Webinar transcript: "Design of Experiments (DoE) to better understand tablet formulations"

webinar visual DOE Nano social media with DMU logo

Feature article


May 31, 2021

On April 21st, Mariana Bezerra, PHD student from De Montfort University (UK), presented her work on Design of Experiments, and the MEDELPHARM Science Lab showed us how it can be performed thanks to a STYL’One Nano. The session ended with a Q&A.

For those who could not join us you can access the transcript below.

Note: Registered members of MySTYL'One can access the full video replay HERE.

#webinar #STYLOne Nano #DOE #Design of Experiments #De Monfort University

Webinar Transcript:

Design of Experiments (DoE): an essential tool to better understand tablet formulations

Read below the webinar's full transcript

[Bruno Leclercq - MEDELPHARM] – Introducing the webinar

Good morning, good afternoon, and good evening to all of you, and welcome to the 3rd MySTYL’One Live session.

Today, we have scientists from all over the globe. Thank you for your interest in this event, and for joining the community of STYL’One users. My name is Bruno Leclercq, I'm a pharmacist working Business Development at MEDELPHARM. And I will guide you through this webinar today, where Mariana Bezerra, STYL’One Nano user, will share her expertise with you on Design of experiments for tablet formulation.

But before we get started, just a few words on this platform, and how you can interact with us. You will notice this webinar is browser based, so if you disconnect for any reason, please just click on the link that you received by email to rejoin the platform. You will find a ‘help’ button at the bottom left corner of your screen to fix any audio issue you may have. To improve your event experience you can hide or show the widget section to increase your screen size, and to turn off & on the notification thanks to these two buttons. Feel free to meet the STYL’One community and MEDELPHARM in the chat box. Ingrid Coyle will be more than happy to see you there. Please use the questions tab to raise any questions or comments you may have during this presentation. One important thing - we will only select questions from this section Q&A.

Now let's start. Today’s session will be divided into 4 parts. First, I will briefly introduce MEDELPHARM and the STYL’One Nano that was used by Mariana from De Montfort University during her PhD work. She will then share with us her work on Design of Experiments and showcase practical example where she used the STYL’One Nano. And before concluding with an open Q&A, we will dive into the STYL’One Nano technology, to highlight how quickly you can explore your design space by changing your product and process parameters and get fast results. 

Introduction to MEDELPHARM

[Bruno Leclercq - MEDELPHARM]

Just a few words about MEDELPHARM, the global leader in compaction simulators. MEDELPHARM’s core business leitmotiv is and has always been to innovate by design, development and manufacture of STYL’One tableting solutions for R&D, scale-up and production support.

We are very proud to input innovative and disruptive solutions to the market with the STYL’One series, including the one that you see on the screen, the STYL’One Evo, the most advanced compaction simulator with multi-lay capabilities, including tab-in-tab, the STYL’One Nano, a benchtop research press with compassion simulation capability which Mariana used during her PhD, and last but not least, containment solutions to answer the growing needs in highly potent API's. 

STYL'One family

Family of STYL’One tableting solutions

Before we leave the stage to Mariana let's watch a short video to introduce a STYL’One Nano research press that she used during her PhD.

[Edit: Our STYL’One Nano video can be found here ]

STYL’One Nano is a benchtop tablet press, can use different types of tooling – B or D.
Setting up the press is very easy – you have a quick change over. You can use shaped or even multi-tips tooling, or even oversized tooling. The setup is very easy and user friendly. We have premium instrumentation on the force and the displacement. The filling can be done with an advanced gravity feeder, or if you just have a small amount of powder available, you can also use a hand-fill mode. This video and others are available on MEDELPHARM's YouTube channel.

Enough for the introduction. Let's me present today’s speaker that you already seen on the screen. We are honored to have Mariana Bezerra, PhD student from the School of Pharmacy at DeMontfort University. During her PhD has investigated how in-line UV-Vis Specstroscopy can be applied to characterize quality attributes of hot-melt extrusion products. She also worked on developing oral solid dosage forms for extruded amorphous dispersion using compaction simulation. Her research is supported by Quality by Design principles and tools such as Design of Experiment and multivariate analysis. So please welcome Mariana for her presentation titled “Design of Experiments: an essential tool to better understand tablet formulation. On to you Mariana.

