Sample identity mix-ups and plate maps

Column six was labelled as the treated animals and contained the controls. The libraries clustered, the base qualities looked fine, and every plot downstream was a fiction. Sample identity fails before chemistry does. A swapped column, a mirrored plate, or an index pasted onto the wrong name will spend a sequencing run with complete confidence. The library mechanics are in next-generation sequencing from library to reads. The pre-run list that includes identity is in what to check before a sequencing run. This page is how to choose the identity checks that match the way your plate is actually handled.

The molecule, the index, and the name are three objects

A sequencing sample has at least three labels, and laboratories talk as if they were one. The molecule is the DNA in the well. The index is the synthetic barcode the instrument will read and use to bin the reads. The sample name is the human phrase in the sheet: mouse 14, donor B, plasmid prep 3. Demultiplexing trusts the index. The scientific conclusion trusts the name. Nothing in the base caller checks that those two still point at the molecule you froze.

A plate map is the table that claims to hold the relationship. Rows and columns, index sequences, names, and sometimes a second barcode printed on the tube. If the map is wrong, the FASTQ files will be neatly named and biologically swapped. Read quality will not save you. High-quality reads of the wrong genome are still the wrong genome.

A1 is a convention, and conventions disagree

On a standard 96-well plate, rows are lettered A to H and columns are numbered 1 to 12. A great many people define A1 as the top-left well when the plate's labelled edge faces them. That is a convention, not a law of geometry. A handler can rotate a plate. A spreadsheet can be filled column-wise while the map is read row-wise. A photograph of a plate can be flipped when it is pasted into a slide. Some software exports well positions starting at a different corner once a plate is turned to match a nest.

The failure mode that ruins a study fastest is a shifted or swapped column. Eight samples move together. Treated and control trade places. Technical replicates stop being replicates. Because the indexes move with the wells or stay on the sheet, you can get either a clean swap or a scramble, depending on whether the DNA moved or only the names moved.

Before any library reagents are opened, pick a well that is easy to check, often A1 and H12, and confirm that the tube, the map, and the index sheet name the same thing. If they do not, stop. Fixing a map after the pool is made is archaeology.

Barcodes on plastic versus barcodes in the read

Tube barcodes and plate barcodes are identity in the freezer. Index sequences are identity in the run. They solve different swaps. A scanned cryovial stops you aliquoting donor C into a tube you will later call donor D, if the scanner is used at the moment of transfer and the result is stored. It does nothing once you have pipetted into a well and thrown the map away.

Write the index sequence, not only the kit's index name. Kit names get reused across chemistries, and a name without the bases cannot be checked for a collision. How those collisions behave once the pool is loaded is the subject of index collisions in multiplexed pools. The identity problem here is earlier: the right index on the wrong DNA.

Chain of custody is the list of hands. Who pulled the plate, who thawed it, who added the indexes, who resealed it. A freezer log that says the plate moved, without saying that the map was checked afterwards, is a travel diary. The useful record says the map matched the tubes at the moment the indexes were assigned.

Sex balance and a variant fingerprint are concepts, not decorations

When the organism has sex chromosomes, the proportion of reads that land on them is a coarse identity check. A sample recorded as male that lacks the expected chromosome, or the reverse, is either a swap, a sexing error in the metadata, or a genuine biological exception. You do not silently edit the metadata to match the reads. You quarantine the well and the neighbours that were handled with it.

A variant fingerprint is tighter. A small set of common positions, genotyped once and stored, can be compared with the same positions in the new reads. Agreement supports identity. Disagreement means this is not that earlier sample. Samples that have never been genotyped have no fingerprint until you decide to create one and keep it. The check is only as honest as the archive.

Neither check notices a swap between two samples that match on those markers. Use them as alarms, then go back to the map.

Branch when a check fails

If A1 does not match, do not load the plate and do not "fix it in the sample sheet" from memory. Re-read the physical plate. If the sex check fails after sequencing, hold the whole batch that shared the map, not only the odd sample. A column swap predicts that the neighbour is wrong too. If the fingerprint fails, compare the failed sample with the other fingerprints in the same run. The missing identity is often sitting in another FASTQ, looking perfectly analysed.

A check that you cannot run, because no prior genotype exists and the species has no useful sex marker, means the map and the custody log are your only evidence. Invest in them before the run, not after a reviewer asks.

A clean demultiplex is not an identity result

The instrument will bin reads by the index it sees. If the sheet is internally consistent and biologically wrong, the undetermined fraction stays low and the quality reports look healthy. People then trust the names. The failure shows up later, as a treatment effect in the wrong direction, or as a replicate that does not correlate with itself. Those are expensive places to discover a rotated plate.

Index hopping and mismatched sheets are real, and they are a different chapter. Do not use them as the explanation for a sex-check failure until you have looked at the map. The dull explanation, a column moved during a thaw, is common.

Governance of samples is an institutional decision

A mix-up in a human research cohort can attach one participant's genome to another participant's consent. That is a governance incident, not a bioinformatics quirk. Animal and plant studies have their own rules about what a wrong label does to a claim. Biosafety classification of the organism does not change because the sequencer is in a different room. This page does not approve a study, a consent form, or a containment level. The biobanking ideas in ISO 20387 are a public place to see how identity and custody are discussed as requirements. They are not a certificate for your freezer.

Humidity, smeared labels, and a map that lives on one laptop

Monsoon humidity and ordinary condensation loosen adhesive labels and blur ink. A plate that comes out of a frost-caked freezer with unreadable rims has already lost the physical half of its identity. Store the durable record the way you store other biological inventories, as described in storing biological samples from fridge to freezer, and keep the plate map somewhere other than a single personal computer. A photograph of a whiteboard is not a map you can diff against the index sheet. When a second site will index the plate, send the file, and ask them to confirm A1 in writing before they proceed. Methods repositories such as protocols.io are where groups write those handling steps down. Copy the decision, not someone else's volumes.

What to name about identity when you enquire

Say how samples are identified today: tube barcode, plate barcode, index kit, or a handwritten map. Say whether a prior genotype or a sex metadata field exists. Say who is allowed to relabel a well. Library plastics and index reagents are catalogue classes in the genomics and sequencing catalogue. The surrounding research use is the genomics research pathway.

If the design is a whole genome, the whole-genome sequencing enquiry reference is where that design can be discussed. If the identity check itself is a short amplicon you want read as a trace, the Sanger DNA sequencing enquiry reference is the matching conversation. Put the map convention and the fail rule in the quote request. A provider can demultiplex only the sheet you send. They cannot reconstruct a column you swapped last winter.