figure Open in new tab Figure 1. Phylogenetic distribution of sampled species. Phylogenetic tree showing the 17 animal species used in this study, spanning seven metazoan phyla: Arthropoda, Annelida, Onychophora, Mollusca, Nemertea, Platyhelminthes, and Nematoda. Species names are listed alongside their respective phyla. For short-read RNA-Seq, between 35 and 60 individuals per species were exposed to a range of experimental treatments. Including all these treatments, diverse transcriptional responses are induced and transcript diversity for downstream analyses is maximized. Specimens were flash frozen immediately after each experiment. In the case of onychophorans, samples were kept in RNAlater or TRIzol at -70ºC until processed. For small animals (for instance nematodes, nemerteans and molluscs), RNA was extracted from the whole specimen. For medium-sized specimens (as annelids or onychophorans), they were dissected into anterior and posterior parts (see Table S3-S19 for detailed description on replicates). RNA extractions after stress experiments were performed using the TRIzol® reagent (Invitrogen, USA) method following the manufacturer’s instructions and using MaXtract® High Density tubes (Qiagen) to minimize DNA contamination prior to mechanical sample homogenization either by plastic rotor pestles or by ceramic mortar, depending on the sample. The concentration of all samples was assessed by Qubit RNA BR Assay kit (Thermo Fisher Scientific). Libraries were prepared with the TruSeq Stranded mRNA library preparation kit (Illumina), and sequenced on a NovaSeq 6000 (Illumina, 2 × 150 bp) for a minimum of 6Gb coverage. The same RNA extractions used for short-read sequencing were pooled together in each species to prepare an Iso-Seq library. RNA samples were subjected to DNAse treatment using the Turbo DNA-free DNase (Invitrogen) following the manufacturer’s instructions. SMRTbell libraries were generated following the procedure ‘Preparing Iso-Seq® libraries using SMRTbell® prep kit 3.0 (PN 102-396-000 REV02 APR2022)’. To enable pooling of multiple samples on 1 SMRTcell, libraries were made using barcoded adapters. The libraries were sequenced on a Sequel-IIe using Sequel II sequencing kit 2.0 and Binding kit 3.1 with 24 hr movie-time. The majority of the species included in this dataset are non-model organisms that are undersampled and poorly characterized both at the genomic and transcriptomic level. This taxonomically diverse and biologically relevant dataset provides a robust framework for evaluating transcriptome reconstruction strategies across different assembly approaches and sequencing technologies. Long and short reads preprocessing Prior to the construction of de novo transcriptome assemblies, we obtained the Full-Length Non Concatemer (FLNC) reads separately from Iso-Seq data of the 17 species using the preprocessing pipeline available from PacBio. The data was demultiplexed with the Iso-Seq pipeline v4.0.0 ( https://github.com/PacificBiosciences/pbbioconda ), and cDNA primers were removed using LIMA v2.7.1. PolyA tails and artificial concatemers were removed using isoseq refine v4.0.0. Raw Illumina RNA-Seq reads were quality controlled with FastQC v0.11.9 ( https://www.bioinformatics.babraham.ac.uk/projects/fastqc/ ) and adapters and low quality base pairs were removed using Trimmomatic v0.39 (MINLEN: 75, SLIDINGWINDOW:4:15, LEADING: 10, TRAILING: 10, AVGQUAL: 30) ( Bolger et al. 2014 ). Trimmed RNA-seq reads were quality controlled with FastQC before further analysis. Construction of de novo reference transcriptome assemblies using clustering algorithms We employed two main methods to construct de novo reference transcriptomes using clustering algorithms. The first method was the isoseq cluster module within the PacBio SMRT suite ( https://github.com/PacificBiosciences/pbbioconda ). Using isoseq cluster v4.0.0, we followed the publicly available PacBio protocol for