Mariana DMU picture 2

Mariana Bezerra, De Montfort University (UK)

Mariana Bezerra presenting her work on Design of Experiments

[Mariana Bezerra – De Montfort University]

Thanks, Bruno. Welcome, everybody. Thanks for attending this webinar. I just share my screen with you, so you can see my presentation. And you'll see my screen. If so, again, thanks, everybody, for attending this webinar. 

Today, I'm going to give you a talk about “Design of Experiments: an essential to better understand tablet formulations”. and here's the content of this presentation. I'm going to give you an overview on steps for product development, an introduction to Design of Experiments, in a walkthrough DoE steps for formulation studies, since constructing the DoE, but also trying to make the model to predict responses. And this we're going to go through examples of my work formulating amorphous solid dispersions into tablets.

Introduction on Design of Experiments

Today, I'm going to give you an overview of steps for product development, an introduction to Design of Experiments, a walkthrough DOE steps for formulation studies - from developing and creating your DoE to building the model to estimate CQAs. And this will all be with real examples of formulating an amorphous solid dispersion to tablets.

So if we could summarize the steps for product development in three phases, we would have planning, control and improvement.

In planning, we would identify the product requirements, which we defined the quality target product profile, and run a criticality assessment. When we're in control, we want to try to understand product design and process designs to come up with a control strategy. So, when the product life cycle goes on and on, our understanding on the product will improve, and the control strategy will be updated. The important information on this design is that DoE is a fundamental tool to understand process and product design. 

To give you a bit of a background on the system that I worked with during my PhD is an amorphous solid dispersion produced by hot melt extrusion. And this means that Piroxicam which is a BCS class 2 drug is going to be solubilized in a polymer carrier by applying heat and mechanical shear forces, and also the polymer is going to stabilize the active ingredient in amorphous state, which will improve its solubility in water. So for this example, the ASDs contain 16% of Piroxicam and 84% of the Polymer. This means that 125 mg of the ASDs are required to achieve a target dose of 20 milligrams.

 So, in this case, there is a high polymer content on these active systems and this brings challenges to develop this formulation.

One is the low harness of the compact, but also the delayed disintegration time.

If you're going to go from the part, the ASD part goes to the tablets, we're going to have lots of steps and I'm going to walk you through that.

So, initially, we are going to mill this extrudated particles, then select excipients, based on their functionality, blend the excipient with the ACD particles and compact to get a tablet.

This diagram shows you a big overview of all the steps to make this product. And from those, we may identify the inputs of the process, the potential critical material attributes and the critical process parameters.
These ones will be the factors of your potential DoEs.

We also might have the outputs of each unit operation, the critical quality attributes and these ones will be the responses in your DoE. In the case of this case study that I'm going to talk about later, I'm going to evaluate how some of the excipients in the formulations would impact the product CQAs which are:Tensile strength, solid fraction, ejection force, disintegration time and friability.

So now that we know what we want to understand - how to approach experiments with formulation?

We could start by having the formulation and wanting to investigate for example, how the percentage of three components affect the responses - disintegrant, compression aid and lubricant. And we might try one factor at a time.

This means that we are going to fix two of the factors - disintegrant and lubricant - and vary the compression aid, and then we will fix compression aid and disintegrant and vary the lubricant. And finally, fix compression aid, fix lubricant, and just vary disintegrant.

We are going to make the tablets and measure the responses. However, if we want to evaluate the effects of the 3 factors at all these five levels, we might end up with 125 experiments. This simple example show how resources, how one factor at a time experiments might consume a lot of resources, and what would be the alternative?

Design of experiments. But what Design of Experiments really is. So that's why I brought an exact definition here for you today.

Design of Experiments: Definition & principles

A designed experiment is a controlled set of tests designed to model and explore the relationship between factors and one or more responses.

Now that we have the DoE, let's compare one factor at a time and DoE, so again, I'll bring you the same example - disintegration percentage, compression aid and lubricant.

