Analytical Validation and Detection Limits - Usage Guide

Overview

Establishes and audits the sensitivity claim of a liquid-biopsy assay using measurement science: the Poisson genome-equivalent sampling ceiling, the error-suppression floor, and the CLSI EP17 LoB/LoD/LoD95/LoQ framework, including the per-locus-vs-panel-integrated distinction and honest LoD reporting conditioned on input mass.

Prerequisites

pip install numpy scipy statsmodels

Quick Start

Tell your AI agent what you want to do: - "How many genome equivalents do I need to detect a 0.1% variant 95% of the time?" - "Estimate the LoD95 from my dilution series of detection calls" - "Compute the panel-integrated LoD for a 30-variant bespoke MRD assay" - "Audit this 'detects 0.1% VAF' sensitivity claim for missing input mass"

Example Prompts

Detection Limits

"Given 25 ng of cfDNA input, what is the probability of detecting a true 0.05% VAF variant at a single locus?" "What input mass puts lambda at 3 for a 1e-4 variant, and why is that the 95% sampling threshold?"

Validation Study

"Fit a probit detection curve to my dilution series and report LoD95 with the LoB from my blank replicates" "Distinguish the LoD from the LoQ for this assay and tell me which VAFs I can report as quantitative"

Panel Design

"Compare the per-locus LoD to the panel-integrated LoD for tracking 16 vs 48 clonal variants" "Explain why a bespoke panel reaches ppm sensitivity when each locus alone floors at 1e-4"

Reporting and Audit

"Rewrite this LoD claim to condition on input genome equivalents, consensus depth, and replicate detection rate" "Flag whether this sensitivity number is sampling-limited or error-limited at the stated VAF"

What the Agent Will Do

Convert input mass to haploid genome equivalents (~330/ng) and compute lambda = input_GE x VAF.
Return the Poisson detection probability and the minimum GE needed for lambda >= 3 (95% sampling-detection).
Identify whether the achieved LoD is sampling-limited or error-limited at the target VAF.
Estimate LoB from blank replicates and fit a probit/logistic detection curve for LoD95 from a dilution series.
Combine per-locus detection probabilities into the panel-integrated LoD under a >=k-of-N positivity rule.
Reframe any bare-VAF sensitivity claim into an LoD conditioned on input mass, consensus depth, and replicate detection rate.

Tips

Anchor to molecules - Always pair a VAF with input genome equivalents; lambda = GE x VAF, not the VAF, determines detection.
Stop buying depth - Past the deduplication plateau the assay is sampling-saturated; add plasma volume or conversion efficiency instead.
Separate detect from quantify - LoD is binary detection; LoQ (CV<=20%) is quantitation. Report near-floor VAFs as detected/not-detected, not as measured values.
Mind the integration level - A per-locus LoD and a panel-integrated LoD differ by orders of magnitude; never quote one for the other.
Duplex for the deepest claims - Single-strand UMI consensus does not remove template-resident deamination/oxidation damage; sub-1e-5 claims need duplex strand-concordance.
Use commutable standards - Validate against contrived SEQC2 Sample A / HCC1395 admixtures fragmented to ~160 bp, and treat commutability with real plasma as the open caveat.

ctdna-mutation-detection - applies these limits to low-VAF somatic calls
longitudinal-monitoring - per-timepoint LoD and left-censoring of undetectable samples
tumor-fraction-estimation - the ~3% CNA-based detection floor as an LoD
experimental-design/multiple-testing - repeated-surveillance specificity and FDR
clinical-biostatistics/power-and-sample-size - validation-study design