Computational biologists have developed an end-to-end bioinformatics pipeline named TE-Seq that integrates long-read genomic sequencing with short-read transcriptomic profiling to resolve the expression of repetitive transposable elements across eukaryotic genomes. Described in Mobile DNA by researchers Maxfield M. G. Kelsey, Radha L. Kalekar, and John M. Sedivy, the software addresses long-standing technical hurdles in quantifying sequences from the so-called dark genome. Transposable elements comprise approximately half of the human genome and similar proportions of other mammalian and eukaryotic genomes, influencing vital physiological processes including chromosome segregation, gene regulation, and telomere maintenance, while also contributing to chronic inflammation, tumorigenesis, and neurodegenerative disorders. Despite their widespread biological importance, repetitive elements are routinely neglected in standard sequencing workflows because identical and highly similar sequences generate ambiguous alignments, known as multi-mapping reads, that prevent standard mapping algorithms from confidently assigning transcripts to individual genomic loci.
The central innovation of the TE-Seq framework lies in its ability to construct sample-specific custom reference genomes that incorporate polymorphic transposable element insertions and mask absent reference sequences before quantifying RNA expression. When conventional pipelines align RNA reads against a standard reference genome, transcripts originating from these unannotated, non-reference loci are inevitably misattributed to homologous reference elements or discarded entirely. By ingesting long-read Oxford Nanopore DNA sequencing data, TE-Seq discovers these sample-specific non-reference insertions and detects structural deletions, establishing an expanded reference assembly that accurately reflects the physical genome of the biological specimen under investigation.
Architecturally, TE-Seq operates through two overarching analytical modules designated as Annotate Reference and Short-read RNA-Seq. The Annotate Reference module accepts user-provided reference FASTA files and gene models, generating comprehensive repetitive element annotations across each chromosome using RepeatMasker when precomputed outputs are unavailable. Recognizing that RepeatMasker splits fragmented repeats into disjointed records, the pipeline executes dedicated scripts to consolidate fragmented features into unified records adhering to tidy data principles. Repeats are hierarchically grouped into superfamilies, families, and subfamilies, and classified as full-length or truncated depending on whether their alignment spans at least ninety-five percent of the consensus sequence. Subfamilies are further stratified as evolutionary young or old based on user-tunable average divergence thresholds, defaulting to divergence below three percent for Long Interspersed Nuclear Elements, eight percent for Short Interspersed Nuclear Elements, twelve percent for Endogenous Retroviruses, and fifteen percent for all other classes.
For young and potentially retrotransposition-competent subfamilies, the annotation module evaluates coding potential and structural integrity. In human sequences, this deep structural curation focuses on L1HS and L1PA2, while in murine models it assesses L1Md, L1Tf, L1Gf, and L1A subfamilies, selecting the seven least diverged subfamilies in other eukaryotic organisms. To determine open reading frame integrity, length distributions of identified coding sequences are tabulated, and lengths achieving a standard score greater than four are selected to reconstruct consensus amino acid sequences. Individual elements are annotated as open reading frame intact if their predicted coding sequences fall within ten percent of the consensus length and maintain less than five percent amino acid divergence. For endogenous retroviruses, the module distinguishes between solitary long terminal repeats generated by homologous recombination and complete proviruses possessing internal sequences, grouping adjacent elements into unique proviral cluster identifiers to track coordinated transcript synthesis.
When supplementary long-read genomic data are supplied, the Annotate Reference module processes Oxford Nanopore sequencing reads using Dorado for basecalling, Minimap2 for alignment, and PycoQC alongside Samtools stats for quality metrics. Non-reference insertions are detected using the transposable element insertion caller TLDR, which identifies clusters of clipped reads whose unmapped segments align to repeat consensus profiles. The pipeline enforces rigorous filtration criteria, requiring non-reference calls to pass all internal caller filters, maintain a median mapping quality score of sixty among supporting reads, possess target-site duplications diagnostic of authentic retrotransposition, and exhibit support from at least five reads, including two fully spanning reads representing at least twenty-five percent of local read depth. High-confidence insertions, along with thirty base pairs of flanking reference sequence, are appended to the genome as independent contigs, while homozygous structural deletions detected by Sniffles2 with at least ten supporting reads are masked using Bedtools.
Following genome augmentation, the Short-read RNA-Seq module performs adapter and quality trimming with fastp, aligning short paired-end reads to the customized genome using STAR with an allowance of up to one thousand multiple mapping locations per individual read. Gene-level counts are compiled using featureCounts against RefSeq gene annotations, while transposable element quantification is delegated to the expectation-maximization tool Telescope. Telescope models the probabilistic assignment of multi-mapping reads to specific repeat loci by iteratively weighing mappings against loci that accumulate uniquely mapped reads, operating under the established premise that loci with unique mappings reflect genuine biological transcription. TE-Seq outputs two discrete count matrices: a conservative quantification restricted entirely to uniquely aligned reads, and a relaxed, expectation-maximization reassigned matrix that captures both unique and multi-mapping read densities across individual repetitive element loci.
