Continuous improvement Process Verification (CiPV)

What you see here is our SmartFactory Rx dashboard that contains a number of different tiles with various functionality that our platform offers. Where I would like to start today is in our overall factory map which is really the high-level view. Oftentimes notifications are received or high-level managers can be monitoring the equipment through this high-level factory map.

The notification that we received was related to a production bioreactor and as you can see it’s a process health issue. The indicator here shows that there is a warning that has been triggered and when we select that particular equipment it brings us to a more detailed page of that equipment and what the warning was. The overall health that is indicating here is that there was an issue that has resulted in a prescriptive action.

Now, this is prescriptive action was indicating that the initial titer prediction the final titer that is an early prediction of the batch was low outside of limits. And that there is a machine learning optimization model that is calculating optimizing the time for a specific feeding that would help to improve that potential yield to a higher level. So we can actually monitor and see the batches, each one of these points here is actually a batch that has been run over time and this is a titer prediction, early prediction and then an optimized value that we would see potentially if this prescriptive action was taken. The portal enables us to actually delve a little deeper into what might be happening within the process.

Each of these indicators shows green if it’s within limits and the inside of this indicator will turn more yellow as the point gets closer to an actual limit boundary. So overall the pH was showing that there were some issues within the process. We see that there’s some very distinct phases within the process of the pH that it began before the pH shift.

There was actually a multivariate model that’s running showing that there was an outlier related to this particular batch and this is what helped to also trigger the potential recalculation or optimization of when that feeding should be. The main contributors as we can click on from this model and see that the cell density was lower than normal than the other batches. We also noticed that the second contributor here is the pH range and near the beginning of the batch and we see that the pH went through a higher range so definitely something to do with the pH control resulting in also a lower cell density for the amount of change in pH that we normally observe.

And those two things combined definitely feed into the machine learning algorithm that helps us to optimize based on culture performance and not just based on a strict time. So we’re optimizing the feeding to be able to adapt to the culture performance at hand. The ability to then view what the quality metrics might be from the QC lab in terms of that, what that titer was at the very end of the batch, it can be viewed through the CiPV screen.

So when we do go to the CiPV which is the continuous process verification we see some of the similar results here this multivariate model for that particular batch. We have other QC results that can be related to that batch whether it’s purity or potency. We have the, again the contributors to that multivariate model.

We’re able to then identify in a very succinct way of the relationship between a multivariate model and a number of different parameters in these SPC charts. And we do have the ability here to see what the actual final titer was from the quality control lab that it was close to 0.74 grams per liter. We are able to view some, various statistics here, in terms of, you know, the distribution of the data.

We have another visualization of the data in terms of a spider chart that helps us to see maybe if there’s any type of interesting progression of points if we start to follow the data in this manner. We also have some charts here showing that the early prediction of the batch was close to 0.4 grams per liter. The optimized prediction was showing that if we fed at the adaptive time we would look to be able to achieve a titer of 0.8 grams per liter and then the actual titer which was 0.7. So these things are also overlaid in this bar chart with the actual overall PCA analysis showing that there was some variation of this batch compared to all the others.

And so overall we get to have a quick view as far as each batch. And as I select a new batch here we see that these parameters change, the values change. We’re able to monitor this batch as it’s progressing or these batches over the course of time.

We have some very important information here in terms of the step that we’re analyzing here, what equipment it was, and then the individual metrics that we’re following or a product. And you can see as we again select a new batch we see that these the name of the batch or the batch id and again all the data surrounding that really helps us to understand the variability within potentially our process and how we can maybe improve upon it. And that’s why we refer to it as our Continuous improvement Process Verification.

When we’re combining these statistical process control charts with our machine learning algorithms and optimization routines as well as our multivariate models, all this comes together to give us the information we need to really improve upon our process. Some of the other things just to review here is that we not only deal with large molecule manufacturing lines where we have upstream and primary recovery and purification, but we also do a lot of small molecule manufacturing. And so we can see that there’s a number of different ways that we can interact or different lines that we can have.

We can even make this more of a site map and have a number of different, you know, lines within one’s particular site. So a lot of ways that we can get into the data. We have different metrics that we can be tracking, whether it’s the equipment health or the process health.

Sometimes OEE is oftentimes a metric or a KPI that we would want to be able to view across whether it’s a line or a facility. Overall, this really helps to get that high-level view to know exactly where to drill down into. From that, we also are able to visualize actually raw data.