transcriptome construction. In this process, two or more Full-Length Non-Concatemer (FLNC) reads are clustered if they differ by less than 100 bp at the 5′ end and less than 30 bp at the 3′ end, with the additional requirement that there are no internal gaps. The resulting consensus sequences define the transcriptome as High Quality Isoforms, while Low Quality Isoforms are discarded from further analysis. For the second method, we used CD-HIT v4.8.1 ( Li and Godzik 2006 ), a fast and flexible clustering tool originally developed for clustering large protein sequence databases at various sequence identity thresholds. CD-HIT significantly reduced execution times compared to earlier programs, though it can introduce some redundancy. We used CD-HIT with sequence similarity thresholds of 95%, 96%, 97%, 98%, and 99% to cluster preprocessed FLNC reads. For each threshold, the longest sequence in each cluster was retained as the cluster representative, forming five distinct reference transcriptomes composed of the most “informative” sequences. While CD-HIT was first designed for protein clustering, its underlying greedy incremental algorithm ( Holm and Sander 1998 ) is also well-suited for DNA and RNA sequence clustering, as implemented in the CD-HIT-EST tool used in our analysis ( Li and Godzik 2006 ). Construction of de novo reference transcriptomes using graph-based assembly methods De novo transcriptome assembly can be achieved through several algorithmic approaches, with graph-based algorithms among the most widely used for reconstructing transcripts from sequencing reads. For this purpose, we chose two widely used software for long-read Iso-Seq data, RNA-Bloom2 ( Nip et al. 2020 ) and isONform ( Petri and Sahlin 2023 ). Then we also used Trinity ( Grabherr et al. 2011 ) to reconstruct assemblies from short Illumina reads for evaluation purposes, as described below. RNA-Bloom2 is a reference-free assembler tailored for long-read transcriptome sequencing data. We used RNA-Bloom2 v2.0.1 to assemble transcriptomes directly from full-length non-chimeric (FLNC) reads, without requiring a reference genome. The process takes the FLNC reads as input and produces a final transcriptome assembly as output. isONform follows a multi-step workflow and operates in conjunction with isONclust. For our assemblies with isONform v0.3.3, we first used isONclust to cluster FLNC reads efficiently, employing a greedy, minimizer-based approach that switches to full alignment as needed ( Sahlin and Medvedev 2020 ). isONform then constructs isoforms from these clusters, generating a set of predicted transcripts in FASTQ format for downstream analysis. Quality evaluation of the reference de novo transcriptome assemblies Trinity is one of the most established reference-free assembly programs for short-read RNA-Seq data ( Tzec-Interián et al. 2025 ). Based on de Bruijn graph algorithms, Trinity enables the de novo assembly of full-length transcripts when a suitable reference genome is unavailable. In this study, we used Trinity v2.11.0 to assemble transcriptomes from short-read Illumina RNA-seq data. These assemblies serve as a key evaluation metric for comparing the performance of the long-read assembly methods. To compare the quality of transcriptome assemblies produced by different algorithms, we used several global evaluation metrics. Assembly completeness was assessed using BUSCO v 5.4.7 ( Simão et al. 2015 ), analyzing the presence of 954 universal single-copy orthologs from the metazoa odb10 database. To evaluate read support, we mapped preprocessed short RNA-seq reads to each reference transcriptome using Minimap2 ( Li 2018 ) and calculated the proportion of mapped reads. Coding sequence recovery was measured with TransDecoder v5.5.0 ( https://github.com/TransDecoder/TransDecoder ), which identifies candidate coding regions and predicts likely protein sequences in each assembly. Finally, we quantified