# Common Mistakes in Genomics Data Analysis and How to Avoid Them Most genomics data analysis mistakes begin before a single statistical test runs. The most common cause is a problem with the data itself, a swapped sample, a coding error in the sample sheet, or a quality problem that nobody checked after [preprocessing](/blog/data-analysis/what-is-data-preprocessing). The first thing to check is therefore not your model or your plots. Confirm that every file belongs to the sample the spreadsheet says it belongs to, and confirm that quality control actually ran at each step rather than only at the end. The pattern is well documented. Reviews of microarray practice have found that checking arrays after normalization rarely occurs, even though unreliable data can seriously distort every downstream result [1]. Studies that audited sample identity in large cohorts keep finding the same thing, a small number of swaps and coding errors hiding inside data everyone assumed was clean [2]. This article walks through the most common genomics data analysis mistakes from the most frequent to the rarest. For each one it explains the mechanism behind the failure, how to confirm it, and how to fix it. A troubleshooting table collects the patterns in one place, a diagnostic sequence gives you an ordered checklist, and the closing sections cover prevention habits. The examples span bulk [RNA sequencing](/blog/guides/rna-sequencing), whole genome and [exome sequencing](/blog/guides/exome-sequencing), single cell assays, microarrays, genome assembly, pangenomics, and multi-omics integration. Every technical term is defined where it first appears. ## Start With the Data, Not the Model ### Check sample identity first ![Decision path for confirming a sample swap, from sex mismatch through replicate concordance to re-extraction and database correction.](/article-images/illustrations/genomics-data-analysis-mistakes-swap-triage.webp) Sample swaps are the classic silent failure. The analysis runs, the p-values appear, and nothing in the output hints that two columns of data belong to the wrong individuals. Genetic identity checks catch this. The amelogenin test, a genetic sex check frequently used in forensics, amplifies a region of the amelogenin gene that differs between the X and Y chromosomes, which makes it a fast way to compare genetic sex against recorded sex. In one biomedical study, researchers ran this test on 1,224 participants and found overall concordance of 99.84 percent, meaning 1,222 samples matched the database. The two discordant samples each hid a different error, one was a codification error in the database and the other was a sample labeling error that only came to light after a new [DNA extraction](/blog/guides/dna-extraction) [2]. Practical identity checks include comparing genetic sex with recorded sex, checking genotype concordance between technical replicates, verifying known control samples, and confirming that duplicate samples from the same individual actually match. Run these checks before you spend compute on downstream analysis, because every result that follows inherits any swap you miss. Safety note: samples of human or animal origin carry biohazard risk. Wear gloves and eye protection, handle extractions in a designated area, and follow your institution's biosafety procedures. ### Check the metadata Metadata is the descriptive information attached to each sample, diagnosis, age, tissue type, collection date, sequencing lane, and library preparation batch. Metadata mistakes are as damaging as sample swaps because they mislabel what you think you are testing. Common causes include unit mismatches, inconsistent coding of categorical variables, merged spreadsheets with shifted rows, and dates recorded in ambiguous formats. The fix is boring and effective, a written protocol for collecting and entering phenotypic data. The African Goat Improvement Network developed its image collection protocol iteratively over three years, across 12 countries and more than 12,000 images, assessing the quality of the resulting images at each round so the data could actually support analysis [3]. You do not need that scale to copy the principle. Write the protocol, test it on a handful of samples, and revise it before the real data arrive. ### Check file integrity Before analysis, confirm that files transferred completely and match their records. Compare file sizes and checksums against the source, confirm read counts or array positions against expected values, and archive the raw data in read-only storage. A truncated FASTQ file, the text format that stores sequencing reads together with per-base quality scores, can fail silently in some pipelines and loudly in others. Either way, you want to catch it on day one. ## Quality Control Mistakes That Corrupt Everything Downstream ### Skipping post-normalization checks Normalization adjusts raw signal intensities so that arrays or sequencing libraries become comparable. Many analysts treat normalization as the last quality control step and never look at the data again. That is backwards, because normalization itself can fail, and the failures are visible if you look. A review of post-normalization checking describes four problem types that these checks catch, clerical mistakes, array-wide hybridization problems, problems with normalization itself, and mishandling problems [1]. Three techniques cover most cases. Dimension reduction methods such as [principal component