So maybe we noticed that the cell density was different in that one batch. So we can actually pull that batch up, want look at the actual trend of that batch, maybe overlay it with the amount of CO2 that’s being used during the course of that batch. So we can actually see, you know, the various trace data as well. When we start thinking about how all this data is being integrated, and I think this is a very important component, is that we have a number of different data connectors that are easily accessible or used that have already been configured, already been developed. And all we need to do is select that particular data connector, and make the correct links to that data, and then be able to then stream that data in. So you can see we have two data connectors here that are configured.

We can also look at the monitoring of those data connectors. So we see that there are a number of different logs. If we were to lose that connection, this oftentimes indicates or results in a notification to make sure that data is not being lost into our system.

So these connectors, we also configure with the interval, the polling interval. We again have very standard connectors for our ODBC type databases, which are more LIMS types of databases, OPC UA. We also have, we support the OSIsoft ACID framework and also just regular OSIsoft PI. We do have a spreadsheet flat file as well that we can actually connect to or link to and be pulling in data from flat files.

So all very important for us to be able to configure the equipment that enables us to build those models. And so you see when we look over here at our equipment list, we have a number of different equipment. So we can go back to that bioreactor that we were just looking at.

The bioreactor itself, there’s a mapping that is done for all these particular tags. So you can see that we’re able to navigate or search through the different data sources. And this is where we start to integrate across our data sources, our data silos potentially, and really build stronger models.

So we might be able to search here a particular bioreactor, visualize all these tags, and it’s just a matter of initially mapping these tags from various data sources in order to be able to bring it all together into one data model. This data model or process type is what we call it, is the bioreactor process type. But we do have a number of different data models already input into our system, whether it’s centrifuges, chromatography systems, the various commonly used granulators, tablet press encapsulation.

So they’re commonly used bio, pharmaceutical, and pharmaceutical equipment. We do have the data models for those. And oftentimes we also are incorporating more than just the data required for that equipment.

So we’re incorporating utilities, we’re incorporating maybe room temperatures. And so all that from those different data types and different sources really enables us to build very strong robust models, whether it’s for the equipment performance and monitoring for predictive maintenance or for the process health and being able to predict process parameters and critical process attributes in order to make changes within the process and improve it. The last feature that I’d like to demonstrate is the ability to import data manually and using this manual import.

What we’ll show here is that we’re going to be importing some metrics here related to the host cell protein. We can see that there is no host cell protein metric right now that has been loaded into the system. And so the idea is to take this flat file, which contains some host cell protein data for the batches that we have run in the system, and load this into our dashboard so we can actually visualize it along with the other quality metrics and also batch parameters or critical process parameters.

So this is accomplished through our manual import where we would, all we have to do is navigate to the file. We have some host cell protein data that we would open, and then once that file is loaded, it’s read and loaded, we would choose to import it into the equipment metrics or into the CiPV dashboard. So once the file has been loaded from our folder, it’s only left to decide whether we want to import that into the equipment health metrics or into the CiPV metrics.

So once we’ve loaded the file that contains the data that we’d like to populate into our dashboard, the next step is to decide if we are going to import it into the equipment health metrics or into the batch metrics for the CiPV. In this case, we’re going to, once we’ve loaded the file, as we can see here, the next step is to decide whether we want to import that data into our equipment health metrics or into our batch metrics for the CiPV dashboard. In this case, it will be through CiPV, and you can see that the import then is successful.

We can go back into our CiPV dashboard now, and we are able then to pull up our search for any metrics to do with host cell protein. We have that metric that has been loaded. We just select it, and it gets added to the bottom of the chart.

So these are just stacked on top of each other. So we can see now that we do have the percentage of host cell protein for each batch. It’s a very easy way to incorporate the data.

It automatically populates into the various distribution or the charts of different views of the data, and we have the ability to filter based on particular products, the equipment, maybe the equipment type, or even down to the batch level to bring in specific batches. Any one of these, once we’ve built this sort of configuration or this report, we can then download that data, the CSV, or we can actually share the link that will always bring us to the most recent version of this. So if there’s new batches or new data comes in, it would automatically populate into this view, being able to display the most recent 12 or the most recent 20, whatever the settings were when we created this actual link.

So very easy to configure these types of reports, download them, share them. This is, I think, a very strong functionality in terms of monitoring our batches. I want to thank you so much for your time, and I’m looking forward to hearing from you in the future.

So with that, I would like to thank you so much for your time. And so that concludes our demo for today, and I would like to thank you so much for your time, and we look forward to hearing from you in the future concerning any needs that you might have, where we can collaborate with you and help you move forward in your digital journey path.