# Sample Annotation Errors in Genomics: Causes and Fixes If your samples cluster by sequencing lane or extraction date instead of by disease status, suspect a sample identity error first. In most genomics projects, the single most likely cause of an annotation problem is not a subtle algorithmic flaw but a mundane one: two samples swapped, a plate map transcribed by hand, or a metadata spreadsheet with the wrong column alignment. The first thing to check is whether sample identities hold up against independent biological evidence, such as sex markers, known genotypes, or pedigree relationships. If those checks pass, move on to reference and annotation build mismatches, which are the next most frequent source of errors that quietly corrupt a whole analysis. This article covers the full range of sample annotation errors in genomics, from mislabeled tubes to non-canonical gene models, explains the mechanism behind each failure, and gives concrete fixes. The advice applies to short-read and long-read projects, to human and non-model organism data, and to anything in between. ## What a Sample Annotation Error Actually Is The phrase covers two distinct families of problems, and confusing them wastes diagnostic time. **Sample-level annotation errors** are mistakes in the information attached to a sequencing sample. This includes mislabeled tubes, swapped library indices, incorrect metadata (wrong tissue, wrong time point, wrong case-control status), and contamination from another sample. A sample-level error poisons every downstream result for that library. **Feature-level annotation errors** are mistakes in the biological labels applied to the data after sequencing. This includes wrong gene coordinates because of a genome build mismatch, gene symbols that map to the wrong organism, homology-based functional annotations transferred from the wrong species, transcript models that do not reflect real biology, and variant annotations that misclassify a clinically relevant change. Both families share a common failure mode: the pipeline runs to completion and produces plausible-looking numbers, so the error survives until someone notices a result that contradicts prior biology. Detection therefore depends on deliberate validation steps, not on pipeline success. Reviews of sequencing standards emphasize that technical artifacts and contamination can enter at every point of sample manipulation, extraction, sequencing, and analysis, which is why benchmarking and error measurement belong in every project plan [1]. ## Troubleshooting Table This table is the working core of the article. Find your symptom, confirm the cause, then apply the fix. The sections that follow explain each mechanism in detail. | Symptom | Likely cause | How to confirm | Fix | |--|--|--|--| | Samples cluster by batch, plate position, or date rather than biology | Sample label swap or plate map [transcription error](/knowledge/molecular-biology/transcription-error) | Check sex markers, known genotype snapshots, or relatedness against pedigree | Trace the physical chain of custody, correct labels at the source, never patch labels in the analysis directory | | Child shows alleles absent from both parents | Sequencing artifacts or true sample mix-up in a trio | Filter on read depth, genotype quality score, and alternate allele ratio, then recheck | Apply quality filters, then investigate remaining errors as possible swaps [2] | | Gene coordinates look shifted, or genes are missing entirely | Genome build mismatch (for example, aligning to GRCh37 while annotating with a GRCh38 GTF) | Inspect the reference in the BAM header and compare against the annotation file version | Realign against the correct build or lift the coordinates with a tested chain file, then revalidate a known gene | | Functional enrichment returns pathways impossible for your organism | Homology-based annotation transferred incorrectly from a distant species | Spot-check annotations against primary databases and literature for the organism | Build a custom organism-specific annotation package from source data [3] | | A flood of novel transcripts appears in long-read RNA-seq | Technology errors indistinguishable from rare real molecules | Look for replicate evidence, full-length support, and consistency across samples | Use methods validated by community benchmarking and treat single-read novel models cautiously [4] | | Indel calls look unreliable on nanopore data | Chemistry error profile, older pore versions | Compare performance against reference samples, check chemistry version | Use current chemistry, apply tuned callers, validate clinically important indels orthogonally [5] | | Mitochondrial genes misannotated, missing stop codons or wrong exon boundaries | Non-canonical start and stop codons, RNA editing, trans-splicing | Compare against a curated, evidence-backed reference for the organism | Use a validated mitochondrial annotation pipeline rather than