the number of longest isoforms present in each reconstructed transcriptome, in order to compare the resulting longest protein isoforms per assembly. Results & Discussion In this study, we compared clustering and assembly strategies for de novo transcriptome reconstruction from long-read Iso-Seq data in 17 mainly non-model animal species. We evaluated four approaches: (i) isoseq cluster, (ii) CD-HIT clustering at five similarity thresholds (95–99%), (iii) RNA-Bloom2, and (iv) isONform, all of them benchmarked with Trinity (the latter using short Illumina RNA-seq reads). In total, we constructed 170 transcriptomes and assessed quality by comparing BUSCO completeness using metazoa odb10, the percentage of mapped short reads using Minimap2, and counting the number of total protein isoforms as long as the longest ones per assembly ( Figure 2 , Table S1). Download figure Open in new tab Figure 2. Workflow for evaluating transcriptome assembly strategies. Workflow showing preprocessing, clustering, and assembly of Illumina and PacBio Iso-Seq reads, followed by quality assessment using BUSCO, Minimap2, and TransDecoder to evaluate completeness, read support, and coding potential. Assembly completeness Our results show that CD-HIT at 99% similarity consistently outperforms other long-read methods in transcriptome completeness, with BUSCO scores rivaled only by Trinity ( Figure 3a , 4a ; Supplementary Figure S1). The higher completeness obtained with CD-HIT likely reflects its use of the longest representative sequence in each cluster, whereas consensus-based approaches such as isoseq cluster and isONform tend to collapse transcript diversity, potentially discarding biologically valid isoforms or introducing artificial consensus sequences not corresponding to conserved genes. Download figure Open in new tab Figure 3. Comparative performance of transcriptome assembly methods across 17 animal species. (a) BUSCO completeness scores (metazoa_odb10 database) summarizing the proportion of complete, duplicated, fragmented, and missing single-copy orthologs across all assemblies. (b) Percentage of Illumina short-reads mapping back to each assembly, as a measure of representation of high accurate short-reads in each assembly. (c) Log-scaled number of transcripts, predicted proteins, and longest protein isoforms identified per assembly. Trinity also achieves high completeness but displays a high rate of duplicated BUSCOs, likely due to transcript redundancy and variability in gene expression levels. These issues are consistent with previously reported limitations of short-read assemblers, including reduced reproducibility and an elevated rate of false positives ( Bankar et al. 2015 ; Wang and Gribskov 2017 ). In contrast, RNA-Bloom2 produces lower duplication rates ( Figure 3a ; Figure S1), though at the expense of overall completeness. As expected, unassembled FLNC reads show very high BUSCO duplication, reflecting the presence of multiple overlapping isoforms prior to clustering or assembly. Finally, we note that CD-HIT assemblies also display increased duplication at higher similarity thresholds, illustrating the trade-off between maximizing completeness and minimizing redundancy. Read mapping support Read mapping analyses show that all CD-HIT and Trinity assemblies capture the highest proportion of short-reads ( Figure 3b , 4b ; Figure S2). Trinity’s strong performance is expected, since its assemblies are generated directly from the short-read data that are also used for the mapping. Among the long-read strategies, RNA-Bloom2 consistently achieves higher mapping rates than Iso-Seq cluster and isONform, indicating that it better represents the transcript diversity detected by Illumina sequencing. In contrast, isONform performs worst across most species, often recovering substantially fewer reads, which suggests that its consensus-based reconstruction discards valid transcript variants present in