The biopharmaceutical process is a relatively complex process, such as vaccines, antibiotics, cell-based treatments, etc. Biopharmaceutical production processes, including fermentation or cell reactors, involve controllable parameters such as pH and temperature, as well as a series of output measurements such as bioburden and impurities. In addition, some other uncontrollable factors can also affect product quality and the repeatability of the production process. These three elements generate a large amount of data throughout the entire production process of a drug.
Biopharm ProcessBiopharm ProcessBiopharm ProcessBiopharm ProcessBiopharm Process
So, what strategies and ways of thinking do we use to process this data? Today we will introduce three main tools: DOE (Design of Experiments), MVDA (Multivariable Data Analysis, also called Multivariable Data Analysis), and the combination of PAT (Process Analysis Technology) and MVDA.
As we all know, QbD (Quality by Design) has been introduced to China more than ten years ago. JMP has already added QbD functions to the software platform and provides the industry's first-class, complete and powerful DOE tools to improve pharmaceutical manufacturing. Provide comprehensive support for industry process development.
Whether in the chemical pharmaceutical or biopharmaceutical industries, immediate release is an ideal. Since immediate release currently faces some challenges, a common recommendation in the industry is to use parameter release, which all involves the application of PAT technology. The result of both DOE and PAT technologies is the generation of large amounts of data. Without proper analysis of the data, the application of these two technologies will not be possible.
DOE , a top priority in drug development
First, let's look at DOE experimental design. DOE is a very important platform in the drug development and production process. A large amount of data will be generated during screening, characterization research, optimization and verification. Through the analysis of multi-variable correlations, a lot of data will be generated.
DOE WorkflowDOE WorkflowDOE WorkflowDOE WorkflowDOE Workflow
More importantly, DOE provides a good platform for the next step of multivariate data analysis. When the QbD concept was proposed in the early days, DOE was a very advanced and useful concept. The design space can be found through DOE, and the control space and control strategy can be found through the design space. In the process of control, the most important thing is to understand the correlation between data, the source of variables, the causal relationship between variables, etc. In this case, DOE is not a node, but a starting point. It allows us to continue to understand the causes of process variables and find space. So, what is the purpose of finding space? It's control. So throughout the process, we need monitoring technology, and more importantly, control strategies. When you formulate critical quality attributes (CQA), when you find critical process parameters (CPP), as long as you control the process parameters, the quality must be in line with expectations (achieve quality goals), thus creating the basis for immediate release, parameter release and true Lay the foundation for data transmission. Therefore, more importantly, through multivariate data analysis, the final process is controlled, and therefore the quality is controlled.
Process control starts with process monitoring. There are several important strategies (tools) for process monitoring: MSPC (multivariable statistical process control), BSPC (batch statistical process control), and PAT process analysis technology.
- BSPC is used for trend analysis by determining whether the production batch is the same as the expected batch.
- MSPC is more about observing the relationship between key process parameters and key quality attributes to see whether the key quality requirements are met.
- The PAT process analysis technology combined with multi-variable data analysis can achieve Real Time Prediction (real-time prediction). At present, we are more about diagnosing problems after they occur, but in the future we will be more about making predictions, that is, we can already predict quality, increase accuracy, and improve production efficiency during the production process, manufacturing and process development. , thereby truly realizing the concept of quality by design (QbD).
Through the application of these technologies and strategies, future technicians will be able to predict the quality of products during the production process. This is also in line with the idea of "from post-supervision to process supervision" launched by the China Food and Drug Administration. NMPA also The overall strategic idea of smart supervision was launched, and the above technologies are all needed in this supervision process.
Next, let’s look at why multivariate analysis is so important?
Multivariate Analysis, full knowledge of process deviations
In multi-variable data analysis, we can not only see the relationship between variables, but also the reasons for deviations between variables. As a powerful tool, multi-variable data analysis can be observed in the early stages of production. to the root cause of the deviation.
The figure below shows the difference between multivariate data analysis and univariate data analysis.
In this example, the temperature and pH appear to be within control limits. However, when two images are plotted against each other in a multivariate format, a trend can be observed. One of the data points is far away from the main data group. This free data point is not displayed in the single variable control chart. The main reason for this is that in most cases, process parameters are related to each other. Because these data are all related to each other, multivariate data analysis can provide a powerful tool for observing the complete picture. Through multivariate data analysis, process data for a complete production batch can be summarized into a single trend. At the same time, each data point contains information about each process variable. This information can be used to conduct in-depth, more focused analysis and problem solving on any particular variable. We can also generate multivariable plots that include limits and can be used to track a production batch through its entire process.