generic gene prediction [6] | | Pathway map rejected by visualization or modeling software | Violations of format rules (for example, SBGN process description rules) | Run a dedicated validator that reports errors with explanations | Correct the map, or apply automated error resolution where feasible [7] | | Rare cell types missing from a single-cell analysis | Clustering method overfitting dominant patterns and technical noise | Look for known rare populations that fail to appear, test on synthetic mixtures | Use embedding and clustering methods designed to recover rare populations and resist noise [8] | | Abundant variants with low or inconsistent allele frequencies in viral samples | Mixed true variants, artifacts, and annotation noise in dispersed output files | Consolidate results into a single annotated summary with frequencies and affected genes | Use summarization workflows that link each significant variant to a stable identifier [9] | | Calls contradict well-characterized reference material | Untested pipeline or unquantified error rates | Benchmark against characterized reference samples before production | Run a production-readiness assessment covering SNVs, indels, STRs, SVs, and CNVs [5] | ## Causes of Sample Annotation Errors, From Most Common to Rarest ### 1. Human label and metadata errors ![Decision path that starts with genotype concordance and ends in tracing the physical chain of custody at the bench.](/article-images/illustrations/sample-annotation-errors-genomics-swap-triage.webp) The most common cause is also the least technical. Samples get swapped during extraction, plate maps get rekeyed by hand, spreadsheet rows shift when columns get sorted, and case-control labels drift between the clinical database and the sequencing manifest. These errors are frequent because sample annotation is the one step in genomics performed primarily by humans under time pressure, usually at the bench, often with paper in the loop. The mechanism is simple. Every downstream record inherits its identity from the label applied at collection. A barcode scanner misread by one character, a tube placed in the wrong rack position, or a header row accidentally sorted with the data creates a divergence between physical identity and recorded identity that no amount of computational care can repair afterward. **Fixes.** Use barcode labels and scanned reads at every physical transfer. Generate plate maps programmatically from the sample registry rather than by hand. Record metadata in controlled vocabularies with fixed column definitions, and freeze the manifest before library preparation begins. After sequencing, run an independent identity check on every sample: compare sex markers against recorded sex, compare genotype calls against any prior SNP array data, and verify expected relationships in structured cohorts such as trios. International standards work stresses that measurement and benchmarking of potential error sources at the initial input stage are foundational to everything downstream [1]. **Diagnostic sequence.** If you suspect a swap, compute pairwise genotype concordance across all samples in the batch. A pair of samples with near-perfect cross-concordance and poor self-concordance against their recorded identities identifies the swap precisely. Then trace the physical chain of custody, because the computational fix is only a stopgap. The durable fix is at the bench. A short safety note applies here. When source material is human blood, buccal cells, or saliva, as in family-based sequencing studies, handle specimens under your institution's bloodborne pathogen and human subjects procedures, and treat genotype data themselves as identifying information that requires controlled access. ### 2. Index cross-talk and cross-contamination during library preparation Even a perfectly labeled sample can acquire someone else's sequence. On patterned flow-cell platforms, index hopping, meaning the assignment of reads to the wrong sample index during cluster generation, transfers a small fraction of reads between libraries multiplexed in the same run. Carryover contamination from previous runs and aerosolized template during library setup add further mixing. Reviews of sequencing standards identify contamination as a persistent artifact class across whole genomes, transcriptomes, and metagenomes, arising at each point of sample manipulation [1]. The annotation consequence is subtle but real: low-level contaminating variants get attributed to the host sample, and in RNA work, transcripts from a contaminant can be misannotated as sample-derived expression. **Fixes.