the data. Download figure Open in new tab Figure 4. Assemblers performance across species. (a) Radar plot showing BUSCO completeness (Complete + Fragmented) (b) Radar plot showing percentage of short-reads mapped per assembly. Coding potential Coding potential and isoform recovery ( Figure 3c ; Supplementary Figure S3) highlight clear differences among assemblers. Trinity assemblies are highly redundant, producing far more transcripts than unique proteins or isoforms, consistent with earlier reports of transcript fragmentation and over-representation in short-read assemblies ( Cerveau and Jackson 2016 ). In contrast, CD-HIT at 99% similarity generates the largest number of transcripts and efficiently preserves coding information, as most transcripts are retained after translation into proteins. RNA-Bloom2 produces fewer transcripts and proteins overall, and in many species transcript counts exceed protein counts, suggesting redundancy and the recovery of non-coding sequences. Isoseq cluster and isONform yield more balanced results, with similar counts of transcripts, proteins, and isoforms, reflecting their ability to retrieve full-length isoforms directly from the long-read data. Published benchmarks of RNA-Bloom2 ( Nip et al. 2020 ) and isONform ( Petri and Sahlin 2023 ), including comparisons against the Nanopore-based assembler RATTLE ( de la Rubia et al. 2022 ), indicate that consensus-based approaches such as the last step of isONform and isoseq cluster may under-represent transcript diversity and fail to capture biologically relevant isoforms, observations consistent with our results. Together, these results reinforce that direct clustering with CD-HIT, particularly at high similarity thresholds, recovers a richer and more accurate representation of coding potential than either consensus-based or graph-based assembly strategies. Caveats and next steps The core aim of this project is to evaluate how effectively transcriptomes can be reconstructed from long-read data de novo, without relying on genomic resources. This is a common scenario in non-model organism research, where high-quality genomes are often unavailable or cost-prohibitive to generate, and it actually represents a bottleneck in long-read-based comparative genomics of non-model organisms. By comparing de novo assembly and clustering tools across a diverse set of animal lineages, this work aims to define what could be achieved under these constraints and to identify which approaches yield the most complete and biologically informative transcriptomes when working de novo. In any case, reference-guided transcriptome reconstruction remains the gold standard for model species with well annotated genomes. As several species in this dataset are now undergoing genome sequencing or annotation, the next critical step will be to directly compare the de novo transcriptomes generated here with reference-guided assemblies from the same species. This will allow us to quantify the performance of these tools in the absence of a genome and to refine our understanding of how well de novo strategies approximate biological reality. Ultimately, this framework, first benchmarking de novo methods in the absence of a genome, then validating them against genome-guided reconstructions, will provide the scientific community with a practical path forward when genomic resources are lacking. Conclusions This study provides the first large-scale comparison of de novo long-read transcriptome assembly strategies across 17 animal lineages, mainly non-model species, in the absence of a reference genome. By benchmarking clustering- and assembly-based tools against short-read assemblies, we show that direct clustering with CD-HIT at high similarity thresholds offers the most complete and coding-rich transcriptomes from Iso-Seq data, consistently outperforming consensus-based methods and approaching the completeness of Trinity. However, unlike