Let’s look at another case study of multivariate data analysis. To further understand the application of multivariate data analysis, the scientists analyzed published technical data on the production of penicillin. This case is data from 13 production batches. Among them, 10 batches produced normally released and qualified products, which were used to make correction models. In addition, three batches failed, and this data was used to test and validate the process model. So, why were these 3 batches not released by QC? Why does the deviation occur?
In this case, we can see that the process variables are very complete, and the quality indicator of the final product is biomass. Biomass is our main observation data. Through multi-variable data analysis, the structure of the data and various correlations are clearly displayed in the "Loadings Plot", which is used to show the correlation between process variables.
Loading PlotLoading PlotLoading PlotLoading PlotLoading Plot
It can be found from the figure that when the concentration of glucose is very low, the concentration of other metabolites is very high. The correlation of other process variables cannot be seen in univariate analysis, let alone the relationship between metabolites, nutrients and biomass. Why is biomass important? Because it is closely related to product quality. On the loading diagram of multivariate analysis, we can intuitively see the correlation between biomass, metabolites, and nutrients.
Next, through the analysis of 10 batches of data, an initial model is generated, and a multivariable control chart is displayed. In addition, 3 test batches for detection were used to see whether the presence of any bias could be detected by using this model. As can be seen from testing, deviations are easily observed.
As can be seen from the picture above:
- The 10 batches that were released normally were basically within the control range;
- In the early stages of the chemical reaction in the production process, the differences between batches are relatively large. As the process proceeds, all the variables are very concentrated at the end of the production stage, and each batch is within the limit. The deviation Not very big. This is where the so-called gold standard is formed.
Through these 10 batches, we can observe the trend of the entire process parameters and the trajectory of deviations. At the same time, you can also see the occurrence of variables between early production and late production.
Let’s take a look at the three batches that did not meet the quality standards. Before conducting multivariate analysis, we usually only know that these three batches do not meet the quality standards, but when did problems occur in these three batches? Why is there a problem? These can only be answered through multivariate analysis.
Through multivariate analysis, it can be found that red, green, and blue are the three rejected batches, and the dotted line is the control range.
- Red and green are outside the control range, and a batch of blue ones are within the control limit. Through the interactive clicking of JMP software, it can be found that the red color is an abnormality of a certain element, and the abnormality occurred in the early stage of production, but has returned to normal in the late stage of the process. The histogram below shows in more detail which variable caused the deviation in the batch with the red line. Without multivariate data analysis, we can see that metabolites and the like were normal during the production process, but in fact mutations occurred very early in the production process, which resulted in the final rejection of these three batches.
- The green batch was already outside the process control range in the early stages of production. Through the display of multi-variable analysis, we can see that in the actual production process, the generation of many deviations and root causes has already started in the early stages of production. If we only do Ending Point detection, we cannot detect these. The origins of variations, specifically the timing of their emergence. Similarly, if you use offline testing at the end, you will find that the results are very good, but in fact, deviations in nutritional mechanisms and metabolites have already appeared in the early stages of production.
As can be seen from the above, multi-variable data analysis technology can help us find deviations early in the production process, allowing us to find ideas and methods to solve problems in time. On the contrary, without the support of these data analysis technologies, the only information we can obtain is that the yield decreases late in the production process. At the same time, the combination of DOE and multivariate analysis technology can change the data analysis strategy of biopharmaceutical companies, increase the speed of drug development, and significantly reduce related costs.
PAT Overlay multivariate analysis to make instant analysis easier
The third technology - PAT technology, especially the combination of PAT and multivariate analysis technology, is very powerful. There are many PAT technologies. Usually, samples are taken during the production process. After taking the samples, they go to the QC laboratory or IPC laboratory for testing, and then return after testing.
The application of PAT technology can enable the entire analysis technology to be completed online. Generally speaking, there are many advantages:
- The key advantage of online analysis is the ability to control critical process parameters (CPP). Why control CPP? Because every critical process parameter and critical quality attribute is related.
- Instant predictions can be achieved. In the above example, if PAT is applied for real-time prediction, deviations in these three batches can already be found in the early stages of production, so that it can be better predicted whether there are problems with these three batches at the end and whether they can be released, etc. .
- Multivariate data analysis can generate models of correlations and correlations.
- PAT has a wide range of applications and can be used in all aspects of drug production development and drug production processes, which is much higher than Ending Point testing. It can also detect many elements that cannot be tested online without using PAT technology, including nutrition. Measurement of element levels, identification of impurities, etc.
The combination of multi-variable data analysis and PAT online analysis technology can form true process monitoring and process control. Advanced data analysis software such as JMP is not only an independent statistical tool, but more importantly, how companies apply the technical platform and tools provided by the software to form a powerful and overall solution to achieve the goal for the pharmaceutical industry. The real technology now is pushed to the future of instant release and continuous production to provide powerful data and technical support, thereby ultimately improving product quality.
