# Why Are My Reads Mapping to the Wrong Genome? The most common reason reads map to the wrong genome is that you aligned against the wrong reference file, most often an outdated assembly build or a reference from the wrong organism that was left over from a previous project. The single fastest check takes under a minute: run `samtools idxstats` on your BAM file and look at the sequence names in the first column. If you see `chr1` in a project you believed used Ensembl naming, or you see chromosome names from a different species, you have found your problem before touching any other tool. This guide is a practical bioinformatics troubleshooting reference for the situation where alignments land on the wrong genome, the wrong chromosome, the wrong gene, or a paralog instead of the true origin. It covers the causes in order of how often they bite real pipelines, the mechanism behind each failure, a diagnostic sequence you can follow, and the fixes that actually work. ## What "Mapping to the Wrong Genome" Actually Means The phrase covers several distinct failures, and each has a different fix. Before you troubleshoot, work out which of these you are seeing, because the symptoms overlap. **Wrong assembly build.** Your reads are from the right species, but the reference is GRCh37 when you assumed GRCh38, or mm9 when the rest of your analysis assumes mm10. Coordinates in the BAM will not match your annotation files, downstream tools will report genes in impossible places, and liftOver-style coordinate conversions will fail silently or produce garbage. **Wrong species entirely.** Alignments go to a different organism. This usually means sample swap, reference mixup, contamination in the library, or an index-building error. If a human sample produces BAM files full of mouse chromosomes, something upstream of the aligner went badly wrong. **Cross-mapping to paralogs or homologs.** The reads genuinely belong to your organism, but the aligner places them at a paralogous gene, a duplicated region, or a closely related sequence in the reference. In metagenomic and metagenotyping work this failure is severe enough that reads can be assigned to the wrong species altogether, and researchers have shown that growing databases of closely related genomes push standard alignment algorithms past their limits, with multi-mapping reads acting as a prevalent source of these wrong-species assignments [1]. **Contamination and ambient background.** A fraction of reads comes from another genome carried into the library by reagent carryover, ambient material, index hopping, or doublets. Single-cell combinatorial indexing methods are especially exposed, because ambient chromatin accumulates through the workflow and distorts profiles most strongly in low-depth nuclei and in minority populations [2]. **Spurious placements from sequencing error.** Reads with enough base errors can align better to a wrong locus than to their true origin. This is a classic RNA-seq error mode when the aligner is asked to pick one placement from many near-equal candidates and simply chooses the one with the fewest mismatches, which can be the wrong assignment [3]. **Reference problems.** The reference itself contains errors, gaps, or mis-assembled contigs, so reads from the true locus either fail to map or map to an incorrect sequence elsewhere. Computationally annotated RefSeq records have historically contained insertions, deletions, and base errors that propagate into any analysis built on them [4]. De novo assemblies from short reads are also prone to wrong assemblies, where sequences are deleted or arranged incorrectly, and detecting those mis-assembled contigs reliably remains an open problem [5]. Knowing which category you are in determines everything that follows. ## Start Here: A Fast Diagnostic Sequence ![Decision path that starts from contig names in the BAM and routes to build, species, contamination or multimapping checks.](/article-images/illustrations/reads-mapping-to-wrong-genome-triage-wrong-genome.webp) Run these checks in order. Most "wrong genome" tickets are resolved by step 1, 2, or 3. **1. Verify the reference you actually used.** ```bash samtools idxstats aligned.bam | head samtools dict reference.fa | head ``` `samtools idxstats` prints the sequence names in your BAM. Compare them to your expectations. `chr1` versus `1` tells you whether you used a UCSC-style or Ensembl-style reference. A count of contigs numbered in the tens of thousands suggests you aligned to a scaffold-level assembly rather than a chromosome-level one. `samtools dict` prints the reference dictionary from your FASTA, which records the assembly you should have used. **2. Confirm the species composition of the alignments.** ```bash samtools flagstat aligned.bam samtools idxstats aligned.bam | awk '{sum+=$3} END {print sum " reads mapped"}' ``` If your BAM contains references from more than one organism, `idxstats` exposes it immediately. If the BAM looks internally consistent but you still suspect a species problem, map a subsample against a second reference and compare mapping rates. **3. Check the alignment summary metrics.