To avoid distortions during statistical normalization, TE-Seq adjusts standard DESeq2 workflows to accommodate the specific distribution of repetitive element transcriptomes. Because repeat annotations contain thousands of transcriptionally inactive or lowly expressed sequences that could skew global size factor estimations, the pipeline forces the DESeq2 median-of-ratios algorithm to compute size factors using RefSeq cellular genes exclusively through the controlGenes parameter. Differential expression across individual loci and genes is evaluated through negative binomial generalized linear models that account for biological covariates and user-defined batch variables. Beyond single-locus modeling, the pipeline performs aggregate count modeling at repeat family and subfamily levels, fitting negative binomial generalized linear models to summed raw counts with library size factors introduced as an offset, alongside gene set enrichment analysis where repeat subfamilies are ranked according to their DESeq2 Wald statistics.
Cells were cultured until entering replicative exhaustion and subsequently maintained in culture for either four weeks to establish early senescence or sixteen weeks to establish late senescence, using actively proliferating, early-passage cells as baseline controls. High-depth Oxford Nanopore DNA sequencing achieved greater than sixty-fold genomic coverage with read length N50 values exceeding twenty-four kilobases. Using the telomere-to-telomere T2T-HS1 assembly as the starting reference, TLDR and Sniffles2 cataloged two hundred twenty-six non-reference Alu insertions, one hundred seventeen non-reference L1 insertions, twenty-four non-reference SVA insertions, three hundred forty-six AluY deletions, and sixty-seven L1HS deletions within the LF1 genome.
Structural evaluation revealed that non-reference retrotransposon insertions were remarkably intact compared to their annotated reference counterparts. Of the sixty-five full-length non-reference L1HS insertions detected in LF1 cells, forty-six preserved intact open reading frames, representing thirty-two percent of all intact L1HS elements identified across the entire personalized genome. Phylogenetic reconstruction placed these non-reference insertions across diverse lineages of the L1HS tree, with a compact group of active reference elements predicted to have seeded the majority of offspring insertions, including one locus on chromosome seventeen that spawned eight novel genomic insertions. Quadruplicate stranded, poly-A selected, one hundred fifty base pair paired-end RNA sequencing revealed extensive transcriptional remodeling during senescence, with 10,204 transposable elements upregulated in early senescence and 7,418 upregulated in late senescence, dominated by elevated transcript counts across LINE, SINE, and LTR retrotransposon superfamilies.
Crucially, the implementation of a custom reference genome reshaped differential expression outcomes and prevented substantial quantitative errors. Comparing differential expression calls between the custom and standard T2T-HS1 references demonstrated that only forty-five percent of differentially expressed L1HS elements in early senescence and thirty-eight percent in late senescence were shared between pipelines. Most divergent calls involved non-polymorphic reference elements whose read distributions shifted once competing polymorphic contigs were introduced into the alignment index, demonstrating that omitting personal non-reference variations distorts mapping accuracy across the wider repetitive transcriptome.
The authors also established analytical heuristics within TE-Seq to differentiate autonomous retrotransposon transcription from passive transcriptional read-through driven by neighboring host gene promoters. The pipeline categorizes repeat loci relative to annotated transcripts as exonic, intronic, gene-adjacent within ten kilobases, or intergenic, and generates genome tracks using Plotgardener to inspect contextual transcription. In senescent LF1 cells, overall intronic signal declined while intergenic signal rose modestly, and full-length L1HS expression changes showed no correlation with neighboring cellular genes, supporting authentic autonomous activation. Implemented in Snakemake with multi-architecture containerization and SLURM workload profiles, TE-Seq offers a robust, reproducible platform for resolving transposable element dynamics, demonstrating that non-reference retrotransposons represent a substantial, previously overlooked fraction of the functional transcriptome in aging human cells.
Subject of Research: Bioinformatics pipeline integrating long-read DNA and short-read RNA sequencing to quantify transposable element expression and non-reference insertions during senescence
Article Title: TE-Seq: a transposable element annotation and RNA-Seq pipeline
Article References: Kelsey, M. M. G., Kalekar, R. L., & Sedivy, J. M. (2025). TE-Seq: a transposable element annotation and RNA-Seq pipeline. Mobile DNA, 16(1), Article 44. https://doi.org/10.1186/s13100-025-00381-w
Image Credits: AI Generated
DOI: 10.1186/s13100-025-00381-w
Keywords: TE-Seq, transposable elements, retrotransposons, RNA-Seq, Nanopore sequencing, cellular senescence, L1HS, Telescope, RepeatMasker, Snakemake, DESeq2, non-reference insertions, genomic repeats, long-read sequencing, differential expression, human lung fibroblasts, bioinformatics pipeline
Cite Scienmag News
Beatrice Stafford. (November 21, 2026). Custom Genomic Reference Workflow Reveals Hidden Transposable Element Activity During Cellular Senescence. Scienmag. https://scienmag.com/custom-genomic-reference-workflow-reveals-hidden-transposable-element-activity-during-cellular-senescence/
Beatrice Stafford. "Custom Genomic Reference Workflow Reveals Hidden Transposable Element Activity During Cellular Senescence." Scienmag, 21 November 2026, https://scienmag.com/custom-genomic-reference-workflow-reveals-hidden-transposable-element-activity-during-cellular-senescence/. Accessed 21 November 2026.
Beatrice Stafford. "Custom Genomic Reference Workflow Reveals Hidden Transposable Element Activity During Cellular Senescence." Scienmag. November 21, 2026. https://scienmag.com/custom-genomic-reference-workflow-reveals-hidden-transposable-element-activity-during-cellular-senescence/








