Real experimental labels
The doublet-driven signal (real labels)
The stated claim
“A rare population shows a distinctive co-expression signature between two conditions.”
benchmark case hto_doubletsDeciding evidence
On a real dataset with experimental HTO multiplet labels, the doublet score reaches AUC ~0.78, recovering ~20% of labelled doublets at the explicit expected-rate cutoff.
Sentinel: doublet-driven signal, scored against wet-lab labels — the one benchmark graded on real experimental truth rather than a planted answer.
A doublet — two cells captured as one — can masquerade as a novel co-expressing cell state. This is the only VERITAS benchmark scored against real experimental labels (hashtag oligos), so the numbers reflect real-world detection, honestly: strong ranking (AUC ~0.78), partial recall at the expected-rate cutoff.
What the engine does
A tempting rare population
A small cluster co-expresses markers of two lineages — exactly what you would expect from a genuine hybrid state, or from doublets.
The doublet sentinel runs
VERITAS scores each cell's doublet likelihood and compares against the dataset's experimental HTO multiplet labels — real truth, not a simulation.
Honest numbers, honest caveat
The score ranks doublets well (AUC ~0.78) but recovers ~20% at the expected-rate cutoff. VERITAS surfaces the risk and reports the recall — it does not claim to detect every doublet.
Verdict: at risk
The signal is flagged as potentially doublet-driven with a re-derivable statistic, so the scientist can decide whether to trust the population.
What the scientist does next
The scientist re-checks the rare population with orthogonal evidence before publishing a novel cell state.
Verify your own finding
Run a bundled sample of this exact failure mode in the console, or open the validation atlas to see how it is graded.