** Use unique dual indices where the platform supports them. Include library-level negative controls and cross-sample contamination checks in routine quality control. For human data, estimate contamination from patterns such as sex marker heterozygosity and ancestry-informative markers. Set a contamination threshold before the project starts, document it in the analysis plan, and treat samples above it as fail-and-reextract rather than fail-and-reanalyze. ### 3. Reference genome build and annotation version mismatch This is the most common feature-level error and the one most likely to affect an entire project at once. The failure mechanism is a coordinate space mismatch. Alignment happens against one reference assembly, for example GRCh37, while gene annotation, variant annotation, or interval arithmetic uses a different assembly, for example GRCh38. Because chromosome coordinates are not shared between builds, a gene model applied in the wrong coordinate space points at the wrong genomic intervals, and downstream annotation lands on the wrong bases. A related failure is version drift within a single build. Annotation databases release updated gene models regularly, and a pipeline that pinned its reference in one year but its annotation in another will produce genes with shifted boundaries, merged or split transcripts, and inconsistent identifiers. **Fixes.** Record the exact reference assembly and annotation release for every dataset, ideally in the file itself as well as in project documentation. Align and annotate within a single consistent coordinate system. If you must convert between builds, use a maintained liftOver chain and revalidate a panel of known genes afterward. Before accepting any annotation output, confirm that a handful of well-characterized genes have plausible exon structures and protein products. A practical diagnostic uses the alignment file itself. The BAM header records the reference, and the sequence dictionary exposes any mismatch with your annotation file: ``` samtools quickcheck sample.bam samtools view -H sample.bam | head -20 ``` If the header names a different assembly than your GTF or GFF3 file, stop and reconcile before interpreting anything. ### 4. Identifier and ID-mapping failures Gene identifiers are the plumbing of genomics, and they leak. A gene symbol can refer to different entities over time as nomenclature changes, versioned accession numbers get superseded, and different databases use different primary keys. When an analysis joins an expression table to an annotation table on a mutable key such as a symbol, entries silently drop out or, worse, join to the wrong record. **Fixes.** Carry a stable, versioned identifier (an Ensembl gene ID with its version, for example) as the primary key through every step, and map to display symbols only at the final reporting layer. Log every identifier that fails to map, and review that list rather than discarding it. When the join is done, verify that the number of mapped features matches expectations for the annotation release. ### 5. Homology-based annotation transfer in non-model organisms For organisms without curated functional annotations, especially bacteria and other microorganisms, a standard practice is to transfer functional labels from a closely related species using similarity searches. This strategy yields useful insight but introduces a specific error class: annotations transferred from an inappropriate source, species-specific functions overlooked entirely, and labels that propagate through the literature without ever being experimentally checked [3]. The mechanism is that sequence similarity does not guarantee functional equivalence, particularly over large evolutionary distances or in protein families with many paralogous members. Once a transferred annotation enters an analysis, enrichment results inherit its errors, and the errors look like biology. **Fixes.** Where curation exists, use it. Where it does not, build organism-specific annotations directly from source genomic data rather than relying on indirect transfer. A demonstrated approach constructs custom annotation R packages from data retrieved from NCBI, imports them into the R environment, and integrates with the clusterProfiler package for gene ontology and [pathway enrichment analysis](/blog/guides/pathway-enrichment-analysis-turning-a-gene-list-into-testable-questions), with Mycobacterium tuberculosis H37Rv shown as a worked example [3]. Whatever transfer method you use, document the source organism, the search parameters, and the cutoff, so that a reviewer can retrace the annotation. ### 6. Sequencing technology error profiles Every platform has a characteristic error spectrum, and annotation pipelines that ignore it misattribute technical noise to biology. Short-read platforms have very low raw error rates but struggle with repetitive and structurally variant regions. Long-read platforms, including Oxford Nanopore and PacBio, resolve those regions but carry higher raw error rates, and their indel performance historically lags their single-nucleotide performance. A production-level assessment across 17 well-characterized reference samples measured Oxford Nanopore accuracy at F-scores of 0.978 to 0.983 for SNVs and around 0.75 for structural