A designed experiment is a controlled set of tests designed to model and explore the relationship between factors and one or more responses
Mariana Bezerra, De Montfort University

We might vary all the levels as I explained in the previous slide, and we wanted to say we want to understand how it will impact tensile strength, and we may find that a certain combination will give you a certain value of tensile strength. However, doing that one factor at a time experiments will not allow us to understand how those factors interact, how for example, a combination of specific values of each factor will give a different value of the tensile strength.

For that we need to change the factor settings simultaneously. This is what DoE will do.

So, we are going to combine minimum and maximum percentage of the three components at the same time, and then we are going to figure out if those factors interact and possibly give another response or another value for the same response-  if the response will behave differently when we vary all the components at the same time.

At this point, we figure out that DoE offers a more effective way of understanding how the factors impact our responses and notably the tablet CQAs.

But at this point, we need to consider some specific requirements of formulations.

So if I bring again the example: that is the formulation that I have, I have my ASD 31.25, Magnesium Stearate, and I want to understand how the proportion of Avicel® PH102 and Ac-Di-Sol® will impact my product CQAs.

So, moving back to my CQAs-  if I apply a classical design, the one that I just showed you in the previous slide, I would, for example, vary the Avicel® from 10 to 20%, and Ac-Di-Sol® from 1 to 5%. I could do that, but we should notice that there's not the level of these two factors that changed. As formulations always have one component, the diluent, that will make up to one specific volume or amount of the formulation, that is not the only thing that I will be able to evaluate.

This in practical terms means that if I do just the DoE with those two factors, I might find effects that are due to the change in the diluent proportion, and not exactly the factors of the DoE, I'm going to have a something happening in the background that I could not estimate.

To be able to account for this variation in the proportion of those three components, I could use a mixture design. The definition of a mixture design is in which the sum of the all the components - the proportion between the components - will sum to a constant value, whatever you set, in the case would be 100 in my case, as some of the formulations, components are fixed, as the dosage and the manganese stearate is 68.25. Another thing to present here for you about mixture design is how they are graphically represented: it is not like the cube we had before, because we need to consider that one, one factor changes or increases for example, the others will decrease proportionally to keep that constant some value.

Making the DoE

So now is when we start making DoE.

So I will just giving you at this point introduction about Design of Experiments, and especially requirements for tablets formulations. Now I want to go step-by-step on what we need to consider on making Design of Experiments.

So we are going to define responses and factors, we are going to specify the model, generate the design and make the table. 

creating the DoE workflow

Creating the Design of Experiments workflow

If we start defining responses and factors: the responses are the variables you want to predict. So anything that you are interested to know about what is going to happen, that is your response. In my case, as I said, the CQAs of the tablet: tensile strength, solid fraction, disintegration time, friability and ejection force. And it's important here to have a target, what you want for that to be maximize, minimize or match a target.

And these goals need to be of lower and upper limits.

For example, if I want an ejection force, and minimize that, I need to have the upper limit from that onward, I don't want to have any, any any samples, I don't want to produce tablets with ejection forces higher than 200.

And the case in the factors - these are all the responses.

But what we are going to pick from our process map that we will be tweaking around to see the effect on these responses: these are the factors.

In my case, as I showed you will be the Ac-Di-Sol, Avicel PH 102 and Pearlitol right here.

So we defined the responses and factors. And now we need to specify the model. 

At this point, I need just to let you know that we are going step-by-step because this might look complicated, but bear with me.

So we want to predict the responses, we want to understand the effect of the factors on the responses, we want to model that. But we are going to generate this model based on the data. And the data is going to be generated from the DoE. So when we need to plan our experiments, we need to know what we want to achieve with it.

For example, if I want to use a linear model, a linear model is the simplest one we can have. And this will be appropriate for any screening design. If I have a couple of factors, I don't know exactly which one or if some of them impact the responses, I might be able to run a couple of combinations between the three factors. And just to figure out which factors are relevant for that I would specify a linear model and it will suit the purpose very well. 