** ```bash picard CollectAlignmentSummaryMetrics R=reference.fa I=aligned.bam O=metrics.txt ``` An overall mapping rate far below expectation for your organism and protocol (for a clean human whole-genome library, most labs expect the large majority of reads mapped) points at contamination, barcode or index problems, or the wrong reference. For RNA-seq, remember that intronic and intergenic reads depress exonic counts without being a mapping failure. **4. Inspect where the reads are going.** ```bash samtools view aligned.bam | head -20 bedtools genomecov -bga -ibam aligned.bam | awk '$4>0' | head ``` Look at a handful of aligned records directly. Read names can reveal sample identity issues. A pileup concentrated on one unexpected chromosome suggests contamination or a duplicated region rather than a systematic reference problem. **5. Subsample and re-map against a control reference.** ```bash seqkit sample -s 100 -p 0.01 reads.fq.gz > subset.fq bwa mem -M -t 8 reference.fa subset.fq | samtools sort -o test.bam samtools flagstat test.bam ``` If the same library maps well to a reference you trust and poorly to the one in your pipeline, the reference is the problem. If it maps poorly to both, the problem is in the library or the reads themselves. **6. Check read quality and adapter content.** ```bash fastqc subset.fq ``` Low per-base quality toward the read ends inflates mismatches and drives spurious mappings. Adapter dimers and overrepresented sequences also degrade alignment. If all six checks pass and you still see biologically impossible placements, move down the cause list below, because the remaining explanations are subtler. ## Troubleshooting Table | Symptom | Likely cause | How to confirm | Fix | |--|--|--|--| | BAM coordinates do not match annotation GTF or BED files | Wrong assembly build (for example GRCh37 vs GRCh38, mm9 vs mm10) | Compare contig names and lengths with `samtools dict`, `cut` on the GTF, and a CHROM-sizes diff | Rebuild the index against the correct build, realign, and re-derive all coordinate-based files | | Alignments spread across many chromosomes with similar scores | Multi-mapping reads in paralogous or duplicated regions | Count secondary and supplementary alignments with `samtools flagstat`, inspect MAPQ distribution | Report multi-mappers honestly, filter by MAPQ, or use a context-aware or probabilistic assignment method | | A subset of reads consistently lands on a different species' contigs | Contamination in the library or ambient background | Map a subsample to a two-species combined reference, count reads per species | Identify the contamination source, decontaminate at read level, exclude affected libraries, or model and subtract the ambient profile | | Many reads with the same barcode behave inconsistently | Index hopping, barcode collision, or doublets | Check barcode whitelist compliance, per-barcode read counts, and multiplet rates | Use demultiplexing tools that model ambient signal, filter doublets, adjust index assignments | | Splice junction reads are lost or genes appear at wrong loci in RNA-seq | Aligning splice-containing reads with a contiguous genome mapper, or per-read greedy mismatch resolution | Compare junction counts and gene assignments across two aligners | Use a splice-aware mapper, or refine initial alignments using context information from other reads | | Long reads align across a breakpoint but call the wrong structure | [Structural variant](/blog/guides/structural-variant), especially an inversion, that the reference does not represent | Inspect split and supplementary alignments manually in a genome viewer | Realign with a method designed for the variant class, and consider a reference that carries the allele | | High mismatch rates at read ends, spurious placements | Sequencing error accumulation, low-quality tails | FastQC per-base quality plots, PHRED distribution review | Trim low-quality tails, apply quality-aware filtering, raise minimum base quality during alignment | | Mapping rate drops after adding new genomes to a database | Database growth has made near-identical sequences multi-mappable | Compare metagenotyping calls across database versions | Restrict to marker genes or use methods designed for dense reference panels | ## Cause 1: The Wrong Reference Build or Version ![Calculation showing about 5 percent more mapped reads, roughly 120,000 extra CpG calls and about 20,000 wrong calls avoided.](/article-images/illustrations/reads-mapping-to-wrong-genome-e14-personal-reference.webp) This is the failure everyone eventually commits. Reference genomes are revised constantly, and every revision moves coordinates, merges or splits scaffolds, adds sequences, and patches errors. GRCh37 and GRCh38 use different coordinate systems for large regions of the [human genome](/blog/guides/human-genome). mm9 and mm10 differ even more substantially for the mouse. If your pipeline mixed builds, for example aligning against GRCh38 while calling peaks against GRCh37 annotation, every downstream coordinate comparison is wrong even though the alignments themselves are fine. The failure has a