variants, with indels weaker at 0.659 to 0.758, while still correctly detecting all four pathogenic indels in the test set and showing improvement with the newer R10 chemistry [5]. The practical lesson is that platform-specific error profiles demand platform-specific quality control, not a generic filter tuned for a different sequencer. Short-read artifact patterns follow the same logic. In a trio sequencing study that used samples from blood, buccal cells, and saliva, variants that violated [Mendelian inheritance](/knowledge/molecular-biology/mendel-heredity-laws-of-inheritance-explained-simply) served as a proxy for artifacts, and filtering on read depth, genotype quality score, and alternate allele ratio removed about 95 percent of Mendelian errors while retaining 80 percent of called variants. Concordance between identical samples from different DNA sources rose from 87 percent before filtering to 99.99 percent after [2]. The mechanism behind those numbers is that true variants carry consistent evidence depth, quality, and allele balance, while artifacts cluster at low depth, low genotype quality, and skewed allele ratios. **Fixes.** Apply filters matched to your platform and assay. For trio or family designs, Mendelian error screening is a powerful, nearly free quality control layer. For long-read data, track chemistry version, use callers tuned for the platform, and validate clinically significant findings with an orthogonal method. For any design, sequence the same sample from an independent source when feasible and measure concordance directly. ### 7. Structural and non-canonical biology that breaks annotation assumptions Some annotation errors are not pipeline mistakes at all. They arise because real biology violates the assumptions baked into annotation tools. Plant mitochondrial genomes illustrate the problem sharply. Their structural dynamism and frequent annotation errors stem from features that standard pipelines do not model: trans-splicing, RNA editing, and non-canonical start and stop codons. A curated high-confidence mitochondrial reference for Rhodiola, built by integrating transcriptomic evidence, public resources, and experimental validation, corrected exon-intron boundaries and validated atypical 5-prime terminal variants in specific genes, and subsequent re-annotation of public mitogenomes revealed widespread structural misannotation including cross-chromosomal gene arrangements [6]. Long-read [transcriptomics](/knowledge/bioinformatics/modern-transcriptomics-bulk-single-cell-spatial) raises the same issue from the opposite direction. Full-length single-molecule sequencing captures tens of thousands of novel transcripts, including lowly expressed, often sample-specific molecules that short-read methods overlooked. Community benchmarking validated many of these, but it also highlighted the core interpretive problem: distinguishing true rare molecules from technology errors requires new experimental and computational methods, and the field must decide how to capture molecular variability while still maintaining reference annotations useful to the community [4]. **Fixes.** Match your annotation strategy to the biology. For genomes with known non-canonical features, use annotation systems that explicitly model them, such as the algorithms developed for mitochondrial assembly that handle trans-splicing, RNA editing, and non-canonical codons [6]. For transcript discovery, require replicate and full-length support before promoting a novel model to an annotation. For genome quality itself, use established completeness metrics, as in a recent seabird genome assembly that reported 99.95 percent BUSCO completeness and chromosomal-level scaffolding against a related species reference [10]. High completeness and careful reference-based scaffolding reduce the chance that assembly errors masquerade as annotation errors downstream. ### 8. Format and schema violations Annotation lives in files, and files have rules. Standardized formats such as SBGN, the Systems Biology Graphical Notation, impose strict rules on how biological entities and their interactions are depicted. In practice, many users overlook or disregard those rules, which produces maps that fail validation and silently misrepresent the biology they intend to convey [7]. The same class of failure applies to GTF, GFF3, VCF, and metadata schemas. An unescaped character in a metadata field, an attribute written with the wrong key, or a coordinate off by one between zero-based and one-based conventions can propagate through an entire pipeline without raising an error. **Fixes.