Data partnerships are central for integrating AI in CGT manufacturing
Across sectors, AI is synonymous with efficiency innovations, and it continues to gain momentum. The pharmaceutical industry has a reputation as a late adopter for new technologies, but even this field is now buzzing over AI’s potential to accelerate advances.
The truth is that AI, which means many different things, has been impacting our work for decades. Recently, it has been deployed behind-the-scenes for improving clinical trials and digitizing paper records, and more prominently, for early-stage drug and diagnostic development.
There remains huge untapped potential to increase efficiency in manufacturing and improve outcomes, and nowhere is that more pressing than for cell and gene therapies (CGTs). The space is in a tenuous position due to the current inability to make commercial therapies affordable, and the difficulties are linked directly to the way they are made. And because CGT technologies have largely grown out of bespoke, manual, and paper-based processes, established in the early stages of drug product development., As an industry, we are just now sorting out the instrumentation and data infrastructure that we will need to support AI advances.
Potential brimming outside a closing window
CGTs approved so far are one-time therapies that are often curative for intractable fatal diseases, like late-stage blood cancers, as well as for dangerous chronic diseases like sickle cell disease (SCD). The horizon is bright, with thousands of CGT clinical trials ongoing that underscore the tremendous promise to expand into other disease areas.
Unfortunately, they are cumbersome, time-consuming, and expensive to make compared with other modalities that allow for batch manufacturing. CGTs are made from human cells, which means a significant amount of complexity and variability for each dose. Commercial therapy – which can cost from hundreds of thousands to millions of dollars – are not affordable enough for the health system to offer them to all eligible patients.
Many in the industry are working to solve the issues, but there is a risk of CGT missing its window. Big pharma companies are getting frustrated and investors seem to be losing some faith. There is a risk that if we take too long to solve commercial feasibility, it could pull the rug out from under the entire field.
The field is broadly engaged in the process of identifying the most important manufacturing parameters. It will likely be different for each type of CGT, and maybe for each therapy, in some measure. As we build this understanding, it becomes possible to learn how best to manipulate aspects of manufacturing like cell growth in real time. The goals include faster cell expansion rates, higher quality of therapies, lower process failure rates, less time to get a dose to patients, and simpler scalability.
AI is uniquely suited to the task. It can help optimize and standardize processes, guiding decision-making at crucial steps in the manufacturing process. When coupled with automation, we expect it will reduce ‘vein-to-vein’ time, lower the reliance on manual labor, free up manufacturing clean room floor space, and make processes more flexible and robust – all key steps for lowering the cost of production.
Data challenges
AI, of course, has its limitations, the algorithms can only ever be as good as the data they have to learn from. That makes the quality and quantity of data critical, which in this space means existing data has to be made more usable, and more data will need to be generated.
One major roadblock is data siloing. Therapeutics developers naturally need to guard their intellectual property (IP), but some siloing is unintentional – for example, paper records that have not been digitized. Another issue is internal siloing due to incompatible systems, either within a company or between them. Enabling technology companies must standardize in order to integrate, which also means data sources have to talk to each other. Data housing is a related concern.
In most cases, developers have more to gain through standardization than they stand to lose by sharing data, and it can be done in ways that do not betray company secrets. Democratization of data is necessary for the industry’s development, as CGT is a rapidly growing field that relies on cutting-edge technology stemming from very recent scientific discovery.
Hardware partnerships
There is only one solution for these issues: collaboration. For many companies, AI is still a novel technology. As they develop the internal expertise required to leverage it, they will require knowledge exchanges with experienced partners to lay the groundwork.
These partnerships will take many forms. Enabling technology companies are starting to work together with data integration in mind. In our case, we have identified several kinds of opportunities that allow us to leverage our technology in ways that can benefit the broader community through better data.
One is increasing the amount and types of data that can be generated by and with our platforms. This is most important during the cell expansion process, the most time-consuming and expensive part of making a cell-based therapy. That is where AI can make a large contribution to optimizing processes, leading to faster manufacturing and lower costs.
Last year, we partnered with 908 Devices, which has technology that enables on-line monitoring of critical process parameters, and integrated it with our CGT cell expansion platform. Now, users can automatically sample glucose and lactate, ingredients needed to control the growth of expanding therapeutic cells in order to make the right-sized dose.
We recently announced a similar partnership with Nova Biomedical. Their device does on-line sampling for additional characteristics, including parameters like cell viability and cell density that could indicate how well a process is working.
Both of these point to another key piece of the AI-enabling puzzle, which is automation. As better data from sensors allows AI to quickly guide decision-making, automated technology then will allow rapid execution of changes, with less in-person time spent in cleanrooms.
Earlier this year, we published the results of a study done with Charles River Labs, which combined another platform of ours, for automated fill-and-finish, with their T-cell workflow. It resulted in increased capacity, meaning more developers can automate the process for more kinds of CGTs. We can envision a time when these systems and their connections to other parts of the workflow can be optimized with AI.
In each of these cases, the integration of technologies is also about integrating the data flows. On-line monitoring increases the number of possible measurements, enabling companies to more easily connect various inputs and ongoing process parameters with outcomes. In the future, in-line monitoring will increase that potential even more.
Generating more data will allow developers to deploy machine learning algorithms that can parse the relationships between the many variable parameters, unlocking new levels of process optimization. In conjunction with automation, AI will make manufacturing decision making and execution easier for multiple doses simultaneously, in various locations.
Data partnerships
Technology partnerships represent part of the solution; the other is data partnerships.
Imagine a different ecosphere, where instead of leaving each developer to optimize processes through years of data generation, general learnings could be shared across the industry. Developers could spend their time and resources innovating, building on what’s unique about their therapies instead of digging for answers to problems that someone else has already solved. This is only possible through data partnerships among developers, as well as between developers and enabling tech companies.
It will require a shift in thinking toward compatibility. Companies will need to develop data infrastructures that include robust data capture in every phase of the manufacturing, from donor or patient screening through patient infusion.
Companies have legitimate concerns about who owns the intellectual property connected to the data. Particularly because CGTs are developed from donor- and patient-derived cells, data privacy and the ethics of AI usage must be carefully addressed. But nobody can thrive without structured sharing.
These hurdles are manageable, and the industry is motivated to overcome them. Regulators will also need to have an important role in enabling the future of data sharing.
Enabling technology companies, like ourselves, will play a central role, too. We can support data integration and housing, allowing our partners to conduct complex analyses and leverage AI for better decision-making. By gleaning insights from across the industry that can be securely shared when appropriate, we have the chance to be the rising tide that lifts all boats.
We can also enable more contributions from academics, who are already important partners. Of course, they are the source of most new CGT therapies and biotechnologies. They also share an interest in using large datasets to improve processes and patient outcomes.
We recently worked with academics at Boston Children’s Hospital and Harvard Medical School, sharing data from our devices that are used in apheresis – the process where starting material is isolated. These experts looked at cell collection from patients with SCD. The blood from patients with SCD behaves differently, which can make it difficult to get enough material for a dose.
The academic team collected data from our devices. Through an analysis of the unique aspects of automated apheresis in these patients, they were able to develop protocols that boosted cell collection efficiency by increasing the mean CD34+ collection efficiency 32 percentage points from 4.9% to 36.8%. These results could mean the difference between a patient sitting for one apheresis session instead of needing to come back for multiple days – a big quality of life improvement, especially considering that apheresis is used several times in preparation for the gene therapies.
Through data partnership and technology integration, we believe that these kinds of improvements will become commonplace, underpinning the growth of CGTs by improving the quality of patients’ lives and helping make them much more accessible to patients in dire need.