second, less obvious face. Even a perfectly correct standard reference can be a poor match for your specific sample. A study of the widely used E14 mouse embryonic stem cell line sequenced the line's own genome with more than 2 billion Illumina reads, built a database of about 2.7 million single nucleotide variants, and then constructed a line-specific reference assembly. Aligning against that personalized reference increased the number of mapped reads by about 5 percent relative to the mm9 genome reference, recovered roughly 120,000 additional CpG calls in reduced representation [bisulfite sequencing](/knowledge/molecular-biology/bisulfite-sequencing), and avoided about 20,000 wrong CpG calls [6]. The lesson generalizes: a reference that carries your sample's actual sequence maps reads that a generic reference misplaces, particularly in assay types such as bisulfite sequencing where mismatches are already abundant. **How to confirm the build problem.** Sequence names and lengths are the fingerprint. Download the chrom sizes for the build you intended and the build you suspect, and diff them. Tools like `samtools dict` and the `@SQ` lines in the BAM header record exact sequence lengths, and any discrepancy against your intended build is dispositive. Also check the alignment index creation date against the reference download date in your pipeline logs. **How to fix it.** Pick one build, document it, rebuild all indices, realign everything, and regenerate all coordinate-anchored files. Do not attempt to salvage a mixed-build analysis with coordinate conversion, because conversions between builds lose information at exactly the regions where builds disagree. For cell lines or strains where a personalized reference is feasible and the assay is mismatch-sensitive, consider building a sample-specific reference from the sample's own variant calls as the E14 work did [6]. ## Cause 2: Multi-Mapping Reads and Paralogs When a read's sequence appears at more than one locus in the reference, the aligner faces an assignment problem. Standard aligners typically report all placements as secondary alignments and pick one as primary, often the placement with the fewest mismatches. That heuristic is exactly the failure mode researchers flagged in RNA-seq work, where choosing the mapping with the fewest mismatches can produce wrong assignments driven by sequencing error rather than biology [3]. The same mechanism, amplified by database size, degrades metagenotyping. As genome collections grow and fill with closely related species and strains, more reads match multiple entries, and a documented trade-off emerges between retaining correct alignments and limiting incorrect ones, with many incorrect alignments sending reads to the wrong species [1]. Paralog-driven mis-mapping matters most when your question depends on per-gene or per-locus attribution. Gene families with high sequence similarity, segmental duplications, [transposable element](/knowledge/molecular-biology/transposable-element) families, and recent duplications all generate multi-mapping. Reads are not wrong. The reference genuinely contains multiple nearly identical positions, and any single-position assignment is a guess. **How to confirm it.** Look at the multimapping rate and the MAPQ distribution. ```bash samtools view -c -f 256 aligned.bam # secondary alignments samtools view -c -q 30 aligned.bam # alignments with MAPQ >= 30 ``` A MAPQ histogram bimodal at 0 and 60, with a large zero bin, tells you the aligner is uncertain about a large fraction of reads. For RNA-seq quantification, compare count distributions with and without multimappers included. Genes whose counts inflate dramatically when multimappers are retained are usually paralogous. **How to fix it.** The right fix depends on the question. For expression quantification, expectation-maximization or probabilistic assignment methods distribute multimapping reads across candidate loci in proportion to evidence, rather than committing each read to one locus. For variant calling, discarding MAPQ-low reads is standard because a read assigned to the wrong paralog creates false variant calls at both loci. For metagenotyping, evaluate strategies that limit incorrect assignments rather than maximizing retention, since the documented trade-off means you must choose a point on that curve deliberately [1]. In all cases, report the multimapping rate in your methods. A pipeline that silently discards half its reads in duplicated regions is not a pipeline anyone can reproduce. ## Cause 3: Contamination, Ambient Background, and Barcode Problems When reads map to a genome you did not sequence, the library itself carries foreign material. The routes differ by assay. In bulk libraries, contamination enters through sample handling, reagent carryover, cross-talk between indexed samples, or a sample swap upstream of extraction. The signature is a coherent subset of reads, often thousands to millions, mapping consistently to a foreign genome. A two-species joint alignment identifies it cleanly. In droplet-based and combinatorial-indexing single-cell assays, the dominant problem