analysis](/blog/data-analysis/principal-component-analysis-explained), or PCA, which compresses thousands of variables into a few axes of greatest variation, reveal outliers and structure you did not expect. False array plots image each array's residuals so that spatial artifacts become visible. Correlograms, which display correlations between arrays as a matrix, expose arrays that behave unlike their group. All three run quickly in the R statistical package, and once you find a problem you can either rectify it or exclude the offending sample [1]. ### Ignoring input quality Downstream statistics cannot repair poor starting material. For microarray work, the success of differential expression analysis depends on the quality and integrity of the starting RNA, and even RNA of reasonable quality can yield unreliable hybridization if the labeling efficiency was poor [4]. A microfluidic assay on the Agilent 2100 Bioanalyzer can assess the quality of directly labeled fluorescent cDNA before hybridization, and researchers validated it by degrading RNA with varying amounts of RNase to simulate damage [4]. The lesson generalizes beyond arrays. Assess the actual molecule you are about to run, after labeling or library preparation, not just the original extract. That single habit prevents the costly mistake of hybridizing a poor product to an expensive run. Safety note: fluorescent labeling reagents and nucleic acid stains can be toxic or mutagenic. Handle them with gloves and eye protection and dispose of them under your institution's chemical waste rules. ### Single cell input mistakes Single cell RNA sequencing, which measures [gene expression](/blog/guides/gene-expression) in thousands of individual cells at once, adds its own input traps. Guidance on the [10x Genomics](/blog/guides/10x-genomics) Chromium Single Cell Gene Expression assay stresses that data quality links directly to the quality of the initial cell or nuclei suspension, which should be fully dissociated, extremely clean, and highly viable [5]. Clumps, dead cells, and debris propagate into the data as doublets, meaning two cells captured and read as one, and as ambient RNA, stray transcript from lysed cells that clouds every droplet. The manufacturer's user guide documents the common mistakes that occur while running the assay, and reading it before the experiment costs nothing compared with repeating it. ### QC metrics worth actually reading For sequencing data, look at per-base quality scores, duplication rates, GC content bias, which is the tendency of reads to cluster at certain proportions of the bases G and C, and coverage evenness, meaning how uniformly reads distribute across the genome. For arrays, look at signal intensity distributions, background levels, and control probes. Every one of these has a normal range for your platform, and the manufacturer documentation states it. A metric outside that range is a question to answer before analysis, not a number to ignore. ## Statistical Mistakes That Manufacture False Positives ### Misspecified main effects in interaction tests Interaction analysis asks whether the effect of one factor depends on another. The standard model includes two main effect terms plus their product, and that model is everywhere in the statistical and epidemiological literature. It fails in a specific, well characterized way. When the main effects are misspecified, for example modeled as linear terms when the true relationship is curved, tests for interaction can suffer severe inflation of the type I error rate, the probability of declaring an interaction that does not exist [6]. The same work tested remedies. For a quantitative outcome with two independent factors, using the sandwich variance estimator, a robust method for computing standard errors, kept both the Wald and score tests at the correct type I error rate. That protection disappeared when the factors were not independent or when the outcome was binary. Flexibly modeling the main effect with a generalized additive model, which fits smooth nonlinear terms instead of forcing linearity, largely reduced or often removed the bias and maintained the correct type I error rate for both quantitative and binary outcomes regardless of the independence assumption [6]. The practical fix is to model main effects flexibly and to treat a surprising interaction as a hypothesis to confirm, not a discovery to report. ### Enrichment analysis pitfalls Functional enrichment analysis, or FEA, takes a list of genes and summarizes which biological pathways it over-represents. It is among the most used techniques in bioinformatics, with about 10,000 scientific articles describing its use each year, and mistakes creep in easily through poor tool design and user unawareness [7]. A recent review catalogued ten mistakes that undermine enrichment analyses in published research and provided practical advice on mitigating each one [7]. The recurring themes are worth internalizing as general practice. The background gene set must match the genes that had a real chance of appearing in your list, not the entire genome. Redundant pathway terms inflate a results section without adding information. Heavily thresholded gene lists discard weak but coherent signals before the test ever runs. When an enrichment result looks implausibly broad or implausibly specific, suspect the inputs before you believe the biology. ### Ignoring multiplicity and structure Genomics tests thousands of hypotheses at once, so