Trinity, CD-HIT avoids excessive redundancy and better reflects the full-length isoforms captured by long reads. Our results highlight the practical value of clustering approaches for laboratories working without genomic resources, where reference-guided strategies are not feasible. In this context, CD-HIT provides a realistic and efficient solution for reconstructing transcriptomes from long-read sequencing. At the same time, the comparison to short-read Trinity assemblies underscores the trade-offs between completeness and redundancy that remain central to transcriptome reconstruction. Future steps including integration of genome-guided methods as high-quality chromosome-level genomes become available, will be essential to validate how close to biological reality the de novo reconstructions are. All together, this approach will establish best practices for de novo assembly from long-read Iso-Seq data to the community. Supplementary Data Supplementary material can be found: Supplementary Material.pdf Acknowledgements R.F acknowledges support from the European Research Council (this project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement no. 948281). We thank Centro de Supercomputación de Galicia (CESGA) for access to computer resources. ChatGPT was used to polish grammar and improve readability of the manuscript. No AI tools were used for idea generation, data analysis or interpretation of results. Funder Information Declared European Research Council, https://ror.org/0472cxd90 , 948281 References ↵ Ali A , Thorgaard GH , Salem M. 2021 . PacBio Iso-Seq Improves the Rainbow Trout Genome Annotation and Identifies Alternative Splicing Associated With Economically Important Phenotypes . Front Genet 12 : 683408 . Available from : doi: 10.3389/fgene.2021.683408 OpenUrl CrossRef PubMed ↵ Balart-García P , Cieslak A , Escuer P , Rozas J , Ribera I , Fernández R. 2021 . Smelling in the dark: Phylogenomic insights into the chemosensory system of a subterranean beetle . Mol Ecol 30 : 2573 – 2590 . Available from : doi: 10.1111/mec.15921 OpenUrl CrossRef ↵ Bankar KG , Todur VN , Shukla RN , Vasudevan M. 2015 . Ameliorated de novo transcriptome assembly using Illumina paired end sequence data with Trinity Assembler . Genom Data 5 : 352 – 359 . Available from : doi: 10.1016/j.gdata.2015.07.012 OpenUrl CrossRef ↵ Bolger AM , Lohse M , Usadel B. 2014 . Trimmomatic: a flexible trimmer for Illumina sequence data . Bioinformatics 30 : 2114 – 2120 . Available from : doi: 10.1093/bioinformatics/btu170 OpenUrl CrossRef PubMed Web of Science ↵ Byrne A , Cole C , Volden R , Vollmers C. 2019 . Realizing the potential of full-length transcriptome sequencing . Philos. Trans. R. Soc. Lond. B Biol. Sci . 374 : 20190097 . Available from: https://royalsocietypublishing.org/doi/10.1098/rstb.2019.0097 OpenUrl CrossRef PubMed ↵ Cerveau N , Jackson DJ . 2016 . Combining independent de novo assemblies optimizes the coding transcriptome for nonconventional model eukaryotic organisms . BMC Bioinformatics 17 : 525 . Available from : doi: 10.1186/s12859-016-1406-x OpenUrl CrossRef PubMed ↵ Conesa A , Madrigal P , Tarazona S , Gomez-Cabrero D , Cervera A , McPherson A , Szcze-niak MW , Gaffney DJ , Elo LL , Zhang X , et al. 2016 . A survey of best practices for RNA-seq data analysis . Genome Biol 17 : 13 . Available from : doi: 10.1186/s13059-016-0881-8 OpenUrl CrossRef PubMed ↵ Davidson NM , Oshlack A. 2014 . Corset: enabling differential gene expression analysis for de novo assembled transcriptomes . Genome Biol 15 : 410 . Available from : doi: 10.1186/s13059-014-0410-6 OpenUrl CrossRef PubMed ↵ Emms DM , Kelly S. 2019 . OrthoFinder: phylogenetic orthology inference for comparative genomics . Genome Biol 20 : 238 . Available from : doi: 10.1186/s13059-019-1832-y OpenUrl CrossRef PubMed ↵ Fernández R , Gabaldón T. 2020 . Gene gain and loss across the metazoan tree of life . Nat Ecol Evol 4 : 524 – 533 . Available