is ambient material. Ambient chromatin and ambient RNA accumulate through the experimental workflow, and their signal distorts profiles most drastically in low-depth nuclei and in minority populations, precisely the cells you most want to detect [2]. Standard cell calling relies heavily on read count thresholds, so ambient signal both inflates apparent cell counts and contaminates genuine cells. Existing decontamination approaches that operate on aggregated count matrices lose the read-level information needed to separate real signal from background. The AmbientMapper method addresses this by mapping reads competitively against multiple references, learning the ambient profile from empty and low-complexity barcodes, and classifying nuclei against background and singlets against doublets with a Bayesian Information Criterion. On a 26-genome reference panel with simulated read-level barcodes where every contaminating read is traceable, the approach achieved a wrong-genome rate of 0.19 percent [2]. That benchmark number is not a target you should expect from a naive pipeline. It is a demonstration that read-level competitive mapping against a multi-reference panel can drive cross-genome misassignment to a level where it stops distorting biology. **How to confirm contamination.** Map a random subsample of reads against a combined reference containing your organism plus the most likely contaminants, then count mapped reads per genome. For single-cell data, examine empty droplets and low-complexity barcodes directly, because their read composition is the best available estimate of the ambient profile [2]. A contamination profile concentrated in barcodes you would otherwise discard is diagnostic. **How to fix it.** Trace the source first. A single contaminated library suggests handling. All libraries from one batch suggest a reagent or instrument. Species-wide low-level signal in single-cell data is usually ambient and should be modeled and subtracted rather than chased physically. Use read-level decontamination tools for combinatorial indexing data rather than count-matrix corrections applied after the fact, and use multi-reference competitive mapping when several possible source genomes exist [2]. A brief data-handling note applies here. Human sequencing data, including contamination-derived reads, can contain identifiable genomic information. Treat raw FASTQ and BAM files from human samples under your IRB protocol and data-use agreement, keep them on controlled-access storage, and never upload them to public servers that do not enforce access control. ## Cause 4: RNA-Seq-Specific Mapping Errors RNA-seq has its own catalog of wrong-genome behavior, and a common RNA-seq error is assuming that a DNA aligner will handle transcript reads correctly. **The splice problem.** A read that crosses a splice junction has no contiguous match in the genome. Aligning RNA-seq reads to the genome with a standard contiguous aligner fails to place these reads, while aligning to a transcriptome reference places them but cannot detect unannotated splice sites. Either choice alone leaves a systematic hole in your data. **The ambiguity problem.** Even after solving splicing, per-read placement among multiple candidate positions is fragile. The ContextMap approach was built on the observation that assigning each read independently, choosing the candidate with the fewest mismatches, produces wrong assignments when sequencing error makes a wrong locus appear cleaner than the true one. ContextMap instead uses the context of a read, meaning information from other reads mapping to the same expressed region, to resolve ambiguities, and it allows a much larger degree of ambiguity in the initial stage so that all candidate positions can be found and then judged together [3]. The underlying insight applies beyond that tool: reads travel in cohorts from expressed regions, and a placement that isolates a single read from its cohort is suspicious. **How to confirm RNA-seq mapping errors.** Run two splice-aware aligners, for example STAR and HISAT2, on the same library and compare mapped-read counts, splice junction counts, and per-gene assignment. Large disagreements concentrated in genes with multiple isoforms or in paralogous families point at ambiguity handling rather than raw sensitivity. Check whether intronic retention events are being called in loci that are actually unannotated paralogs receiving misassigned reads. **How to fix it.