unadjusted p-values guarantee false discoveries. Control the false discovery rate, the expected proportion of false positives among your declared hits, and state which method you used. Population structure and batch effects create associations with no biological basis, and exploratory plots colored by batch, date, and ancestry variables will expose most of them before they reach your results section. ## Variant Calling and Sequence Analysis Mistakes ### Trusting one caller and default thresholds Calling variants means deciding where your sequence data differs from the reference genome, and it looks more like a solved problem than it is. Detection of minority variants, variants present in only a small fraction of the reads, is complicated by errors introduced during both sample preparation and data analysis [8]. A study using synthetic RNA controls and simulated data tested seven variant calling tools across a range of allele frequencies, the fraction of reads carrying the variant, and simulated coverages, the average number of reads spanning each position. The choice of caller and the use of replicate sequencing had the most significant impact on single nucleotide variant discovery, and both allele frequency and coverage thresholds moved false discovery and false negative rates [8]. When replicates are not available, the recommended approach is a combination of multiple callers with more stringent cutoffs [8]. Defaults are a starting point, not a decision. Record every threshold you set and report it, because a threshold nobody can see is a threshold nobody can audit. ### Mixing up genome builds and references Genomic coordinates are only meaningful relative to a specific reference build, and comparing coordinates across builds scrambles results. Record the reference genome and annotation release for every analysis, and when you must convert between builds, use a dedicated lift-over step and verify a sample of positions afterward. Online resources reduce the risk of comparing the wrong things. Ensembl provides a structured route to gene sequences, exon and intron structure, and conserved noncoding regions for an orthologous gene, while the [UCSC genome browser](/knowledge/bioinformatics/ucsc-genome-browser-beginner-tutorial-and-tips) supports sequence homology searches to locate orthologs [9]. Orthologs are genes in different species that descend from a common ancestral gene, and they are the unit of comparison in cross-species genomics. ### Mitochondrial pseudogenes masquerading as heteroplasmy Heteroplasmy means more than one [mitochondrial genome](/blog/guides/mitochondrial-genome) variant coexists within an individual, and it matters enormously for mitochondrial disease research. Nuclear mitochondrial pseudogenes, called numts, are fragments of [mitochondrial DNA](/blog/guides/mitochondrial-dna) that integrated into the nuclear genome over evolutionary time, and they are a classic source of false heteroplasmy calls. The scale of the problem is real. Researchers cloned and fully sequenced 46 paralogous nuclear fragments representing the entire mitochondrial genome, amplified with primers designed for the mitochondrial genome itself, and noted that earlier BLAST sequence similarity searches had suggested the nuclear genome harbors between 400 and 1,500 such copies [10]. Multiple numts amplified simultaneously with the mitochondrial target and increased the pseudogene signal in the reactions, and the common assumption that mitochondrial template swamps numt amplification depends heavily on which region of the mitochondrial genome you target [10]. Authenticate mitochondrial biomarkers against paralogous nuclear sequences before trusting a heteroplasmy call, and check every primer pair against the nuclear genome. ### Ignoring your sequencing technology's error profile Every platform has characteristic errors. Oxford Nanopore Technology long reads, for example, carry relatively low accuracy and sequence mistakes driven by high GC content and sequence repetitions, conditions frequently found in bacteria that carry biosynthetic gene clusters [11]. A biosynthetic gene cluster, or BGC, is a group of genes that together produce a bioactive natural product such as an antibiotic or anti-tumor compound. Despite those error modes, a pipeline combining ONT long reads with antiSMASH software for BGC identification succeeded in identifying and cloning a previously unknown BGC from a previously unsequenced myxobacterium [11]. The point is not that one technology is better than another. The point is to match the technology to the question, understand its error modes, and validate the calls that matter. ## Assembly and Pangenomics Mistakes ### Treating a draft assembly as a finished genome An assembly is a model of a genome, and models can be wrong in structured ways. Plant mitochondrial genomes illustrate the trap. The Stephania japonica mitochondrial genome, assembled from a combination of second and third generation sequencing data, is 555,117 base pairs and forms a complex polycyclic structure, meaning multiple interlocked circular molecules, rather than one clean circle [12]. An analyst who assumed a single circular chromosome would misrepresent the genome from the first plot onward. Combine data types when the genome is repetitive or unusual, inspect the assembly structure directly, and validate it rather than assuming it. ### Pangenomics shortcuts Pangenomics studies the union of all genomes in a clade, a