The confidence score#
The final stage of database processing assigns a confidence score to each TCR:peptide:MHC complex, computed from the reported method entries.
The score evaluates TCR sequence confidence, identification confidence and verification confidence, by the following criteria:
Ensuring TCR sequence is correctly identified according to
method.sequencingandmethod.singlecell(1-3 points)sanger - several cells sequenced (2+ cells sequenced according to
method.frequency) - 2 points, otherwise 1amplicon-seq - frequency is at least
0.01- 3 points, otherwise 1. Wheremethod.frequency.countis known it must also be at least 2 reads; where it is not (a float or a percentage carries no count) the frequency decides alonesingle-cell - 3 points if performed
Initial identification of TCR:pMHC is correct according to
method.identification(0-1 point)sort-based - frequency is higher than
0.1according tomethod.frequency)culture-based - frequency is higher than
0.5limiting dilution/culture prior to sequencing -
method.frequencyis ambiguous here, check whether it is higher than0.5
Verification T-cell specificity (0-3 points)
direct method - 3 points, e.g. has PDB id (
meta.structure.idis not empty) or some other method that directly evaluates TCR:pMHC bindingtarget stimulation-based - 2 points
staining-based - 1 points
If verification is performed, the TCR sequence is assumed to be known, so the score from part
1.is set to 3
The final score is the minimum of the part 1. score and the sum of the part 2. and part 3. scores.
Among records that point to the same unique complex entry, that is the same set of unique complex fields - independent submissions, replicas, and so on - the maximal score is taken.
score |
description |
|---|---|
0 |
Low confidence/no information - a critical aspect of sequencing/specificity validation is missing |
1 |
Moderate confidence - no verification / poor TCR sequence confidence |
2 |
High confidence - has some specificity verification, good TCR sequence confidence |
3 |
Very high confidence - has extensive verification or structural data |
Score rules#
The table below is rendered from vdjdb.score.confidence while this page builds.
frequency()method.frequencyas a fraction in [0, 1]. Unparseable or absent is 0.0.cell_count()The numerator of an
n/mfrequency. A percentage gives no cell count, so 0.sequencing_score()How much the sequence can be trusted: single cell > Sanger > amplicon depth.
row_score()The per-row score, before the per-signature maximum.
The score is a maximum over the sample signature cdr3.alpha, v.alpha, j.alpha, cdr3.beta, v.beta, j.beta, species, mhc.a, mhc.b, mhc.class, antigen.epitope, so the same clonotype assayed twice takes the better of the two.