Experimental design on product development
Introduction What is the traditional developing method? What is experimental design? What do we need and what kind of possibilities do we have for designing? Why is it useful during product development? Where and how can we use or apply the design results?
Traditional method Developing a hypothesis Conducting an experiment to test the hypothesis Modifying the hypothesis on the basis of the experiment results Conducting an additional experiment based on the modified hypothesis
Drawbacks of the traditional method Limited understanding of variable effect No information about interaction with other components The approach is extremely inefficient Requires too many experiments and too much time Expensive
The developer must work until... a satisfactory result is found all possibilities are exhausted all available time and money are exhausted... Because the... data processing is slow the number of errors is high
What kind of possibilities do we have? Data from previous investigations Experience, knowledge and intuition Computer software neural networks designing models
Biological neuron structure Dendrites Cell body Axon Electrical signal Synapses
Signal transport between neurons Eletrical signal Axon Presynaptic membrane Neurotransmitter material Synaptic gap Postsynaptic membrane Electrical signal
Relationship between biological and artificial neural network system (ANN) Linear Treshold gate Treshold value
What is artificial neural network? A modelling tool that discovers a relationship from a database of examples An automatic mathematical construction method for modelling preparations directly from data A cost-effective possibility for designing a new preparation
What is needed for design? Pre-examination results of the substances to be used Know the final dose or the final pharmaceutical form Determination of preparation parameters Test results of the preparation
Application areas of ANN Pattern recognition Economic and social models Financial sector Marketing modeling Optimization of investments Telecommunication Signal Analysis Data compression Environmental Protection Industry Weather forecast Quality control Biology Manufacturing Planning Fault Diagnosis
General characteristics of ANN 1. It consists of nodes and links between the knots the weighted input of the input signals is calculated the amount is compared with the threshold(s) linear or nonlinear transmission function their "behavior" changes their behavior and the links between the knots 2. It can be divided into three main parts an interconnected network of nodes the node activation rule the learning rule for nodes
Applied ANNs Associating networks Feature extracting networks Nonadaptive networks Back-propagation model
One and more "hidden" layers of neural systems
Bayesian network (Generalized Regression Neural Network) Hidden layer Regression layer Input layer Output layer
Data modeling system Stuttgart Neural Network Simulator
Other applied ANNs Box-Behnken Central composite Algorithm: back-propagation Pseudo-random Resilient Propagation (Rprop) Resilient Propagation with MAximum-Posterior (Rprop-MAP)
Determining the number of hidden neurons M.N. Jadid et al: Eng.Appl.Artif.Intell. 9 (1996) 303-319 J.C. Carpenter et al: AI Expert 10 (1995) 31-33
The optimization of the node number of the "hidden" layer depends on: the number of inbound and outbound neurons the number of data used to teach the "noise level" of the desired value the complexity of the function you want to teach the type of neural network from the activation value of hidden nodes from the teaching algorithm
How does a neural network work? Data collection Determination of variables and parameters Data input into the computer Teaching and checking Application of knowledge
What do we need? Input parameters Output parameters effective materials disintegrants fillers binders glidants lubricants etc. particle size flow properties hardness friability disintegration time dissolution rate etc.
Predicted and experimental dissolution values Y1 Y2 Y3 Y4 P E P E P E P E F1 47,00 47,03 74,66 74,69 92,61 92,62 103,4 103,4 F2 21,38 21,36 47,17 47,16 91,38 91,42 102,3 102,3 F3 13,81 13,81 21,61 21,61 33,95 33,94 47,49 47,48 F4 17,03 17,03 26,23 26,22 47,02 46,99 67,21 67,19 F5 14,06 14,06 21,81 21,81 32,86 32,85 49,55 49,54 F6 23,12 23,12 50,35 50,36 80,40 80,44 99,16 99,19 F7 17,02 16,66 30,08 29,81 49,79 49,87 67,44 68,76 F8 17,02 17,37 30,08 30,35 49,79 49,71 67,44 66,11 F9 14,47 14,46 26,46 26,44 42,44 42,41 62,08 62,06 F10 36,11 36,11 65,80 65,80 77,18 77,17 99,43 99,43
Release profiles of ASA from model formulations
Response surfaces of the influence of the percentage of Eudragit L 100 and tablet hardness on the percentage of ASA released after (A) 1 hour, (B) 2 hours, (C) 4 hours, and (D) 8 hours, predicted using the GRNN.
Contour plots of the influence of percentage of Eudragit L 100 and tablet hardness on percentage of ASA released after (A) 1 hour, (B) 2 hours, (C) 4 hours, and (D) 8 hours, predicted using the GRNN.
Predicted and experimental observed Aspirin release from optimal formulation.
Crack velocity and film opacity response surfaces
Experimental design
What does experimental design involve? A practice that employs statistical tools and methods in scientific experimenting Variables via which we are attempting to make a correlation or regression with a measurable input and output we are trying to predict
What is necessary for designing Preliminary investigations of the materials The dosage and the dosage form of the final product Determination of the parameters of the preparation process Results of the investigations of the preparations
Preliminary investigations
Designing a pellet
Designing the experiment
Design method selection Central composite Face centered Box-Behnken Simplex Equiradial Random Simplex centroid Simplex Lattice Hybrid
Detailing of the experiments
Parameterizing the ANN
Setting the control parameters
Setting up the parameters
Investigated compositions
Results of investigations 1.