group of organisms descended from a common ancestor. Its analyses have become easy to run, and that accessibility increases the chances of mistakes and inflated or misleading results, especially among beginners [13]. A published set of seven quick tips for gene-focused computational pangenomic analysis describes common mistakes and best practices, with a focus on bacterial pangenomics [13]. The core habits are to assess assembly and annotation quality before computing gene presence and absence, to be explicit about how draft genomes are handled, and to remember that annotation errors masquerade as biology. ### Mapping-by-sequencing without a plan Sequencing a pooled population of recombinants can locate a causal point mutation quickly, an approach called mapping-by-sequencing. Reviews of the strategy note that its wider use is often hindered by a lack of knowledge about how to manipulate and process the resulting data, and they compile case studies in plants and model species including Caenorhabditis elegans, Saccharomyces cerevisiae, and Drosophila melanogaster [14]. Decide the cross design, the pooling strategy, and the mutation filtering scheme before you sequence, because those choices determine which mutations you can possibly see. ## Multi-omics Integration Mistakes Integration combines data from proteomics, the study of proteins, metabolomics, the study of metabolites, metagenomics, the study of microbial communities, phenomics, the study of physical traits, [transcriptomics](/knowledge/bioinformatics/modern-transcriptomics-bulk-single-cell-spatial), the study of RNA, and epigenomics, the study of chemical marks on DNA. Because these data types are deeply heterogeneous, mistakes during the integration phases are common enough that a dedicated set of ten quick tips exists, drawn from errors the authors experienced or noticed in published studies [15]. Three mistakes account for a large share of the damage. First, sample identifiers that do not match across modalities, which silently pairs the wrong measurements. Second, normalizing all modalities by the same rule, which lets the modality with the largest numeric range dominate the result. Third, building a model on all the data and reporting its performance on that same data, which measures memory rather than signal. Harmonize identifiers across every dataset, normalize each modality on its own terms, and hold out data for validation before you trust an integrated result. ## Benchmarking and Simulation Mistakes ### Adopting tools without testing them Pipelines for sequencing data involve many technical and methodological decisions, choices about coverage, assembly, mapping, variant calling, and phasing, which means assigning variants to the parental chromosomes they came from. Those decisions can influence the final biological result, including the evolutionary tree in a phylogenomic analysis [16]. Simulation makes those choices testable. NGSphy simulates Illumina reads and read counts from haploid and diploid genomes across thousands of gene families, with options that model [sequencing coverage](/knowledge/molecular-biology/sequencing-coverage) heterogeneity across species, individuals, and loci, including off-target or uncaptured loci [16]. Before you adopt a pipeline, simulate data whose truth you know and check whether the pipeline recovers it. ### Benchmarking with unrealistic data Tool performance can be undervalued or overvalued when benchmarking datasets are too small, a problem documented for [single cell DNA sequencing](/knowledge/bioinformatics/single-cell-dna-sequencing-applications-and-workflow-considerations) tools [17]. SCSilicon addresses this by generating synthetic single cell DNA reads carrying controlled single nucleotide variants, small insertions and deletions, and copy number aberrations, together with the ground truth of segmentation breakpoints and subclone labels [17]. A sanity check of single cell copy number callers using such synthetic data found SCYN to be the most robust of the tools tested [17]. A benchmark without ground truth or realistic error is decoration, not evidence. ## Genomics Data Analysis Mistakes and How to Fix Them: The Troubleshooting Table Run down the left column until you find your symptom, then confirm the cause before applying the fix. Most of these genomics data analysis mistakes have a cheap confirmation step and an expensive one, and you should always run the cheap one first. | Symptom | Likely cause | How to confirm | Fix | |--|--|--|--| | Samples cluster by date, lane, or operator instead of biology | Batch effect or metadata error | Color a PCA plot by batch variables and test associations between principal components and metadata | Include batch in the model, balance the design, correct the metadata | | Recorded sex disagrees with genetic sex | Sample swap or database coding error | Sex marker check such as amelogenin, X and Y coverage ratios, replicate concordance | Re-extract and re-genotype the sample, correct the database record [2] | | One sample is an extreme outlier on every metric | Poor input quality or handling failure | Per-sample QC metrics and post-normalization residual plots | Exclude the sample or re-run it from freshly prepared material [1] | | Enriched pathways are all generic, metabolism and signaling everywhere | Background set mismatch or redundant terms | Check that the background matches the queried gene list and measure term overlap | Use a matched background, collapse