** Use a splice-aware mapper for genome alignment. If your biological question involves isoform usage or genes in duplicated families, use a quantification or refinement step that pools evidence across reads from the same expressed region rather than committing each read independently, in line with the context-based strategy [3]. For novel splice junction discovery, align to the genome rather than a transcriptome, and require junction support from multiple independent reads before trusting a novel junction. ## Cause 5: Structural Variants the Reference Cannot Represent When a sample carries a structural variant, reads spanning the breakpoint have no single contiguous position in the reference. The aligner either splits the read, leaves it unmapped, or, worse, places both halves somewhere plausible but wrong. Inversions are the hardest case, because the anchors from inside an inversion region are nonlinear relative to anchors in regions without the variant, which defeats chaining heuristics tuned to collinear alignments. Long reads from PacBio or Oxford Nanopore platforms span structural variant breakpoints more frequently than short reads, which means existing long-read mapping methods generate wrong alignments and wrong variant calls more often on such variants, and a benchmark across simulated datasets at different genomes and sequencing coverages showed that the invMap algorithm located aligned regions and called inversion variants more accurately than competing methods, with the NA12878 human dataset showing more candidate inversion calls as well [7]. **How to confirm it.** Collect reads with split or supplementary alignments and inspect them in a genome viewer. An inversion shows a characteristic pairing of read segments from two loci with reversed orientation. Discordant read-pair clusters with unexpected orientation are the short-read equivalent. If a cluster of weirdly aligned reads concentrates at one locus, suspect an unrepresented variant rather than contamination. **How to fix it.** For long-read data on variant-dense samples, realign with a method designed to detect inversions within the alignment process rather than expecting a general-purpose aligner to surface them, then re-call variants [7]. For known variant-dense regions, a reference containing the alternate allele removes the conflict entirely. Report both the reference-restricted call and the variant-aware call when the biological question hinges on those loci. ## Cause 6: Errors Inside the Reference Itself Sometimes the reads map correctly to a reference that is itself wrong. Two failure classes are documented. **Annotation and sequence errors in curated databases.** Computationally derived RefSeq model records have contained single-base and segment-level insertions, deletions, and mutations at the cDNA level, identified by homologous BLAST searches against non-redundant databases and then validated by EST clustering support, splice-boundary consistency with the GT/AG rule, chromosomal mapping, and RT-PCR confirmation followed by cDNA sequencing [4]. Any of your reads aligning through such an erroneous region inherits the error. **Mis-assembled contigs in de novo assemblies.** Short-read de novo assembly produces contigs prone to wrong assembly, with sequences incorrectly deleted or arranged, and mis-assembly detection tools evaluated across different genome sequencing data still leave substantial room for improvement, which is why a machine-learning-based complement to existing detection tools has been proposed [5]. If you mapped against a draft or de novo assembly, a fraction of your alignments can be biologically meaningless regardless of how well they scored. **How to confirm reference problems.** For a suspicious locus, fetch the corresponding sequence from an independent source, for example a different assembly or a curated record, and align the two to each other. Discrepancies localized to one contig indicate an assembly problem. For gene-level questions, check that exon and intron boundaries in the reference annotation obey the canonical GT/AG splice rule, a validation step straight from the RefSeq error-correction methodology [4]. If your alignments to one genomic region behave impossibly, for example reads from a single transcript assigning to two unrelated loci with high scores, examine the underlying reference sequence directly. **How to fix it.** Prefer chromosome-level, well-curated assemblies for alignment. For de novo references, run mis-assembly detection before treating the assembly as ground truth, and treat flagged contigs as provisional [5]. For gene models, cross-validate against expression evidence before building an analysis on a computationally predicted record. ## Cause 7: Sequencing Error Driving Spurious Placements Base-calling errors are not evenly distributed, and clustering them matters. Analysis of Sanger reads compared against a known consensus found that low quality values do not perfectly flag miscalled bases, and window-based analysis showed that a PHRED quality cutoff of 7 masks most errors without masking genuinely correct windows, a result aimed at window-based applications such as seeding and clustering [8]. High-throughput platforms have their own error profiles, but the principle transfers: errors cluster, quality scores imperfectly predict them, and windows that look clean are more trustworthy than isolated high-quality bases flanked by bad ones. The mapping consequence is direct. A read whose true locus contains several variant bases relative to the reference can, after accumulating errors, match a paralogous or repetitive locus better than its origin. Greedy per-read placement then assigns it wrongly, a mechanism identified as a driver of wrong assignments in RNA-seq mapping [3]. **How to confirm it.