Sanity checks, by assay

A PCA plot that separates by prep date instead of treatment is telling you the truth about your experiment, not a bug to plot around.

The pseudocount you pick for low-count genes can move a fold change more than the biology does, and DESeq2's own statistical model never wants a log at all.

Zero genes at padj < 0.05 with a clean PCA is a design or filtering problem to diagnose, not a cutoff to loosen.

Median-of-ratios, TMM, CPM and TPM answer different questions, and the wrong one fails quietly when your treatment shifts the whole transcriptome.

PCA can't tell you whether a stray point is a dead library or your best result; your QC metrics and metadata have to do that job for it.

A misshapen histogram is the cheapest bug report your DE test will ever hand you, and almost nobody opens it before trusting padj.

One minute with infer_experiment.py or salmon -l A tells you which strand flag to use, before a wrong guess quietly eats your counts.

TPM already threw away the count magnitude DESeq2 and edgeR need to estimate dispersion, and rounding it to dodge the integer check does not bring that information back.

When hemoglobin, immunoglobulin or a neighboring cell type's markers show up in every cluster, the soup is talking; here is how to prove it before you annotate anything.

Before you pick Harmony, RPCA or scVI, find out whether batch is separable from your condition at all, because no method fixes a confounded design.

A bridge cluster co-expressing two lineage markers is not a transitional cell state until you've ruled out that it's two cells sharing one droplet.

Harmony and CCA are built to erase batch, point them at a design where batch and disease are the same variable and they will erase your disease effect too.

A pseudocount of 1 versus 1e-9 can turn the same CD19 fold change from 1.24 into 5.64, know which number your tool used before you trust it.

A FindMarkers table full of RPL, RPS, MT- and FOS genes isn't a cell type. Here's how to catch it before the label ships.

Bonferroni zeroes out your gene list, Benjamini-Hochberg inflates it, and neither number means anything until you fix pseudoreplication first.

LogNormalize, SCTransform, CPM, TPM and dsb answer different questions; picking the wrong one quietly rewrites your clusters and your DE calls.

Louvain and Leiden will hand you as many clusters as your resolution allows; the only test that matters is whether each one has markers you can name and cells from more than one sample.

A p-value of 1e-200 from FindMarkers usually means you tested cells instead of donors, and pseudobulk is the fix a reviewer already knows to ask for.

The percent.mt < 5 you copied from a PBMC vignette is quietly deleting the tumor cells, cardiomyocytes, or plasma cells you were funded to study.

Distances between islands, island size and the direction of a smear on your UMAP are artifacts of the embedding, and here is how to test each claim before it reaches a figure legend.

Before you run Harmony on your frozen-tissue nuclei, find out whether the batch axis is technical noise or the condition you came to study.

A cluster that co-expresses two lineage markers isn't always a transitional cell state, and the fix depends on telling homotypic from heterotypic doublets before you delete anything.

FindMarkers will happily hand you a stress signature, an ambient RNA gradient, or a pseudoreplication artifact and let you name a cluster after it.

Leiden will happily hand you fifty clusters from thirty cell types if the resolution lets it; the object itself can't tell you which cuts are real.

5,000 nuclei from two donors is not 5,000 replicates: pick pseudobulk or a mixed model before you trust a single p-value out of FindMarkers.

Your snRNA-seq filter probably came from a whole-cell PBMC vignette, and in nuclei a high mito fraction means cytoplasmic contamination, not a dying cell.

A slide is not a replicate, learn to tell tissue-section artifact from real biology before you trust a single cluster.

The UMAP where every condition mixes perfectly isn't always good news, sometimes it means Harmony erased the exact difference you set out to measure.

FindMarkers and rank_genes_groups will happily hand you ribosomal genes and dissociation stress as your top "cell type marker", here is the checklist that catches it before it reaches a figure legend.

Library-size normalization removes the exact tissue-architecture signal you're trying to map, and the scRNA-seq default you already know is often the wrong call here.

Leiden and Louvain will happily cut your tissue into as many pieces as the resolution allows; the algorithm has no opinion on whether any of those pieces is a real cell type.

A wall of p-values near zero on pooled spots or cells almost always means you counted the wrong thing as your sample size.

Copying the 10% mito cutoff from a PBMC tutorial into your tumor Visium slide will quietly delete every high-mitochondrial tumor spot you came to study.

UMAP islands, cluster sizes and arcs are optical illusions in spatial data; the tissue coordinates and a permutation test are the only evidence that counts.

Zero significant peaks after BH correction is usually the statistics doing exactly what you asked, not a broken pipeline.

The method you pick can turn 33 significant regions into 24,450 on the same data, so decide it on purpose and check it with an MA plot.

Your differential accessibility table is only as good as the shape of its p-values, and one hist() call tells you whether to trust it before you look at a single peak.

No input control exists for ATAC-seq, so ENCODE blacklists, FRiP, and a five-minute browser check of your top peaks are what stand between you and peaks called in centromeres.

Before you pay for another lane, find out whether your ATAC-seq libraries are out of depth or out of complexity, and whether a third replicate would help more.

Your LSI plot separates by processing date instead of cell type, here's the order of checks that tells you whether that's a fixable artifact or a confound no tool can undo.

A cluster that looks like a transitional cell state is often two nuclei in one droplet, and scATAC-seq doublets need fragment-based tools, not the scRNA-seq playbook.

A marker with a huge fold-change can still be sequencing depth or dissociation stress in disguise, here's how to tell before you write the label on the UMAP.

Louvain and Leiden will keep splitting your point cloud for as long as you keep raising resolution, here's how to find out whether the clusters you kept are real.

Your top-scoring peaks might be centromeric repeats, not regulatory elements, and the pooled peak set your pipeline handed you is quietly erasing your rarest cell type.

Five thousand cells from two donors are not five thousand replicates, and your p-values are lying to you about it.

The QC axes that matter in scATAC-seq aren't the ones your scRNA-seq muscle memory reaches for, and a copy-pasted vignette cutoff will quietly delete a real cell population.

The islands, the gaps between them, and the smear you're calling a trajectory are navigation aids, not measurements, so treat them that way before you write the discussion section.

RPKM, CPM, and median-of-ratios all assume nothing changed genome-wide, an assumption that quietly kills real histone-mark shifts before you ever see them.

Flat with a spike, a hill, a U, or a pile-up near 1: each shape of your differential binding p-values points at a specific broken assumption, and you can see it in one line of R.

A matched input control and the ENCODE blacklist are the two things standing between your peak list and a stack of copy-number and repeat artifacts you'd otherwise write up as binding sites.

Your treated sample has a fraction of the control's histone mark, and your bigWigs look identical: the normalization you picked erased the biology.

MACS2 and an IgG control were built for ChIP-seq's noise floor; point them at CUT&RUN's much cleaner background and you'll either call nothing or call the centromere.

Your BED file and your VCF look fine and bedtools says they share nothing: the cause is usually one prefix, and the repair is a text edit, not a re-run.

A low mapping rate isn't "bad sequencing" until you've checked what the unmapped reads actually are.

A BED file from the wrong build gives you confident, wrong overlaps, and the header check that catches it takes thirty seconds.

X heterozygosity, chrY coverage and pairwise genotype concordance take minutes to compute and catch mislabels that no alignment or variant-quality QC report will ever flag.

Mean coverage looks fine, the VCF looks clean, and you still cannot say whether the missing variants are absent or just unsampled.

A VCF position and a BED start are not the same number, and the mismatch silently drops the first base of every target interval.