redundant terms, rerun [7] | | Significant interaction with an implausible effect size | Misspecified main effects inflating type I error | Refit the model with flexible main effects and compare the estimates | Model main effects with a generalized additive model [6] | | Variant appears in one caller but not others | Caller-specific bias or threshold artifacts | Compare genotypes across callers and inspect the read alignments | Combine multiple callers with stringent cutoffs, use replicates when possible [8] | | Apparent heteroplasmy at many unrelated sites | Numt co-amplification or paralogous mapping | Search primers and amplicons against the nuclear genome with a similarity tool | Redesign primers and filter reads that map to pseudogenes [10] | | Assembly is fragmented with low completeness | Coverage gaps, repeats, or unsuitable technology | Check assembly metrics and the coverage distribution | Combine long and short reads and resequence low coverage regions [12] | | Single cell clusters track quality metrics, not cell type | Doublets, ambient RNA, or dying cells | Plot per-cell QC distributions and check marker gene expression | Filter low quality cells, model ambient RNA, remove doublets [5] | | Integrated model performance depends on one modality | Scale dominance or identifier mismatch | Measure each modality's contribution and verify sample pairing | Normalize per modality and validate on held-out data [15] | ## A Diagnostic Sequence You Can Follow When a result looks wrong, work through these steps in order. 1. Verify the sample sheet against the data files, one row at a time for any sample involved in the anomaly. 2. Run identity checks, genetic sex against recorded sex, replicate concordance, and control samples. 3. Inspect raw data quality before any preprocessing, read quality scores for sequencing and signal distributions for arrays. 4. Inspect the data after every transformation, not only at the end of the pipeline. 5. Color exploratory plots by batch, date, operator, and every other metadata variable you have. 6. Reconcile units, counts, and scales across samples and modalities. 7. Check statistical assumptions, including main effect specification, independence, and the multiple testing method. 8. Run a sensitivity analysis with an alternative tool, threshold, or model and see whether the finding survives. 9. Validate the key finding with an independent method, an independent dataset, or the bench. ## Common Mistakes and Limitations Beyond the specific failures above, several habits cause repeated trouble. - Trusting software defaults. Defaults are generic, your data is not. Record and justify every parameter. - Running QC only at the end. Errors compound, so check after every step. - Confusing [statistical significance](/blog/guides/statistical-significance) with biological importance. A tiny effect can be significant with enough samples, and a large effect can miss significance in a small one. - Re-analyzing poor input and expecting different output. Re-analysis cannot repair a degraded sample, a failed labeling reaction, or a swapped tube. Fix the data or collect new data. - Overfitting and data leakage. If a model is evaluated on data that shaped its construction, its performance is a memory test, not a prediction. - Ignoring population structure. Ancestry and relatedness create correlations that look like associations. - Treating draft assemblies and annotations as ground truth. They are estimates with error, and their errors become your errors. - Skipping controls. Negative and positive controls are how you know the pipeline worked at all. Limitations. This article covers computational methods, not clinical interpretation. Genomic findings that touch patient or animal care need review by a qualified professional, and animal cases need a veterinarian with the relevant expertise. The translation challenge is real, because researchers working in systems medicine have documented the tension between individualized algorithmic recommendations and evidence-based clinical decision making [18]. Tool versions also change quickly, so verify current behavior against official documentation before relying on any specific package. ## Prevention Practices That Stop Mistakes Before They Start The solutions below are habits rather than one-time fixes, and together they remove most of the causes described in this article. - Write and validate the sample sheet before the experiment begins, and treat it as part of the experimental design. - Build QC checkpoints into the pipeline after every step, with explicit pass and fail criteria. - Keep raw data read-only and analyze copies, so every result can be traced back to untouched source files. - Track code, software versions, and parameters in version control, and use a workflow manager so the pipeline runs the same way every time. - Benchmark new tools with simulated data or spike-in controls before trusting them on real data. - Validate computational predictions experimentally. One integrative strategy that combined coexpression, functional enrichment, cis-regulatory element annotation, meaning annotation of DNA regions that control nearby genes, and [comparative genomics](/blog/guides/comparative-genomics) between plants and humans nominated 339 putative conserved proliferation genes, and experimental analysis confirmed that 20 of 40 selected genes affected plant [cell proliferation](/blog/guides/cell-proliferation), with conserved roles corroborated