** Validate files at ingestion, not after analysis. Tools designed for this purpose exist across the ecosystem. In pathway modeling, one open-source web service validates a map in SBGN process description format, documents each error with an explanation, displays errors color-coded on an automatically laid out map, and can attempt automated error resolution where feasible, updating both the error list and the map image [7]. Apply the same principle everywhere else: run a validator, read the errors, and fix the file before the pipeline consumes it. ### 9. Computational pipeline, variant interpretation, and clustering errors The final category includes errors introduced by the analysis itself. In viral genomics, next-generation sequencing results arrive dispersed across genome assemblies, variant calls, and functional annotations, and virologists must manually sift through extensive lists of low-frequency variants, a process that is time-consuming and error-prone. Consolidating key results into a concise annotated summary, with each significant variant assigned a unique identifier linked to a high-confidence variant table with frequencies, flanking nucleotides, and affected genes and proteins, directly reduces manual interpretation errors [9]. A related risk comes from building pipelines that nobody has stress-tested. Somatic variant calling remains difficult because of cancer heterogeneity, complex mutational profiles, and unpredictable sequencing errors, and a systematic analysis of nine somatic sample simulators found that no single tool provided complete control over both biological parameters, such as variant type, number, position, and clonality, and technical parameters, such as sequencing errors, coverage, and base qualities [11]. Simulation remains the only practical way to benchmark a variant caller against known truth, and the benchmarking step is where annotation bugs surface before they reach patients or publications. In single-[cell genomics](/blog/guides/cell-genomics), annotation errors take the form of missed biology. Existing dimensionality reduction and clustering methods can overfit dominant patterns while missing unique ones, which impairs detection and annotation of rare cell types, and they can fit technical noise rather than biological signal. Robust self-supervised approaches address this by using reconstruction error to redirect attention to poorly embedded cells and by using balanced consensus learning to dampen the influence of low-quality samples [8]. **Fixes.** Benchmark before production using characterized reference samples [5] and truth-set simulations [11]. Consolidate and cross-reference variant outputs so that every reported variant carries a stable identifier and traceable provenance [9]. For single-cell work, verify that known rare populations are recoverable before trusting annotations of unknown populations [8]. ## A Diagnostic Sequence You Can Follow When an annotation error appears, work through this sequence in order. The steps move from cheapest to most expensive and from most likely to least likely causes. 1. **Confirm the error is real.** Reproduce it from the raw files in a clean directory with a recorded software environment. Many reported annotation errors are irreproducibility artifacts. 2. **Check sample identity.** Run sex marker checks, genotype concordance against prior data, and relationship checks in structured cohorts. In trio data, quantify Mendelian errors and apply depth, genotype quality, and allele ratio filters before concluding that a sample is swapped [2]. 3. **Check for contamination.** Estimate cross-sample contamination and look for unexpected taxonomic signal. Standards reviews identify contamination as an artifact that can enter at any manipulation step [1]. 4. **Verify the reference and annotation versions.** Read the BAM header, compare the assembly against your GTF or GFF3, and confirm that both come from the same release. Revalidate a panel of known genes. 5. **Verify identifier mapping.** Count mapped features, inspect unmapped identifiers, and confirm that symbols resolve to the intended records in the intended organism. 6. **Inspect platform error profiles.** Compare observed variant quality distributions against platform expectations. For long-read data, confirm the chemistry version and check indel performance specifically, given its documented weakness relative to SNV performance on nanopore platforms [5]. 7. **Check for format violations.** Run validators on every annotation artifact, including pathway maps [7], gene models, and metadata tables. 8. **Question the annotation source.** For non-model organisms, audit homology-transferred functional labels against primary evidence and consider building a custom annotation package from source data [3]. 9. **Reconcile with independent biology.** In mitochondrial and other structurally unusual genomes, compare against curated references that model trans-splicing, RNA editing, and non-canonical codons [6]. For novel long-read transcripts, require replicate or full-length support [4]. 10. **Fix at the source, then rerun end to end.