Execution of complex Bayesian workflows with the DDMoRe Interoperability Framework: a case study in the diabetes area
Cristiana Larizza (1), Elisa Borella (1), Lorenzo Pasotti (1), Palma Tartaglione (1), Mike Smith (2), Stuart Moodie (3), Paolo Magni (1)
(1) Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy; (2) Dept Statistical Pharmacometrics, Pfizer Global Research & Development, United Kingdom; (3) Eight Pillars Ltd, Edinburgh UK
Objectives: The DDMoRe Interoperability Framework (IOF)[1] is an integrated infrastructure that enables the exchange and integration of models across different modelling languages and tools. The IOF is based on two interchange standards: Pharmacometrics Markup Language (PharmML)[2] and Standard Output (SO)[3]. It allows users to encode models via Modelling Description Language (MDL)[4], convert them into PharmML, run different tools via R script, and store and reuse results via SO. We updated the previously developed PharmML-to-BUGS converter, NMTRAN-to-BUGS data converter, and the WinBUGS connector[5,6] to support new features. The public-released IOF[7] was updated with this new software suite. The objective of this work is to demonstrate its use to design and execute a complex interoperable workflow based on two diabetes-related models.
Methods: The selected case study is here summarized: 1) identification of a multivariate regression model, relating demographic covariates with the parameters of a linear 2-compartment model of C-peptide (CP) kinetics, on a 207-subject dataset[8]; 2) estimation of the CP kinetic parameters for a new subject by using the identified regression model; 3) identification of the glucose-insulin minimal model (MM) and estimation of insulin secretion rate (ISR) by using the estimated CP kinetic parameters, and CP and glucose plasma concentration data collected in the considered subject after an IVGTT[9]. The IVGTT data have been analysed by testing in the DDMoRe IOF three modelling approaches, whose results have been compared to those reported in literature: i) a full non-Bayesian approach, providing point estimates of MM parameters in a given individual; ii) a mixed approach where a Bayesian identification of the MM was considered; iii) a full-Bayesian approach where also the uncertainty on CP kinetics parameters was taken into account in the MM identification. NONMEM and PsN were used for non-Bayesian tasks, whereas WinBUGS was used in the Bayesian approaches.
Results: Parameters estimates obtained at points 1) and 3) were consistent with the published values, which were originally obtained via Matlab.
Conclusions: This work demonstrates the usefulness of the IOF to execute complex Bayesian workflows, using standard languages for model encoding and task execution, even running target tools with different estimation methods. This also overcomes the encoding difficulties of complex PK-PD models in WinBUGS that are normally written via a nontrivial combination of BUGS and Pascal code.
References:
[1] http://www.ddmore.eu/official-release-interoperability-framework
[2] Swat, M. J., Moodie, S., Wimalaratne, S. M., Kristensen, N. R., Lavielle, M., Mari, A. et al. (2015). Pharmacometrics Markup Language (PharmML): opening new perspectives for model exchange in drug development. CPT: pharmacometrics & systems pharmacology, 4(6), 316-319.
[3] PAGE 25 (2016) Abstr 5954 [www.page-meeting.org/?abstract=5954]
[4] PAGE 25 (2016) Abstr 5899 [www.page-meeting.org/?abstract=5899]
[5] PAGE 24 (2015) Abstr 3565 [www.page-meeting.org/?abstract=3565]
[6] PAGE 25 (2016) Abstr 5829 [www.page-meeting.org/?abstract=5829]
[7] https://sourceforge.net/projects/ddmore/files/install/SEE/
[8] Magni, P., Bellazzi, R., Sparacino, G., & Cobelli, C. (2000). Bayesian identification of a population compartmental model of C-peptide kinetics.Annals of biomedical engineering, 28(7), 812-823.
[9] Magni, P., Sparacino, G., Bellazzi, R., Toffolo, G. M., & Cobelli, C. (2004). Insulin minimal model indexes and secretion: Proper handling of uncertainty by a bayesian approach. Annals of biomedical engineering, 32(7), 1027-1037.