However, if I know a little bit more about the system, and I know there are some subject knowledge that these factors interact, I will need a non-linear model, I will need to do a screening design where we have more points and these more points will allow me to estimate interaction terms in the non-linear model.

In another situation where I know my system, and I want to find a combination between the factors that maximize or minimize my responses, I also would need a non-linear model – Quadratic terms.

And to be able to estimate this mode, as you can see from the graphics down there, you will need to acquire just the corners of the cube, but also intermediate points to be able to estimate this curvature terms.

The summary of this is that, to be able to design the experiment, you need to know how much you know about your system, you need to know the resources you have to be able to pick this one, and then we would summarize this by the selection of the model, and the design are intimately linked. So you're going to run your experiments considering what you want to understand from the data, your object, your experimental goals. 

So we have specified here in my example, that I have many effects and interaction terms in the model. I also wanted to double check to have one certain point - right there in the middle - and add five replicates to be able to estimate experimental error. So this means that I ended up with 18 rounds to make my DoE.

DoE Table

And this is what results from it-  the DoE table, your guide through experimentation.

So I'm going to have the factors, all the possible combinations between those three excipients, I'll have the slots to fill in the response values, you can see in the table as well that they are ordered in the way they are randomized, which means the computer will select a random order of runs that will average out noise from the experiment.

And you can see here as well, there are some colored dots. These colored dots are the replicates that I selected before, and I'm going to compare statistically how close this data is to each other.

The graphical representation of this DoE is shown below:

You see that they’re in a plot - as shown before, and this gray area is the mixtures that I didn't explore.

 And that's because I had a very strong constraint for Ac-Di-Sol. I just wanted to understand Ac-Di-Sol effect on the formulation in the percentage that I know or expect that to work from one to 5%. So that's why the experimental region is squashed. That was the side of the ternary diagram.

If you take a look, you're going to see that the number of points in the ternary is not the same in the table,and that's because when you had replicates, you just had those points overlapped. 

So we have the table. Now it's time to make experiments.

Making the experiment

So at this point, I would say that I have the opportunity to use the STYL’One Nano, which facilitated a lot to carry on all those experiments and replicates and also the Sotax S50, which allowed me to measure the dimensions hardness and weight of the tablet very quickly and feed it back to the STYL’One Nano to be able to calculate size, strength and solid fraction and analyze the data.

More tablets were also made to estimate friability and disintegration time.

Now we carry out the experiments, we measure the responses and we input to the table, we have a completed DoE table. 

Analysing the data - Model Workflow

So now is the second part of my presentation, where I expect to show you the steps necessary to be able to estimate a model from a DoE data set. We are going to start visualizing the data, we are going to fit the model, analyze the variance of the responses, plot the prediction profiler, and analyze the contour plot. I'm going to take you through in the next few slides.

Visualising the data

So to visualize data is a very simple but effective exercise. So we are going basically to block all the responses across the factors ranges. Our objective here is trying to visualize the main trends of the data set. So how the factors affect the responses, which trend which responses will be probably more easy to model or if we spot outliers that will require further investigation.

And then we are going to fit the model. This part will be a little bit different according to the multivariate platform you have available - in my case, I use Jump [edit: JMP]. In Jump, you would input the model terms and then run the model. And then the software will produce a series of reports very detailed reports on many statistical tests that will allow us to critically evaluate the model.

One example of this plot is the ‘actual by predicted’, in this case, you can see the red line, which is the model and the actual data points. You can see by how the line overlaps the data points, that the response in tensile strength was fitted quite well by the model.

But let's go further on our analysis.

Fit the model

So, we are going to propose here three main statistical indexes for you to evaluate how good is your model, and then we are going to use for example, R² adjusted to understand the variation of the response explained by the model - the closer to 1, the better-

 we are going to measure the signal to noise ratio of this of each response -and the higher the better as well. And we are going to test if the effect of that factor on the responses are significant or just came by random chance. In this case, we have a threshold: So we are looking for values below 0.05.

If we take a look to the data from the case study in the table below, you will find that some responses had their R² adjusted of 99. 64, we will interpret this as 99.64% of the data variation in tensile strength, the tensile strength variation was explained by the model.