** Correct labels at the bench, rebuild references in one version, and rerun the full pipeline. Never hand-edit intermediate results, because the patch will not survive the next rerun. ## Common Mistakes and Limitations These are the errors that experienced analysts still make, stated specifically so you can check for each one. **Treating the manifest as ground truth.** The spreadsheet that arrived with the samples is itself a candidate for the error. Independent biological evidence outranks recorded metadata every time. **Patching labels in the analysis instead of at the source.** Renaming samples in R or in a config file fixes one analysis and guarantees the same swap reappears in the next one. The correction belongs in the sample registry and in the physical record. **Mixing genome builds silently.** Aligning to one assembly while annotating with another produces no warning in most tools. Check the header and the annotation release on every new dataset. **Trusting unfiltered variant calls.** Without quality filters on depth, genotype quality, and allele balance, a large fraction of calls in family-based data are artifacts, and concordance between identical samples sequenced from different tissues can sit near 87 percent rather than above 99 percent [2]. **Assuming annotation completeness.** Well-annotated model genomes still yield tens of thousands of novel transcripts under [long-read sequencing](/knowledge/bioinformatics/long-read-sequencing-technologies-pacbio-and-oxford-nanopore), and the field is still deciding how to fold molecular variability into shared reference annotations [4]. Novelty in your data may be biology, but it may also be error, and the discrimination step is not optional. **Ignoring non-canonical biology.** Generic annotation pipelines miss trans-splicing, RNA editing, and non-canonical start and stop codons, which produces systematic misannotation in mitochondrial and other nonstandard genomes [6]. **Skipping benchmarking.** A pipeline that has never faced characterized reference samples or simulated truth data has an unknown error rate. Production readiness is demonstrated, not assumed [5]. **Over-trusting automated clustering annotations.** Single-cell label transfer can miss rare populations and absorb technical noise, so an annotation that looks clean may simply reflect the dominant signal [8]. **One limitation worth stating plainly.** Generic advice covers the systematic patterns. Individual datasets need project-specific validation, and clinical decisions in particular require confirmation by qualified professionals using validated clinical pipelines. ## Prevention Through Bioinformatics Data Management ![Four prevention principles: checksummed raw data, versioned references, programmatic sample tracking and hard validation gates.](/article-images/illustrations/sample-annotation-errors-genomics-prevention-pillars.webp) Most annotation errors are prevented before sequencing begins, through disciplined [data management](/blog/guides/data-management-basics-principles-processes-and-best-practices) rather than clever analysis. **Immutable raw data with checksums.** Compute a checksum such as an MD5 or SHA-256 digest for every raw file at creation, store it alongside the file, and reverify before every major analysis: ``` md5sum sample_R1.fastq.gz > sample_R1.fastq.gz.md5 md5sum -c sample_R1.fastq.gz.md5 ``` Raw data are never modified. All transformation happens in derived files that can be regenerated from the raw layer. **Version everything.** Record the reference assembly, annotation release, software versions, and parameter settings for every result. Containerize pipelines so that the environment is part of the record. A result that cannot be tied to its exact inputs and versions is not reproducible, and irreproducibility is annotation error waiting to happen. **Programmatic sample tracking.** Generate manifests, plate maps, and per-sample metadata from a single authoritative registry. Never rekey identifiers by hand. Every physical transfer of a tube or a library gets a scanned confirmation. **Validation gates in the pipeline.** Build automated checks that stop the pipeline rather than warn and continue. Sample identity concordance, contamination estimates, reference and annotation version agreement, and file format validation all belong as hard gates. The experience of pathway validation shows the value of explicit error reporting, where each violation is documented with an explanation and displayed in context rather than dumped as a log line nobody reads [7]. **Genomics data storage practices.** Store data with a clear tiering scheme: immutable raw reads in durable archival storage, aligned and processed files in working storage with regeneration scripts, and result tables in a versioned, queryable form. Apply retention policies that match the sensitivity of the data, and treat human genomic data as identifiable health information with corresponding access controls. Backups are only real if restoration has been tested. Metadata, including sample annotations, belong in the same disciplined storage regime as the sequence data, because an annotation error and a corrupted file are equally fatal to an analysis. **Standards as the baseline.