** Look at the per-base quality profile and the mismatch distribution along the read. Spurious alignments dominated by mismatches at read ends are the classic signature of untrimmed low-quality tails. Compare mapping rates before and after aggressive quality trimming. If a large read set that previously placed nowhere starts mapping convincingly after trimming, error load was the [limiting factor](/blog/guides/limiting-factor-definition-types-and-examples). **How to fix it.** Trim adapters and low-quality tails before alignment. Set a minimum base quality during alignment so error-prone bases cannot seed alignments. For workflows that depend on exact windows, such as primer design, k-mer filtering, or seeding, apply a window-level quality criterion informed by the PHRED window analysis rather than judging each base in isolation [8]. Distinguish trimming from over-trimming by monitoring both mapped fraction and per-base quality retention. ## How to Fix Reads Mapping to the Wrong Genome: A Consolidated Checklist Work through this list when you know the cause category but need an actionable sequence. 1. **Lock the reference.** One organism, one assembly build, one annotation release. Record filenames, checksums, and download URLs in the pipeline log. 2. **Rebuild indices after any reference change.** Aligner indices, splice-junction indices, and annotation indices must all come from the identical FASTA. A STAR index built from one genome paired with a GTF from another is a guaranteed source of wrong placements. 3. **Realign and compare.** Keep the old BAM until the new one passes your summary metrics. Compare mapping rate, duplicate rate, insert size distribution, and per-chromosome read counts. 4. **Treat multimapping as information, not noise.** Decide explicitly whether your question needs single-placement resolution or probabilistic assignment, and set filters accordingly. Document the choice. 5. **Model contamination instead of discarding it.** For single-cell combinatorial data, estimate the ambient profile from empty and low-complexity barcodes and remove contaminants at read level [2]. 6. **Validate suspicious loci by hand.** Open the region in a genome viewer, inspect individual alignments, and compare the reference sequence against an independent source. 7. **Re-run variant calling after any reference or alignment fix.** Variant calls are downstream of alignment and inherit every alignment error. Wrong CpG calls in bisulfite data from a mismatched reference illustrate how deep the damage runs [6]. ## Prevention Practices **Pin versions in the pipeline.** Store the reference FASTA, its checksum, the index build date, and the annotation release as pipeline artifacts. Every mapping failure in this article becomes harder to debug without these records. **Run a species sanity check on every new project.** A one-line `samtools idxstats` review at project start costs nothing and catches reference mixups, the single most common cause, before they contaminate a full analysis. **Map a subsample before committing compute.** Align 1 percent of reads, check the mapping rate and chromosome distribution, and only then launch the full job. **Keep a small contaminant panel.** A combined index with your organism plus common lab contaminants, for example the other species handled in your lab, allows a subsample-level contamination screen in minutes. **Report multimapping and filtering decisions.** State the aligner, the reference build, the MAPQ thresholds, and the multimapping policy in every methods section. Analyses that hide these choices cannot be audited when a wrong-genome artifact appears later. **Re-examine references before blaming the aligner.** The reference carries its own error burden from database annotation mistakes and assembly mis-assemblies [5], [4]. A locus that misbehaves across tools and libraries deserves a look at the reference sequence itself. ## Common Mistakes and Limitations - **Diagnosing from the mapping rate alone.** A healthy overall mapping rate says nothing about placement correctness. Reads can map at high rates to the wrong build or the wrong paralogs and still look fine in `flagstat`. - **Mixing naming conventions.** `chr1` and `1` break joins between BAM and annotation files even when the underlying assembly is identical. Normalize contig names once, at the reference level. - **Assuming the aligner is at fault.** Most wrong-genome outcomes trace to inputs: the wrong FASTA, the wrong index, the wrong annotation, or contaminated reads. The aligner is usually doing exactly what it was told. - **Discarding all multimappers reflexively.** In duplicated gene families this removes real signal and biases quantification. Choose the multimapping policy to match the question. - **Correcting contamination at the count-matrix level.** For single-cell combinatorial assays, aggregated corrections cannot recover read-level traceability, and read-level competitive mapping against multiple references is the stronger approach [2]. - **Trusting a draft assembly as ground truth.