Title: Expected Design Space: a Bayesian perspective based on modeling, prediction and multicriteria decisi
1Expected Design Space a Bayesian perspective
based on modeling, prediction and multi-criteria
decision method
Pierre Lebrun ULg, BelgiumBruno Boulanger UCB
Pharma, SA, BelgiumPhilippe Lambert Ulg,
BelgiumBenjamin Debrus ULg, BelgiumPhilippe
Hubert ULg, Belgium
Friday , October 23
2Overview
- The process
- Liquid chromatography
- Multivariate regression correlated responses
- Definition
- Design Space
- Objective functions
- Bayesian model
- Introduction
- Priors
- Predictions
- MCDM acceptance limits and desirability
- Results
- Conclusions
3Example of application
- - A chromatographic method is to be optimized
using DOE and response surface models - - P3 peaks to be separated in the shortest time
Gradient time (min.)
pH
pH 2.6 6.3 10
Gradient (min.) 10 20
30
(N x 3P)
Design Space set of conditions (pH, Gradient,)
in the domain, such that separation and short
run are guaranteed for the future
These 9 responses are correlated
4Overview
- The process
- Liquid chromatography
- Multivariate regression correlated responses
- Definition
- Design Space
- Objective functions
- Bayesian model
- Introduction
- Priors
- Predictions
- MCDM acceptance limits and desirability
- Results
- Conclusions
5ICH Q8 (may 2006) definition
-
- ? The Design Space is the set of conditions
giving solution within Acceptance Limits - the established range of process parameters and
formulation attributes that have been
demonstrated to provide assurance of quality. - Working within is not considered as a change in
the analytical method. - n.b. If the Design Space is large w.r.t. control
parameters or conditions, the solution is
considered as robust
6Proposal definition of Design Space
- When the process is known
- Design Space (DS)
- domain of factors
- set of combinations of factors
- the responses obtained for the
condition - the set of acceptance limits (e.g.
resolutiongt1.2 ) - the quality level (e.g. P( resolutiongt1.2) gt
0.8)
However - in development validation, the
process is unknown, its performances are
estimated with uncertainty - purpose predict
the space that will likely in the future provide
most outputs within acceptance limits
Peterson, J. Qual. Tech, 36, 2, 2004
7Proposal definition of design space
- When the process is unknown
- Expected Design Space (DS)
- The predictive probability of achieving the
acceptance limits is larger than , the
quality level - Given the process parameters
- The DS is located using predictions from models
estimated during development validation
experiments
8Objective functions
Specific problem Criteria / Objective functions
- - Sum, product, min/max of the responses
- - Discontinuity
- - Non linearity
Ex
? i.e. DS is the set of conditions, such that the
probability that objectives will be
simultaneously (jointly) within the Acceptance
Limits is higher than
9Overview
- The process
- Liquid chromatography
- Multivariate regression correlated responses
- Definition
- Design Space
- Objective functions
- Bayesian model
- Introduction
- Priors
- Predictions
- MCDM acceptance limits and desirability
- Results
- Conclusions
10Bayesian model
- Multivariate multiple linear regression model
- The joint posterior distribution of the
parameters is obtained as follow
- and are assumed independent a priori,
therefore
Go to results !
11Bayesian model
Posterior distribution of the parameters
Non informative
priors
(Box and Tiao, 1973)
Informative priors
reflects the certainty put in , the
prior correlation matrix
Elements of
Meff takes into account the correlation between
the responses (Sattertwhaite, 1946)
Meff lt M
12Prediction
- Plausible values of one prediction ,
conditional to the available information
predictive posterior distribution
- A draw from the joint posterior of parameters
- A draw from the Normal (model) conditionally to
the posterior of parameters
13MCDM acceptance limits and desirability
- From the predictive distribution
- of responses
- of objective functions (criteria to optimize)
- Use of desirability functions for Multi-Criteria
Decision Making (MCDM)
Ex Desirability functions of Le Bailly and
Govaerts Each criteria, conditional to x, is
transformed gt Domain 0,1, using the CDF of
the Normal distribution, F
D(O) is the global Desirability Index
Weights allow flexibility and balanced decision
14MCDM acceptance limits and desirability
- Joint predictive distribution of two objectives
at a given x0
Brown Arithmetic mean Red Geometric
mean Green Harmonic mean Blue Acceptance
limits
15MCDM acceptance limits and desirability
- Desirability to acceptance limits
- Acceptance limits can be viewed as a special case
of classical desirability-based MCDM
bz 0 az
Acceptance limit
? Step desirability functions
az
az
16MCDM acceptance limits and desirability
bz sd(crz) az
bz sd(crz)/2 az
bz sd(crz)/5 az
bz sd(crz)/10 az
bz 0 az
When bz is 0, weights have no importance any
longer
-Using acceptance limits has the advantage to
have clear limits expressed in the criteria
space -Using classical desirability functions
allows trade-off between objectives ? this is
the experimenter choice
17Overview
- The process
- Liquid chromatography
- Multivariate regression correlated responses
- Definition
- Design Space
- Objective functions
- Bayesian model
- Introduction
- Priors
- Predictions
- MCDM acceptance limits and desirability
- Results
- Conclusions
18Results
Gradient time fixed at 20 min.
Reflects the within peak correlation
03 Mean minimal resolution
Objective function is non-linear
19Results - MCDM - uncertainty
- From the joint predictive distribution of
objective, using acceptance limits
? correlation is taken into account ?
uncertainty
- Ex - separation gt 0, resolution gt 1.2, Run lt 7
min. - - p gt 0.8
(Predictive) probability map that the three
objectives are achieved
X0
20Validation
Mean predicted Real
21Considerations
- DS can be subject to subjectivity
- Model choice (consider Bayesian model averaging)
- Choice of prior distribution
- Choice of responses or criteria used in MCDM
(separation, resolution or both ?) - What is not subjective
- Data and parameters (model) uncertainty
- Good statistical models are prerequisite
- Good data as well
- Design of experiments
- Control of non modelled but influential factors
(to reduce noise) - If it is assumed the responses do not follow the
same regression equation, an alternative is to
use seemingly unrelated regression (SUR) - In some cases, constraints apply on the responses
or criteria. They can be included - using truncated distributions (e.g. Geweke, 1991)
- via rejection sampling in MCMC, if constraints
are complex
22Conclusions
- Design Space must be defined on prediction of
future results given past experiments - The uncertainties in the model should be taken
into account in predictions - Bayesian multivariate multiple regression is
powerful and flexible to model correlated
responses and to manage uncertainty - The Design Space is straightforward to obtain
with Bayesian models - The joint predictive posterior distribution of
objective functions allows the development of
Multi-Criteria Decision Methods (MCDM) - About expected future performance
- under uncertainty
- taking into account dependencies between criteria
- Bayesian models in chromatography can take
advantage from the long history of the domain,
e.g. to set up informative priors - Acceptance limits is a special case of
desirability-based MCDM
23