for two human orthologues [19]. Computational prediction is where the work starts, not where it ends. - Learn bioinformatics in a structured way. The barrier that most often hinders new sequencing users is knowledge of how to manipulate and process the data [14]. A [bioinformatics course](/blog/careers/bioinformatics-courses-and-certifications-a-practical-guide) that makes you run quality control, alignment, and statistics on real datasets will teach more than passive reading, and practicing on public datasets lets you make mistakes cheaply. ## Frequently Asked Questions ### What are the most common genomics data analysis mistakes? Sample swaps and metadata errors, skipped or late quality control, statistical misspecification such as misspecified main effects, and blind trust in a single tool with default thresholds. Most are caught by verifying sample identity and running QC after every processing step. ### How do I check whether my samples were swapped? Compare genetic sex against recorded sex, check genotype concordance between replicates and known controls, and re-extract any sample that fails. Sex marker tests such as the amelogenin test exist precisely for this purpose. ### Why does my enrichment analysis return only generic pathways? Usually the background gene set does not match the genes that had a chance of appearing in your list, or redundant terms are crowding out specific ones. Fix the background, collapse redundant terms, and rerun before interpreting the biology. ### Do I need technical replicates for variant calling? Replicates are among the most impactful factors for reliable variant discovery. When replicates are impossible, combine multiple callers with more stringent cutoffs and report every threshold you used. ### Can I rescue a dataset with poor input quality by re-analyzing it? No. Re-analysis cannot restore degraded RNA, a failed labeling reaction, or a contaminated cell suspension. Confirm the problem with QC, then re-prepare the sample. ### How do I learn bioinformatics and avoid these mistakes? Take a structured bioinformatics course that has you run complete workflows on real datasets, read the official documentation for every tool you use, and practice on public data where mistakes cost nothing. ### What should I report so others can trust my analysis? Report QC metrics at each step, software versions and parameters, the reference build, the [statistical methods](/blog/guides/statistical-methods), and your code or workflow. Reproducibility is the strongest signal of trustworthiness. ### When should I simulate data instead of using real data? Simulate when you are choosing among tools or pipeline settings, because simulated data comes with known truth. Use real data afterward to confirm performance under conditions simulation cannot fully capture. ## Sources 1. [Post-normalization quality assessment visualization of microarray data.](https://pubmed.ncbi.nlm.nih.gov/18629006/) 2. [Amelogenin test: From forensics to quality control in clinical and biochemical genomics.](https://pubmed.ncbi.nlm.nih.gov/17716640/) 3. [Using the community-based breeding program (CBBP) model as a collaborative platform to develop the African Goat Improvement Network-Image collection protocol (AGIN-ICP) with mobile technology for data collection and management of livestock phenotypes.](https://pubmed.ncbi.nlm.nih.gov/37745840/) 4. [A qualitative assessment of direct-labeled cDNA products prior to microarray analysis.](https://pubmed.ncbi.nlm.nih.gov/15762992/) 5. [Guidance on Processing the 10x Genomics Single Cell Gene Expression Assay.](https://pubmed.ncbi.nlm.nih.gov/36495443/) 6. [Interaction analysis under misspecification of main effects: Some common mistakes and simple solutions.](https://pubmed.ncbi.nlm.nih.gov/32101638/) 7. [Ten common mistakes that could ruin your enrichment analysis.](https://pubmed.ncbi.nlm.nih.gov/42018548/) 8. [Optimized quantification of intra-host viral diversity in SARS-CoV-2 and influenza virus sequence data.](https://pubmed.ncbi.nlm.nih.gov/37389439/) 9. [Comparative bioinformatics for mouse and human genes: getting started.](https://pubmed.ncbi.nlm.nih.gov/18428352/) 10. [The pseudo-mitochondrial genome influences mistakes in heteroplasmy interpretation.](https://pubmed.ncbi.nlm.nih.gov/16859552/) 11. [Enabling Access to Novel Bacterial Biosynthetic Potential From ONT Draft Genomic Data.](https://pubmed.ncbi.nlm.nih.gov/40034067/) 12. [Assembly and analysis of stephania japonica mitochondrial genome provides new insights into its identification and energy metabolism.](https://pubmed.ncbi.nlm.nih.gov/39994544/) 13. [Seven quick tips for gene-focused computational pangenomic analysis.](https://pubmed.ncbi.nlm.nih.gov/39227987/) 14. [Getting started in mapping-by-sequencing.](https://pubmed.ncbi.nlm.nih.gov/25359627/) 15. [Ten quick tips for avoiding pitfalls in multi-omics data integration analyses.](https://pubmed.ncbi.nlm.nih.gov/37410704/) 16. [NGSphy: phylogenomic simulation of next-generation sequencing data.](https://pubmed.ncbi.nlm.nih.gov/29534152/) 17. [SCSilicon: a tool for synthetic single-cell DNA sequencing data generation.](https://pubmed.ncbi.nlm.nih.gov/35546390/) 18. [Clinical decision-making and secondary findings in systems medicine.](https://pubmed.ncbi.nlm.nih.gov/27209083/) 19. [Identification of putative cancer genes through data integration and comparative genomics between plants and humans.](https://pubmed.ncbi.nlm.nih.gov/22218400/)