chapter 06 · validation surface

AI in bioanalytical.

ICH M10 v2 is anticipated to open narrow scope for AI/ML in peak detection, integration, calibration-fit (industry-projected Step 2b ~Q4 2026; ICH has not published an official v2 timeline). A standalone ICH AI/ML method-validation document is anticipated 2031–2033 (iFeed projection, not ICH-confirmed). Until either lands, every AI use case lives inside the PCCP framework.

/ 10

AI in bioanalytical.

Where AI is changing the validation surface · 2026

ICH M10 v2 (target Step 2b Q4 2026) opens narrow scope for AI / ML in peak detection, integration, and calibration-fit. A standalone ICH document on AI / ML method validation is expected 2031-2033. Until then, every AI use case in the validation surface lives inside the PCCP (Predetermined Change Control Plan) framework.

Production · since 2018

Peak detection & integration.

Replaces manual peak-boundary judgement; reduces operator variability. Deployed in Sciex, Waters, Shimadzu instruments since 2018. ICH M10 v2 narrow-scope acceptance: algorithm version, training data, validation set documentation required.

Optional in vendor SW

Calibration-curve fit selection.

Replaces weighted-regression manual choice. Optional in Analyst, Waters MassLynx. PCCP framework extends to bioanalytical AI / ML around 2027.

Piloting · CROs

ISR sample selection.

Intelligent randomisation across concentration range, time points, demographics. Piloted by Covance, Charles River. Inspector acceptance projected 2029 (post-pilot lag ~3 years).

Emerging

Anomaly detection.

Outlier identification, assay drift detection. Emerging in LIMS / EDC vendors. Documentation of AI model versions and retraining triggers not yet specified in ICH M10.

Low maturity

RCA automation.

Failed-ISR root-cause analysis. Mostly manual as of 2026. No specific framework yet. 483 citation risk if AI-generated RCA lacks human review.

Supportive evidence

PBPK / MIDD.

BCS Class III biowaiver justification via model-informed evidence. Accepted as supportive (not primary) in FDA / EMA 2024-2025. FDA 2024 MIDD guidance frames the boundary.

Established

Liquid-handling robots.

Tecan, Hamilton sample-preparation optimisation since ~2010. Replaces manual SOP design. Standard QPCR / LC-MS/MS practice. No specific ICH M10 friction.

Out of scope through 2030

Generative authoring.

FDA / EMA / PMDA signalled unacceptable for validation reports through 2030. Document authorship requires human-in-the-loop.High