from : doi: 10.1038/s41559-019-1069-x OpenUrl CrossRef PubMed ↵ Fernández R , Laumer CE , Vahtera V , Libro S , Kaluziak S , Sharma PP , Pérez-Porro AR , Edgecombe GD , Giribet G. 2014 . Evaluating topological conflict in centipede phylogeny using transcriptomic data sets . Mol Biol Evol 31 : 1500 – 1513 . Available from : doi: 10.1093/molbev/msu108 OpenUrl CrossRef PubMed Web of Science ↵ Fu L , Niu B , Zhu Z , Wu S , Li W. 2012 . CD-HIT: accelerated for clustering the next-generation sequencing data . Bioinformatics 28 : 3150 – 3152 . Available from : doi: 10.1093/bioinformatics/bts565 OpenUrl CrossRef PubMed Web of Science ↵ Gonzalez-Garay ML . 2016 . Introduction to isoform sequencing using pacific Biosciences technology (Iso-seq ). In: Translational Bioinformatics . Dordrecht : Springer Netherlands . p. 141 – 160 . Available from: http://link.springer.com/10.1007/978-94-017-7450-5_6 ↵ Grabherr MG , Haas BJ , Yassour M , Levin JZ , Thompson DA , Amit I , Adiconis X , Fan L , Raychowdhury R , Zeng Q , et al. 2011 . Full-length transcriptome assembly from RNA-Seq data without a reference genome . Nat. Biotechnol . 29 : 644 – 652 . Available from: https://www.nature.com/articles/nbt.1883 OpenUrl CrossRef PubMed ↵ Haas BJ , Papanicolaou A , Yassour M , Grabherr M , Blood PD , Bowden J , Couger MB , Eccles D , Li B , Lieber M , et al. 2013 . De novo transcript sequence reconstruction from RNA-seq using the Trinity platform for reference generation and analysis . Nat Protoc 8 : 1494 – 1512 . Available from : doi: 10.1038/nprot.2013.084 OpenUrl CrossRef PubMed ↵ Holm L , Sander C. 1998 . Removing near-neighbour redundancy from large protein sequence collections . Bioinformatics 14 : 423 – 429 . Available from: https://academic.oup.com/bioinformatics/article/14/5/423/225479 OpenUrl CrossRef PubMed Web of Science ↵ Kovaka S , Zimin AV , Pertea GM , Razaghi R , Salzberg SL , Pertea M. 2019 . Transcriptome assembly from long-read RNA-seq alignments with StringTie2 . Genome Biol 20 : 278 . Available from : doi: 10.1186/s13059-019-1910-1 OpenUrl CrossRef PubMed ↵ Li H. 2018 . Minimap2: pairwise alignment for nucleotide sequences . Bioinformatics 34 : 3094 – 3100 . Available from: https://academic.oup.com/bioinformatics/article/34/18/3094/4994778 OpenUrl CrossRef PubMed ↵ Li L , Stoeckert CJ Jr . , Roos DS . 2003 . OrthoMCL: identification of ortholog groups for eukaryotic genomes . Genome Res 13 : 2178 – 2189 . Available from : doi: 10.1101/gr.1224503 OpenUrl Abstract / FREE Full Text ↵ Li W , Godzik A. 2006 . Cd-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences . Bioinformatics 22 : 1658 – 1659 . Available from: https://academic.oup.com/bioinformatics/article/22/13/1658/194225 OpenUrl CrossRef PubMed Web of Science ↵ Nip KM , Chiu R , Yang C , Chu J , Mohamadi H , Warren RL , Birol I. 2020 . RNA-Bloom enables reference-free and reference-guided sequence assembly for single-cell transcriptomes . Genome Res 30 : 1191 – 1200 . Available from : doi: 10.1101/gr.260174.119 OpenUrl Abstract / FREE Full Text ↵ Petri AJ , Sahlin K. 2023 . isONform: reference-free transcriptome reconstruction from Oxford Nanopore data . Bioinformatics 39 : i222 – i231 . Available from : doi: 10.1093/bioinformatics/btad264 OpenUrl CrossRef PubMed ↵ Pootakham W , Uengwetwanit T , Sonthirod C , Sittikankaew K , Karoonuthaisiri N. 2020 . A novel full-length transcriptome resource for black tiger shrimp (Penaeus monodon) developed using isoform sequencing (Iso-seq) . Front. Mar. Sci . 7 . Available from: https://www.frontiersin.org/article/10.3389/fmars.2020.00172/full ↵ de la Rubia I , Srivastava A , Xue W , Indi JA , Carbonell-Sala S , Lagarde J , Albà MM , Eyras E. 2022 . RATTLE: reference-free reconstruction and quantification of transcriptomes from Nanopore sequencing . Genome Biol 23 : 153 . Available from : doi: 10.1186/s13059-022-02715-w OpenUrl CrossRef PubMed ↵ Sahlin K , Medvedev P. 2020 . De Novo Clustering of Long-Read Transcriptome Data Using a Greedy, Quality Value-Based Algorithm . J Comput Biol 27 : 472 – 484 . Available from : doi: 10.1089/cmb.2019.0299 OpenUrl CrossRef PubMed ↵ Simão FA , Waterhouse RM , Ioannidis P , Kriventseva EV , Zdobnov EM . 