Domain-validated residual XGBoost surrogates for model-informed biopharmaceutics and population pharmacokinetic inference.

Authors

Sergio Sánchez-Herrero, Diego García-Álvarez, Laura Calvet, Ángel A Juan

Published in

European journal of pharmaceutics and biopharmaceutics : official journal of Arbeitsgemeinschaft fur Pharmazeutische Verfahrenstechnik e.V. Pages 115260. Oct 27, 2026. Epub Oct 27, 2026.

Abstract

Mechanistic population pharmacokinetic (popPK) models support model-informed biopharmaceutics, but repeated exact simulations can make nonlinear mixed-effects inference computationally expensive. We developed a domain-validated residual XGBoost surrogate embedded in stochastic approximation expectation-maximisation (SAEM) for a two-compartment popPK model implemented in PhysPK®. The emulator learned local residual deviations from a mechanistic baseline and was activated only after held-out validation against exact simulations. Fixed-effect updates were constrained within a log-parameter trust region, with exact fallback outside the validated dosing and parameter domain and exact-model recomputation of final diagnostics. Evaluation used a published six-animal preclinical monoclonal-antibody dataset, internal cohorts, 200 balanced noisy parametric bootstrap replicates of 30 virtual animals each, and sparse-sampling variants. In the external proof-of-concept cohort, surrogate-assisted SAEM reproduced the exact reference solution with profile-level mean relative error of 0.0041 and fixed effects within 3.1%. All 200 bootstrap runs completed and met the prespecified fixed-effect stability criterion by iteration 49. Bootstrap emulator error remained low (mean MRE 0.0027; maximum 0.0028), whereas θ3 (Q) showed the greatest uncertainty. Exact local and Morris sensitivity analyses supported weak practical identifiability rather than surrogate bias as the source of this instability. Single-profile evaluation was approximately 49-fold faster, supporting auditable surrogate-assisted inference within the validated applicability domain.

PMID:
42946231
Bibliographic data and abstract were imported from PubMed on 28 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.