Thalassemia caller

Sign in to continue.

Your password must be changed before you can continue.

Reset your password

Enter your user name or email address. If the account exists, a link to set a new password is sent to it.

Two-factor code

Open your authenticator app and enter the six-digit code for this account.

Lost your phone? Enter one of your recovery codes instead. Each works once.

Set your password

At least 12 characters. A phrase you can remember beats a short string of punctuation you cannot.

Thalassemia caller alpha & beta · nanopore amplicon · hg38 · build 379a6cec9e02

Runs

1 · Choose samples

Drop sample folders here one folder per sample · or loose .fastq.gz files
On the sequencer fastq_pass barcode81 FBC38010_pass_0.fastq.gz FBC38010_pass_1.fastq.gz barcode82 FBC38010_pass_0.fastq.gz FBC38010_pass_1.fastq.gz and so on drag Into the drop zone Drop the whole folder One run per barcode barcode81 uploads, then runs barcode82 uploads, then runs barcode83 uploads, then runs
  1. Pick the barcode folders. From the sequencing output, use fastq_pass — drop the whole folder in and every barcode inside it becomes a sample, or drop individual barcodeNN folders.
  2. One folder is one sample. The folder name becomes the run name. Loose files dropped on their own are treated as a single sample.
  3. Check the list below. Untick anything you do not want in this batch, and set the sample name and record fields. Both can be changed after the run.
  4. Start runs. Upload and analysis run per sample; progress shows underneath. You can leave the page once it starts.

Accepts .fastq and .fastq.gz (also .fq), in any read orientation, up to 20 GB per sample. Output from a previous run found inside a sample folder is ignored and listed as skipped.

2 · Sample records

One run per folder. Untick a folder to exclude it. Fields on the left apply to every included sample; the table overrides them per sample. All fields stay editable after the run.

3 · Running

starting0%

Queue

Select a run from the list on the left.

Sample record

Audit trail

Calls

more than one allele called

Junction browser

Read pools

reads grouped by primer pair
One gzipped FASTQ per primer bin.

Quality checks

Per-bin read counts

Sequencing reads

Read length distribution

Reads per primer bin

Bin profile across samples

Principal components

Alleles seen

What the columns mean

also on hover

Sample records

Edit any cell, then save. These fields group the Summary charts. Expected result is used for the concordance check. Fields left blank stay blank; nothing is inferred.

How a breakpoint is found

All runs follow these steps. Counts on the right are from barcode85 (--THAI) and show how many reads survive each step.

Raw reads 73 FASTQ files, either orientation, ~6% error 660,885 Primer binning seed-and-verify fuzzy match, both primers, both strands 27,072 carry both ATHAI primers Length window modal product size measured per bin, keep ±25% 26,618 product is 1,287 bp Abundance filter a bin under 20% of the top bin is barcode crosstalk 1 bin −a3.7 rejected at 1.0% Map to the locus minimap2 map-ont against chr16 or chr11, chosen by the primer bin find the junction. The first method that confirms is used split aligner already splits the read --SEA --FIL -GB and 7 of the 8 beta alleles clip far side too short, clipped off --THAI −α4.2 −α3.7 Hb Lepore inversion segments face opposite ways β-Asian-Indian chr11 only anchor read looks entirely normal fallback, currently unused Confirm rebuild the deletion allele, map back, need >50% across the join 98.0% spanning Place the join slide it through the homology block, keep the best fit 0.9906 identity Call: --THAI, 33,453 bp, chr16:149,858 → 183,311

All three find the same junction and are confirmed the same way. No single method catches every allele, which is why there are three. The Method column records which one was used.

What each method sees

One deletion, three ways it appears in the alignment. Grey is the reference, blue is where the read aligns, dashed is the deleted DNA. An inversion looks like the split case except that the second segment runs backwards.

split aligner reports both halves deleted one read, two alignments. minimap2 emits the second in the SA tag clip far side too short, so it is thrown away deleted the 25–30 bp tail is below the aligner's threshold, so it is soft-clipped and lost — re-aligning the clipped tails on their own recovers it anchor nothing split, nothing clipped homology homology the read aligns cleanly and contiguously to one copy. Nothing marks the junction but cut it up: the pieces that place uniquely sit further apart than the read is long, and that difference is the deletion

Quality gate

Applied to every run. Thresholds derive from six samples and are provisional. Weak evidence flags a run for review; only a failed assay fails it. Thresholds can be overridden per site.

Confirmation

Every candidate junction is re-tested independently of the method that found it. The two flanks are spliced into a synthetic deletion-allele contig and the reads are mapped back to it. A correct breakpoint lets reads run straight across the join; a wrong one leaves them clipped. The join is then slid through the homology block to its best-fitting position, which is what the Identity figure reports. All fifteen samples reach at least 98% spanning, thirteen of them 100%.

Why the deletion is reconstructed

The contig cannot show a gap: the deleted bases are absent from it, so reads run straight through. Mapping to the reference directly shows no gap either. For −α4.2 and −α3.7 the whole amplicon aligns contiguously to one copy of the duplication, which is why those breakpoints were missed originally. The Deletion view therefore re-expresses each confirmed alignment in chr16 or chr11 coordinates, with a deletion of the called size inserted at the breakpoint. The coverage track shows it most clearly: a solid block either side, nothing between. An inversion has no such view: the span between its breakpoints is rearranged rather than absent, so only the breakpoint view is offered.

Barcode crosstalk

Index hopping leaks a few percent of one sample's reads into every other barcode on a flow cell. A leaked read carries a genuine junction and produces a plausible call, so the filter uses abundance, not junction quality. A bin below 20% of the sample's dominant diagnostic bin is rejected and reported in the per-bin table. Measured here, contamination never exceeded 9.2% of the top bin while a real allele is 100%, and spiking confirms the line: rejected at 1–15%, called at 20% and above.

What this does not tell you

Zygosity is not determined. The assay is allele-specific PCR, so read ratios carry no copy-number information. Inside a homology block the exact junction base is ambiguous: the deletion size is well determined, the coordinate is one valid representation of it.

Your sign-in

Other people's accounts

Accounts are local to this instance. A new account is invited to set its own password, so nobody else ever knows it - not even whoever created the account. The link works once and expires. Disabling an account ends its sessions immediately.

Sign-in log

Account history

every change to an account, and every sign-in