We will say also that for this response, the signal to noise ratio was very high. So that's very good. And that the models were considered statistically significant for all responses. The way to use this is really to compare between the different responses or different models if you want to test them.

The way do use this really to compare the different responses or different models if you want to test them
Mariana Bezerra, DeMontfort University

So we evaluated how the model fit the data, fit each response, how good I would say they are.

So now we need to visualize really the trends, how the factors will impact the responses.

Visualise & Analyse trends


And the prediction profiler is a fantastic thing to do that. In this prediction profile, we will have on the x axis, the factors on the y axis the responses. And in the plot, the black line here represents the model, and the gray area represents the variability of this predicted value. The interesting thing about prediction profilers in Jump [edit: JMP] is that they are dynamic too. This means that we can tweak the difference factor settings, and visualize what happens with the response. These allow us to understand the different effects, the factors, a combination of factors have on the responses.

But considering that we want to understand the effect of the different combinations of factors on the different responses, we might need another tool.

And this is the contour plot.

In the contour plot, I will have my ternary diagram, but something more. So let's start by, again, remembering that the gray area here is the part of the experimental area we did not explore, due to the constraint into this interim percentage. And then here, we might need to just remember, what are the response goals? So for each response, where are the areas where I don't want to get from the product. And these are projected on the experimental design based on the model curves. So these areas colored in red, green, blue, and orange means all the combinations of those three excipients, which predict the tablets will produce responses that I really don't want to have.

And in opposition to that, in the white region, we are going to have all the combinations of those factors, which conditions are expected to produce formulations that comply with the response goals - the product QTTPs.


So, our conclusions for the day are:

First DoE provides a control set of tests designed to model and explore the relationship between factors and responses.

Selecting the type of design depends on the level [of understanding] one has on the system, the experiment objectives and the resources available.

And finally, about closing the case study, a design space was obtained where the measured responses met the QTTP.

So just to finalize, what will be the next steps of the study. Now that we understood how the formulation impacts the tablet CQAs, we might start considering how those critical process parameters will impact the tablet CQAs. We might get a screening design, to try just to scan from a list of those factors, which ones really affect the CQAs. And we might take a sequential approach to these experiments, which means we are going to get a very simple design, just to select which ones are the better, and then carry on with the ones that are significant, and try to get more experiments that are more complex, but also we generate models that will allow us to understand interactions and maybe optimization.

So just to finalize here, I would just like to acknowledge the huge support I have been receiving from MEDELPHARM, from SOTAX, to be able to use those very nice instruments to carry on my studies, all the academic and industrial collaborators that helped give insight and provided that incredible learning experience. And a special thanks to my supervisor, Professor Walkiria Schlindwein that had a huge impact on all my research and the preparation of these slides for you today. Thank you and I welcome your questions on the Q&A.

[Bruno Leclercq - MEDELPHARM] Okay, thank you Mariana for this very nice presentation. We will see Mariana back at the end of the session for the Q&A. And now we'll switch to a live demo. Let me just share my screen.

MEDELPHARM’S STYL'One Nano & Alix software

Demonstration of Alix integrated software

I will start the live demo by showing how a process parameter can be easily set up - as you've seen, during Mariana’s work, she modified some of the parameters, and I will show you how you can do that in practice.

The ALIX software has been developed for controlling the STYL’One Nano, collecting data during the compression and analyzing the results. As you will see the software has been designed to guide you naturally through the different steps that you need to set up your experiment.

Alix software: choose your tooling

STYL’One Nano & Alix software: choose your tooling

On the screen, what you see is that we have tooling that's been installed – Euro B flat base 11-28. This is a tooling which is typically used for material characterization. However, if you actually need to use other tooling for your development work, you can choose it from the list of tooling that you already entered into your into your database. For example, if you want to use a punch which is oblong, you can just select it and then this will be the tooling you will be using for your experiment.