** De novo contigs can carry mis-assemblies, and detection tools remain imperfect, so validate before building conclusions on a draft reference [5]. - **Expecting quality scores to be perfect error predictors.** Quality values flag errors imperfectly, and window-level criteria outperform per-base judgments for some applications [8]. - **A limitation of this guide.** Every dataset differs in organism, assay, and reference availability, and individual troubleshooting decisions should be reviewed by someone with direct experience in your data type, ideally a core facility or senior analyst, before you commit to a re-analysis plan. ## Frequently Asked Questions ### Why are my reads mapping to the wrong genome when my pipeline ran without errors? Software errors are rare in this failure mode. The pipeline executed exactly as written, and the written instructions pointed at the wrong reference, the wrong build, or an index built from a different file. Verify the reference identity with `samtools idxstats` and the `@SQ` header before suspecting the aligner. ### How do I find out which genome build my BAM file uses? Read the `@SQ` lines in the BAM header with `samtools view -H` and compare sequence names and lengths against the chrom sizes of each candidate build. Sequence lengths differ between builds, so an exact length match identifies the build unambiguously. ### What is a normal multimapping read percentage? There is no universal number, because it depends on genome repetitivity, read length, and assay. What matters is the trend within your project. A multimapping rate that jumps between libraries from the same protocol, or a gene whose counts depend entirely on whether multimappers are kept, indicates a region where assignment is unreliable and needs an explicit policy. ### Can contamination make reads map to a different species? Yes. Foreign nucleic acid introduced through handling, reagents, index cross-talk, or, in single-cell assays, ambient material and doublets produces reads that align coherently to a foreign genome. Competitive mapping of reads against a multi-reference panel, with the ambient profile learned from empty and low-complexity barcodes, is the strongest read-level detection and correction approach [2]. ### Should RNA-seq reads be aligned to the genome or the transcriptome? Both placements have blind spots. Genome alignment misses reads crossing splice junctions unless the aligner is splice-aware, while transcriptome alignment cannot discover unannotated splice sites. Use a splice-aware genome aligner for discovery and add context-aware refinement when per-read ambiguity among candidate loci threatens correct gene assignment [3]. ### Why do my long reads align across a breakpoint but call the wrong variant? The reference does not carry the structural variant, so reads spanning the breakpoint have no single correct placement, and inversions are especially difficult because their anchors are nonlinear relative to collinear regions. Long-read mapping methods designed to detect inversions during alignment reduce these wrong alignments and variant calls [7]. ### Can errors in the reference database itself cause wrong mappings? Yes. Computationally annotated reference records have contained documented insertions, deletions, and base errors, and de novo short-read assemblies can contain mis-assembled contigs with wrongly arranged or deleted sequence. Cross-check suspicious regions against an independent assembly and validate splice boundaries before trusting a locus, [4]. ### Does read quality trimming improve mapping accuracy? Often yes. Error accumulation, especially at read ends, can make a read match a wrong locus better than its origin and is a recognized driver of wrong assignments when placement is resolved per read by fewest mismatches. Trim adapters and low-quality tails, and apply window-level quality criteria where your method depends on reliable sequence windows [8], [3]. ## Sources 1. [Pitfalls of genotyping microbial communities with rapidly growing genome collections.](https://pubmed.ncbi.nlm.nih.gov/36657438/) 2. [Empirical Estimation of Ambient Contamination in Combinatorial Single-Cell Methods Using Multi-Reference Mapping.](https://pubmed.ncbi.nlm.nih.gov/42779630/) 3. [A context-based approach to identify the most likely mapping for RNA-seq experiments.](https://pubmed.ncbi.nlm.nih.gov/22537048/) 4. [[Analysis, identification and correction of some errors of model refseqs appeared in NCBI Human Gene Database by in silico cloning and experimental verification of novel human genes].](https://pubmed.ncbi.nlm.nih.gov/15478601/) 5. [Identifying wrong assemblies in de novo short read primary sequence assembly contigs.](https://pubmed.ncbi.nlm.nih.gov/27581937/) 6. [High-throughput single nucleotide variant discovery in E14 mouse embryonic stem cells provides a new reference genome assembly.](https://pubmed.ncbi.nlm.nih.gov/25004115/) 7. [invMap: a sensitive mapping tool for long noisy reads with inversion structural variants.](https://pubmed.ncbi.nlm.nih.gov/38058196/) 8. [Evaluation of window cohabitation of DNA sequencing errors and lowest PHRED quality values.](https://pubmed.ncbi.nlm.nih.gov/15688315/)