** International standards work in genomics emphasizes that from initial input material through computational pipelines, standards, tools, and metrics exist to quantify the accuracy of sequencing methods, and that these measurements are essential for robust clinical genomics and precision medicine [1]. Adopting such standards early is far cheaper than diagnosing annotation errors after publication. ## Frequently Asked Questions ### What is a sample annotation error in genomics? It is a mismatch between the recorded identity or biological label of a sample and its true identity or true biology. It covers both mislabeled or swapped samples and incorrect gene, transcript, variant, or functional annotations applied to the data. ### What is the most common cause of sample annotation errors? Human error in labeling and metadata is the most common cause, including swapped tubes, hand-transcribed plate maps, and spreadsheet misalignment. Reference build mismatches between alignment and annotation files are the most common feature-level cause. ### How do I detect a sample swap? Compare each sample against independent biological evidence such as sex markers, prior genotype data, and known family relationships. In trio data, children should inherit one allele from each parent, and violations that survive quality filtering can flag swaps or artifacts [2]. ### How do I fix a genome build mismatch? Identify the assembly recorded in the BAM header, compare it against the annotation file, and realign or lift coordinates to a single consistent build using a maintained conversion chain. Then revalidate known genes before rerunning the analysis. ### Are long-read sequencing errors a source of annotation errors? Yes. Long-read platforms carry higher raw error rates than short reads, with documented indel weakness relative to SNV performance, so error profiles must be handled with platform-tuned callers and quality control [5]. Long reads also surface rare transcript molecules that must be distinguished from technology errors [4]. ### Why do non-model organism annotations contain so many errors? Because functional annotations are often scarce and are frequently transferred by homology from related species, which introduces incorrect labels and misses species-specific functions. Building custom annotations from primary source data reduces this error class [3]. ### How can I prevent sample annotation errors in my own project? Use barcode-based sample tracking, programmatic manifest generation, checksummed immutable raw data, versioned references and annotations, and automated validation gates that stop the pipeline on identity, contamination, or format failures. ### When should an annotation error be considered clinically significant? Any annotation that informs a patient's diagnosis or treatment requires confirmation in a validated, production-assessed pipeline with documented performance against reference materials, because even strong platforms have error profiles that vary by variant class [5]. ## Sources 1. [International Standards for Genomes, Transcriptomes, and Metagenomes.](https://pubmed.ncbi.nlm.nih.gov/28337071/) 2. [The struggle to find reliable results in exome sequencing data: filtering out Mendelian errors.](https://pubmed.ncbi.nlm.nih.gov/24575121/) 3. [Enhanced RNA-Seq Expression Profiling and Functional Enrichment in Non-model Organisms Using Custom Annotations.](https://pubmed.ncbi.nlm.nih.gov/42367215/) 4. [Notable challenges posed by long-read sequencing for the study of transcriptional diversity and genome annotation.](https://pubmed.ncbi.nlm.nih.gov/40032585/) 5. [Assessing the readiness of Oxford Nanopore sequencing for clinical genomics applications.](https://pubmed.ncbi.nlm.nih.gov/41672886/) 6. [RhoMitoAnnotator and Polypods, Bioinformatics Tools for the Rhodiola Mitochondrial Gene Assembly, Annotation and Phylogenetic Analysis.](https://pubmed.ncbi.nlm.nih.gov/42196420/) 7. [SyBValS: a validation and error resolution service for biological pathway maps.](https://pubmed.ncbi.nlm.nih.gov/41454279/) 8. [Robust self-supervised machine learning for single cell embeddings and annotations.](https://pubmed.ncbi.nlm.nih.gov/40502088/) 9. [vvv2_align_SE, vvv2_align_PE/vvv2_display: Galaxy-Based Workflows and Tool Designed to Perform, Summarize and Visualize Variant Calling and Annotation in Viral Genome Assemblies.](https://pubmed.ncbi.nlm.nih.gov/41157655/) 10. [First whole-genome assembly of the Galápagos Petrel (Pterodroma phaeopygia) using Oxford Nanopore sequencing to advance conservation genomics in a critically endangered seabird.](https://pubmed.ncbi.nlm.nih.gov/41589066/) 11. [MOV&RSim: computational modelling of cancer-specific variants and sequencing reads characteristics for realistic tumoral sample simulation.](https://pubmed.ncbi.nlm.nih.gov/41310436/)