But if we just go back to the tooling that was already installed, Euro B - this is the one that we're going to use. Obviously, when you have determine your tooling, your second step is to choose your powder. We already entered one power for today for the webinar, but if you need - like in Mariana’s work she has been working with different formulations, if you want to put a new formulation into the system, you can just change the powder and introduce a new powder, either from a list that you have in your drop-down menu or you can type it to introduce a new powder.

When you choose a tolling and the powder you want to use in your experiment, the next step obviously, would be to get the kind of compression cycle you want to use. In this software we have different cycles you can use. We have the V-shape, the extended dwell- time, or the small rotary tablet press. And to be clear for you, we’re going to launch a small video showing you the different profiles that can be used. We see the V-shape profile, we see the pre- and main compression profile, and last the square profile to measure visco-plasticity.

Access STYL'One Nano video HERE

Now let’s choose a profile mimicking a small rotary tablet press – in this case one of the process parameters you may want to look at is the actual speed - in this case we can mimic different speeds of the rotary tablet - ranging from one revolution per minute to 70 revolution per minute. In this one, we're going to choose, for example 20 revolutions per minute, for our experiment.

The next step would be that we need to adjust the weights to achieve a certain target weight. And to do that, we just need to modify the filling height, in this case, we choose 12 mm. But this would need to be varied depending on the target weight you want to achieve.

Also, if needed, you can also modify the overfill. When you have kind of sorted your dosage, the next step would be is that you need to decide which pre- compression and main- compression you want to apply. In this case, we decided to have a pre-compression of 1 kN and the main compression of 20 kN. Obviously, if you don't want to do an experiment with pre-compression, you can just remove the pre-compression and you just add the main compression. For the main compression, you can target either force, but you could also target in pressure, or even if needed compression, so the thickness of your targets. But in our case, let's target mid-compression of 20 kN, and lets add a pre compression of let's say, 1 kN. When this is done we can actually do our experiment, at this experiment that you could be running will be  mimicking rotary tabbing press with 20 RPM, filling height of 12 mm, pre-compression of 1 and main compression of 20.

And as you can see, you can easily kind of separate all the different parameters to do your experiment, and this could be done quite quickly. And that was actually done by Marianna when she did the different DoEs. Now I will turn to Adrien Pelloux, the Science Lab Manager based in our application lab in Beynost in France, and he will do a short demo of the STYL’One Nano. Up to you Adrien.

Demonstration of STYL'One Nano & Alix software

[Adrien Pelloux - MEDELPHARM]

Thank you Bruno for the nice introduction. So welcome everybody to MEDELPHARM Science Lab, where we test prototypes for the engineering team and where we make some trials for our customers - compaction trials and powder characterization trials. So feel free to contact us if you have any issues on your powders or on tableting. 

So here we have the typical setup you have to have if you want to perform experiments with the STYL’One Nano: we have the STYL’One Nano, we have the computer that drives the equipment, and where we retrieve data from the STYL’One Nano, we have the SOTAX ST50, the same as the one used by Mariana during her studies. And here we have also a vacuum cleaner where we can plug the vacuum directly to the STYL’One Nano and it will avoid the spread of the material if you want to work with active products. 

Typical set-up STYL'One Nano in lab

Typical set-up of the STYL'One Nano in a lab

So here let's show you a small manufacturing of samples. So here the STYL’One is equipped with some standard oblong tooling. It could be the same that you can use on your rotary press/ manufacturing unit. And here we have the advanced gravity feed shoe where we have put some powder inside, and so as you can see the door is equipped with security feature, that you need to reset if you want to use the equipment again. 

So let's make a few tablets. Here you'll see first the feeding part, with the pre-compaction, main compaction and ejection – [again] filling, pre-compaction, main compaction, and ejection. Here we are set the small rotary press profile at 10 RPM, it's quite a slow speed, but it's for you to show how the machine works. But keep in mind that the STYL’One Nano is able to work seven times faster, and to work like a small rotary press at its maximum speed. 

 So now the manufacturing of the samples is over, we can retrieve the tablet, but let's have a look at the software to know what we have.

So here the software has retrieved both the displacement data from the displacement sensors, and also the data from the force sensors. And so from this screen, we can monitor the evolution of the ejection forces during the trial. And we can also monitor the evolution of the lower punch force as an example.