Results of investigations 2.
Finding the best or worst sample
Importance of ingredients
Querying a new sample
Predicting the properties 1.
Predicting the properties 2.
Material and method Granule: API: Filler: Disintegrant: Binder: Dilthiazem HCl (Dilt) Vivapur 101 (V101) Era-Tab (Era-T) Pharmacoat 603 (P603) Desing mode: Preparation: Hybrid design Freund CF-360 granulator
Design the core Dilt Era-T V101 P603 Min. Max. Min. Max. Min. Max. Min. Max. 100 g 500 g 300 g 500 g 100 g 500 g 10 g 40 g
Designed compositions Dilt Era-T V101 P603 (g) (g) (g) (g) MST-1 315 339 346 23 MST-2 144 440 417 36 MST-3 471 413 116 17 MST-4 426 408 166 23 MST-5 374 453 173 19 MST-6 342 378 280 23 MST-7 365 420 215 21
Granule preparation
Preparation conditions Liqu. add. Rotor Temp ml/min rpm C MST-1 10 140 45 MST-2 10 140 45 MST-3 15 160 45 MST-4 15 160 55 MST-5 20 200 55 MST-6 20 200 65 MST-7 20 200 65
Investigated properties MST-1 MST-2 MST-3 MST-4 MST-5 MST-6 MST-7 Density Roundness Hardness Friability (N) (%) 1,649 1,19 0,751 0,15 1,753 1,17 0,747 0,12 1,583 1,29 0,729 0,18 1,651 1,16 0,746 0,23 1,678 1,25 0,757 0,31 1,731 1,22 0,781 0,17 1,939 1,19 0,731 0,19
Shape of granules
Parameterizing the ANN
Editing the training parameters
Structure of ANN
Control parameters
Prediction of density
Prediction of hardness
Prediction of friability
Prediction of roundness
Searching for samples or parameters
Directed search
Areas where it is useful Area Application Applied DoE TYPE Oral drug delivery Tablet formulation development Multivariate design (fractional factorial design in 14 variables, 2 14 9 design, 35 experiments) Multivariate design + simplex optimization by Modde Optimizer Fractional factorial designs (two studies); design space definition using a simplified Bayesian Monte Carlo simulation Oral drug delivery immediate release (IR) Mixture design Dispersible tablets development Several factorial experiments at 2 3 factors, 2 3 levels Immediate release tablet platform Resolution V 2 5 1 fractional factorial design Fast dissolving pellets 2 5 1 fractional factorial design, five factors (four numeric and one categorical), two levels Oral drug delivery modified release (MR) Gastroretentive dosage form 3-level-3-factor, Box Behnken design
Area Application Applied DoE TYPE Inhalation drug delivery Powder for inhalation (formulation and process development) Half-fractional factorial design with five factors at two levels with resolution V Face centered central Composite Design with three factors at three levels Risk assessment by Lean QbD Software Transdermal drug delivery Patch development 2 4 full factorial design Iontophoretic delivery Face-centered central composite design (total number of experimental combinations 2 k +2k + n 0, with k = number of independent variables and n 0 = number of repetitions of the experiments at the center point) Cutaneous drug delivery (Topical) Ocular drug delivery Nanoemulsion for leishmaniasis (formulation development) Microsponge-based gel for surgical wounds (development) PEGylated PLGA nanospheres (optimization and characterization) 2 2 full factorial design 3-factor, 3-level Box Behnken design Central composite factorial design Liquid crystalline nanoparticles (formulation optimization) Fractional factorial design 2 5 1 ; simplex-lattice experimental design
Area Application Applied DoE TYPE Injections Biopharmaceuticals Formulation for parenteral nutrition (development) Toward QbD implementation in the biopharmaceutical industry QbD for biopharmaceuticals QbD for risk assessment and management (Case Study on Monoclonal Antibody) D-optimal experimental design (mixture design) Nanopharmaceutics Antibody Formulation Robustness Protein formulation Development and industrialization of polymeric targeted nanoparticle drug delivery platforms Multivariate study (full factorial) including three factors at two levels Solid Lipid Nanoparticles for Inhalation (process development) Two-level full factorial design (with no center points and three repetitions for each level)
Area Application Applied DoE TYPE Pharmaceutical processes Dry powder inhaler capsule filling D-optimal model with design statistics G- efficiency with three replicates Excipient micronization Freeze-drying of injectables Film-formation by spraying Full factorial (three variables at two levels, eight runs) Design space calculation (according to a cited model) Rechtschaffner Res V two-level fractional design for four variables with center point Nano-precipitation and nanospray-drying Design model based on integrated-variance optimal design for surface response Analytical methods Method (e.g. chromatographic) development Test methods Adhesion test (for patches) Randomized response surface design, five factors, 38 runs In vitro aerosol deposition in human cast (for inhaled products) Half-factorial design Generic drug products Generic product equivalence Fractional factorial design with triplicate center points
Conclusions It is necessary to know the properties of the materials The dosage form and the final dosage of the preparation must be known for the preliminary investigations It is possible to determine the optimal parameters of the preparation process It can generally be used for the development of new products
The benefits of computerized design Short data-processing time Economical design Fewer errors Better preparation More profit