Wednesday, October 28, 2020

COVID Vaccine Production Could Accelerate the Use of AI in Biomanufacturing

 Discussions about using machine learning in biomanufacturing have increased recently as COVID vaccine production gets closer to its anticipated scale up. Technology providers see a huge opportunity to bring AI products to a market that might have sat in development had there not been the immediate need to get a COVID vaccine out the door.

Illustration © Visual Generation | Dreamstime.com
                                                          

Different suppliers have different approaches. While some are looking to machine learning to detect patterns in treatments, others are looking at measuring tools needed to monitor biologic production. Either way, biomanufacturing and traditional pharma offer plenty of opportunities to reduce cost, to improve quality, or to increase speed. Moreover, many observers believe biomanufacturing processes hold the greatest near-term promise for AI within the larger biopharma industry outside of drug discovery. But machine learning in this field looks very different than it does in other industries. Not surprisingly, detecting the behavioral patterns of bacteria requires very different approaches than detecting the behavioral patterns of people.

Measuring Cell Cultures isn't like Measuring Website Visitors

More advanced analysis of cell cultures is now taking place now in academic research in order to increase the quality of these cultures that are used to produce biologic drugs. Computational biologists are applying algorithms to improve yields on E.coli by optimizing batch quantities without monitoring actual cells. This approach isn’t based on the predictive analytics that are often used in AI-based customer analysis in marketing and website optimization, but rather trial and error analysis on data from a “virtual” cell culture. Essentially, the algorithms provide rapid pattern recognition on activities in the culture that would take people much longer to assess. The next step though, is to move out of the virtual world and bring the techniques to real bioreactors.
With AI in biomanufacturing heading down a different path than the heavily software-based approaches used in marketing and website analytics, it is also heading to an environment where algorithms have to learn continuously from ongoing manufacturing processes, not just by sifting through massive databases. Measuring cell cultures also requires monitoring hardware that does not contaminate the dividing cells.

Hardware Plays a Crucial Role in Biomanufacturing Machine Learning

Taking advantage of photonic products used to monitor and manage fiber optic networks in telecommunications, Raman spectroscopy is one technology that can measure product features using lasers and infrared wavelengths. Chinese Hamster Ovary cells, which are commonly used to grow antibody therapies, emit biophotons during bioprocessing that can be measured by Raman spectrometers without interfering with production. While these devices add to the capex needed to buildout a facility, they can reduce opex through labor savings and more precise quality measurements during both upstream and downstream processing.
Beyond Raman spectroscopy, many other IoT (Internet of Things) sensors are built for higher volume manufacturing environments with less sensitivity. Chemical manufacturers and oil refineries use these devices to a large extent, but these same IoT sensors can’t be placed in a bioreactor without impacting the cell culture. Outside of basic environmental monitoring in the space surrounding the bioreactor, customizing industrial sensors for biomanufacturing would be very expensive for the device suppliers, and leave them with limited volumes over which to cover their fixed development costs. As a result, these types of devices are not used as widely in biomanufacturing.
Sensors and spectrometers are needed on the hardware side, on the software side there is growing use of AI and analytic tools to help automate biomanufacturing. While the additional data can be useful in places, it is still faster and easier to adopt in less regulated industries. Biomanufacturing process analytics can certainly take advantage of more frequently sampled data, but development of Process Analytics Technology must be coordinated with the FDA.
Dependent on learning through experience, and needing to conform to the needs of a regulated industry, biomanufacturing AI will likely see its growth accelerated by COVID vaccine scale up in the next 12 months, but perhaps not at the torrid pace some are hoping for. Nonetheless, there could be greater uses a few years down the road as biomanufacturing faces new unit cost, quality, and scale up challenges with the higher volumes brought about by biosimilars, cell and gene therapy, and even gene editing. COVID might just be a stop along the way to a deeper industry embrace of AI.


No comments:

Post a Comment

Note: Only a member of this blog may post a comment.