From there, you can also have a look on the compression timeline, on how it has been made during the experiment. So in yellow, you have the movement of the lower punch, and in blue you as the movement of the upper punch.

So here we have the feeding part, here we have the pre-compaction, here we have the main-compaction, and here we have the ejection signal. If you want to make a close up on one part of the cycle, that's very easy, and you can see how the punch forces is evolving during the cycle. 

Another thing that people enjoy is the evolution of the force according to the distance between the punch: that is what we call the energy plots. But once you have made your experiment, you test it with your tablet tester, and the software is able to retrieve the data from the tablet tester and to allocate it directly to tablets but have just been manufactured. So let's have a look on how it works when you have performed a lot of experiments like Mariana did.

Data with STYL'One Nano

Easily generate the data you need with STYL’One Nano & Alix software

We have the project, and we create some study. And so here we have all the runs with all the settings that have been used, and also the data retrieved from the tablet tester. And when you are here, you can create some standard charts, and you will see the same metrics than the one that used Marianna used as a response in her DoE, like the tensile strength here, or the solid fraction as a function of the lower punch peak pressure. Or you can monitor the ejection forces according to the lower punch force.

If you want to create more advanced charts, you can create some custom charts where you just have to set your y and your x of your chart, and you will draw the charts automatically.

And those charts can be directly exported as a picture to be uploaded on your reports, so it's very convenient when you make a lot of experiments and to have interesting data from the same experiments.

So I will let Bruno continue on this webinar.

Key Learnings

[Bruno Leclercq - MEDELPHARM]

Okay, thank you Adrien for this live demo, especially because it was the first we actually did during a webinar. Now, let me conclude by sharing my screen.

As a key learning:

- DoE should be a key element in your development process.

- No need to rush into experiments. Take your time to carefully think about what you want to achieve and maybe discuss with some of your colleagues to see what should be an ideal Design of Experiment

- And DoE outcome will support your decision-making process 

- The STYL’One Nano , a key technology to explore your design space by changing your product and your process parameters and it’s fast results to drive your decision making.

And this is tied to quick product and tooling changeover, intuitive software for fast, experiment set-up and build-in editing graphs, and reports. 

It’s now time to start the Q&A session. Don't forget that you can still ask a questions in the Questions tab. 

We can ask Mariana and Adrien to come on stage.

Q&A Session

[Bruno Leclercq - MEDELPHARM] 

Can we have more than 3 factors in a mixture design?

[Mariana Bezerra – De Montfort University]

Yes, that's a very good question Bruno -   as I just showed, mixture design with just three components is the minimal number you can use the mixture designs, because it's the easiest way to visualize these experimental region. However, you can use mixture designs with a couple of more variables with more factors.

So you can have a formulation of seven components, and you can have a mixture design with several components. The only thing to have in mind there that the more number of factors you have, the more complex the experimental design, and the higher the number of experiments to make. So you would try to just try to estimate a linear model, just to screen from that seven factors, which ones most impact the responses and then carry on a more complex design.

[Mariana Bezerra – De Montfort University]

If I could, and there was a question or two questions in the chat that I think might be relevant Bruno.

So a lot of people ask me what the software I used to analyze the data.

I used one single subject to create the design and analyze the data, and the software is called JUMP, which is JMP.

And another question is, if I could mix process parameters with formulation parameters in a single design?

And this question is very interesting, because you might think that you want just to put everything in one single DoE, and you can do that.

But the problem is – [how can we] understand those different factors and the responses you might get.

So we are going to mix in this hybrid model dependent terms and independent terms. The independent terms are from the process parameters and the dependent terms, because the formulation parameters will compensate each other, and having all those factors in one single model, you increase the complexity of your model, and will require more runs, while number of different combinations of those factors to be estimated. So in practical terms, what we would advise is you do a sequential approach, as I proposed here, you first evaluate the process or first evaluate the formulation, or maybe with the best formulation, you carry on this formulation, to experiment concerning the process parameters.