2015 . BUSCO: assessing genome assembly and annotation completeness with single-copy orthologs . Bioinformatics 31 : 3210 – 3212 . Available from: https://academic.oup.com/bioinformatics/article/31/19/3210/211866 OpenUrl CrossRef PubMed ↵ Smith-Unna R , Boursnell C , Patro R , Hibberd JM , Kelly S. 2016 . TransRate: reference-free quality assessment of de novo transcriptome assemblies . Genome Res 26 : 1134 – 1144 . Available from : doi: 10.1101/gr.196469.115 OpenUrl Abstract / FREE Full Text ↵ Tzec-Interián JA , González-Padilla D , Góngora-Castillo EB. 2025 . Bioinformatics perspectives on transcriptomics: A comprehensive review of bulk and single-cell RNA sequencing analyses . Quant. Biol . 13 . Available from: https://onlinelibrary.wiley.com/doi/10.1002/qub2.78 ↵ Volden R , Schimke KD , Byrne A , Dubocanin D , Adams M , Vollmers C. 2023 . Identifying and quantifying isoforms from accurate full-length transcriptome sequencing reads with Mandalorion . Genome Biol 24 : 167 . Available from : doi: 10.1186/s13059-023-02999-6 OpenUrl CrossRef PubMed ↵ Wang B , Jia P , Gao S , Zhao H , Zheng G , Xu L , Ye K. 2025 . Long and Accurate: How HiFi Sequencing is Transforming Genomics . Genomics Proteomics Bioinformatics 23 . Available from : doi: 10.1093/gpbjnl/qzaf003 OpenUrl CrossRef ↵ Wang S , Gribskov M. 2017 . Comprehensive evaluation of de novo transcriptome assembly programs and their effects on differential gene expression analysis . Bioinformatics 33 : 327 – 333 . Available from : doi: 10.1093/bioinformatics/btw625 OpenUrl CrossRef PubMed ↵ Wang Z , Gerstein M , Snyder M. 2009 . RNA-Seq: a revolutionary tool for transcriptomics . Nat Rev Genet 10 : 57 – 63 . Available from : doi: 10.1038/nrg2484 OpenUrl CrossRef PubMed Web of Science ↵ Wijeratne S , Gonzalez MEH , Roach K , Miller KE , Schieffer KM , Fitch JR , Leonard J , White P , Kelly BJ , Cottrell CE , et al. 2024 . Full-length isoform concatenation sequencing to resolve cancer transcriptome complexity . BMC Genomics 25 : 122 . Available from : doi: 10.1186/s12864-024-10021-x OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted September 20, 2025. Download PDF Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following An evaluation of clustering and assembly strategies from Iso-Seq data in the absence of reference genomes in non-model animals Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share An evaluation of clustering and assembly strategies from Iso-Seq data in the absence of reference genomes in non-model animals Klara Eleftheriadi , Marçal Vázquez-Valls , Rosa Fernández bioRxiv 2025.09.18.677004; doi: https://doi.org/10.1101/2025.09.18.677004 Share This Article: Copy Citation Tools An evaluation of clustering and assembly strategies from Iso-Seq data in the absence of reference genomes in non-model animals Klara Eleftheriadi , Marçal Vázquez-Valls , Rosa Fernández bioRxiv 2025.09.18.677004; doi: https://doi.org/10.1101/2025.09.18.677004 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Evolutionary Biology Subject Areas All Articles Animal Behavior and Cognition (7635) Biochemistry (17697) Bioengineering (13895) Bioinformatics (41951) Biophysics (21456) Cancer Biology (18594) Cell Biology (25520) Clinical Trials (138) Developmental Biology (13381) Ecology (19903) Epidemiology (2067) Evolutionary Biology (24323) Genetics (15612) Genomics (22510) Immunology (17738) Microbiology (40401) Molecular Biology (17184) Neuroscience (88622) Paleontology (667) Pathology (2833) Pharmacology and Toxicology (4825) Physiology (7644) Plant Biology (15158) Scientific Communication and Education (2046) Synthetic Biology (4296) Systems Biology (9825) Zoology (2271)