 [Bruno Leclercq - MEDELPHARM] 

Okay, thanks Mariana, I'm sorry for the interruption. So we have another question:

Why is the dissolution of tablets not included as a CQA? 

[Mariana Bezerra – De Montfort University]

That's a good question. So my point with this first case study was just trying to understand what works, what doesn't work in terms of the diluents in compaction aid system. And I wanted to have very quick results to be able just to put aside which didn't work. So the point here of this solution is now taking this formulation that work and caring for another DoE, which instead of disintegration time, I will have the dissolution as a response. Thank you for the question.

 [Bruno Leclercq - MEDELPHARM] 

Is it advisable to mix formula and process factor in the DoE?

 [Mariana Bezerra – De Montfort University]

Yes, that's the question that I just answered, if you should have process parameters and formulation parameters in the same, as I said, you can do that, but be mindful that you might generate a very complex model, very rich, very difficult to estimate. And these require a large number of experiments to be able to achieve.

[Bruno Leclercq - MEDELPHARM] 

Is there any plan to integrate DoE features in the STYL'One Alix?

[Adrien Pelloux – MEDELPHARM & Bruno Leclercq - MEDELPHARM]

Not yet – the answer is not yet, however, as you’ve seen in the software, you can easily change the different parameters, then if you want to play with process parameters like speed, or pre-compression & main-compression. This can be done quite easily, and that’s what you did Mariana during the second part of your work.

[Mariana Bezerra – De Montfort University]

Yes, was quite easy, for example, to swap between one formulation and the other, to be able to conduct the randomized orders of my DOE, it was very simple to manipulate and get the experiments going.

[Bruno Leclercq - MEDELPHARM]

There's another question: which method was used to define true density of mixture? I guess maybe Adrien can answer to that one.

[Adrien Pelloux – MEDELPHARM]

Yes, at MEDELPHARM Science Lab, we perform the measurement of the true density with a helium pycnometer from QuicPik. And it's a very nice device, you will need just a few grams of your material and you as a result within 20 minutes.

 [Mariana Bezerra – De Montfort University]

There was one question, if you allow me to just jump in, about the contour profile, can I just share my screen? Somebody asked me what these different colored regions means.

 And just to emphasize here, they represent the different responses, so each color corresponds to one response. So if I pick up, for example, ejection force, the area of color in here means that the formulations with this high proportion of Pearlitol in the formulation resulted in an ejection force higher than 200 Newtons. So if I want to avoid achieving this ejection force, I might avoid having formulations with high proportions of the Pearlitol.

That is the basic interpretation of this.

[Bruno Leclercq - MEDELPHARM]

Okay. And maybe one last question, I guess one question was addressed to you Mariana: With the DoE, how do you overcome internal variation error from the overall design?

[Mariana Bezerra – De Montfort University]

Yeah, I tried to, to estimate the variation on the experiments using that replicate runs. Depending on how many runs you can afford to make, you can actually replicate the whole design, but that will depend on how you're going to or the facilities, the resource, you have to run the experiments. And you might have other options in the statistical software to evaluate the experimental variability, like the root mean squared error or some things like that. But I could not go in deep into the analysis of the data because we just didn't have enough time.

[Bruno Leclercq - MEDELPHARM]

Okay, thank you Mariana. I think we, we couldn't answer all the question because it’s already quite late, but we will try to answer your questions by email.


[Bruno Leclercq - MEDELPHARM]

And just to finish the presentation:

I would like to thank everybody for your attendance and your good questions. We never get so many question. And thank you Mariana for the hard work at this very nice presentation. And thank you MEDELPHARM team for the organization.

The Q&A session will continue on my MySTYL’One forum with Mariana, myself and Adrien. 

This presentation and the replay will be available on MySTYL’One.com. [MySTYL'One members, access video replay HERE]

Stay tuned for our next event – MySTYL’One live session #4 will be held next June and we will keep you updated.

And be sure to check the STYL’One Nano brochure in the resource section and to visit our website.

We at MEDELPHARM wish you a pleasant day, and do not hesitate to contact us directly for any additional questions or comments. Take care and thank you for listening.

[Edit: End of transcript]

Want to read more?