What insights did you gather over the course of your career that helped inspire the creation of Quartic AI?

I have lived with Industry 3.0 automation all my career. While automation evolved, control and automation is limited to a zoomed-in view of equipment and process. Existing automation hit its limitation when dealing with a zoomed-out, elevated view of production – this is why big data and AI are needed for the next generation of automation. When I started looking at how AI was being brought into manufacturing, I felt that technology companies are coming at automation from a top down approach. There was a big gap between that strategy and the production floor with the machines, equipment and processing. That is both in terms of understanding the nature of how these processes and machines work, and also the nature of the data that they generate.

For long term sustainable value to come from these technologies. AI cannot be seen as an isolated discipline in terms of skills, or an isolated process. The people who can apply AI and ML best are the people who understand the manufacturing process. Much of the focus of tech development has been for people in IT and data science. This creates a big barrier to adoption for chemists, biologists or chemical engineers, it is unsustainable. Overcoming those two barriers is the inspiration behind Quartic.ai.

To what extent have pharmaceutical manufacturers shown a willingness to adopt AI based technologies?

The desire to use these technologies is generally very high. The level of urgency to adopt is not as high. There is a perception that the biopharma industry, is data rich. This is true, but a lot of the data that was being generated by machines, sensors, and equipment was not being captured because there was no need to capture it in the past. It is not that data is not being generated, there is no way to easily make it accessible to machine learning and analytics. That is the number one barrier. The other impediment is the regulatory part, which is that most of these processes are validated processes, equipment is validated. To make any kind of changes, particularly in those areas that are validated that fall under GMP requirements, requires you to go into a lot of change management and there is naturally hesitation. Companies have the perception that there will be a difficulty getting changes approved for new technologies. A big catalyst, that is making people more willing today, is the FDA Emerging Technologies Task force (ETT), which is incentivizing acceptance and approval.

How will Quartic’s partnership with Bright Path Labs expand its presence in the life sciences?

Bright Path Labs has a technology that brings speed to the commercialization of API production. Continuous manufacturing is a big need in the pharmaceutical sector and Bright Path labs technology accelerates that process. In order to enable and commercialize the technology they needed modern automation. It became a natural fit for us to work together, as their method of manufacturing can be accelerated with use of machine learning and AI. In moving from one molecule to the other molecule in the same reactor it is a challenge to get to quality production quickly. They are using inline monitoring (ILM), which are analytic measurements, generally done offline. That is typical for a batch process. However, if you have a continuous process you need to do those measurements online. Quartic is able to use online measurements to create closed loop control. This creates a semi autonomous control system. It has basic automation that is going to control the equipment but the basic automation is being guided by AI models, which will be trained for particular molecules. For each molecule, we know what the perfect set points will be to achieve a quality target quickly.

What long term impact do you see the Covid-19 having on pharmaceutical manufacturing?

As a result of Covid-19, there has been a realization that these types of events can happen again and they can be very disruptive. The current situation requires speed of experimentation and speed of commercialization and secondly, requires agility. Repurposing your existing manufacturing assets to make something different or having to move production from one geography to another. That is agility. This requires a higher level of automation and a much better understanding of data. Agility absolutely, benefits from the higher orders of automation, machine learning and AI. The second part is speed to market. We have all been contemplating ways to commercialize these new drugs through internal supply chains or contract supply chains, very quickly. Quartic is doing this with a technique called transfer learning, which enables us to successfully make something in one manufacturing facility, train the machine learning model on that manufacturing facility, let's say in North America, and then when I open an additional manufacturing facility in Taiwan, the process and product knowledge that I gained is captured into an algorithm and applied to the new process. I see commercialization tech transfer as perhaps the biggest and easiest path to achieving returns to the bottom line from AI and machine learning. First, you need to run a lot fewer production runs during process qualification and validation, this saves a lot of time and money, second, with algorithmic knowledge transfer, your subsequent facilities for large scale manufacture come to market much faster. Both have a tangible impact on the cost and time of commercialization – benefiting the patients, and increasing the manufacturer’s margins.

What are some applications of AI that are currently aspirational that might add a lot of value in the future?

The one at the top is autonomous bioreactors. We have proven these strategies using a digital twin. So, we simulated a bioreactor, which is very difficult. On that simulation, the recipe can be controlled automatically. We are now working with a couple of manufacturers of biologics who are all very interested in autonomous bioreactors. It might take time to get wide adoption, but Quartic currently possesses this capability. If you want to make this technology autonomous it requires deep learning and reinforcement learning. What that means is that the models have to be able to learn while the process is running. In order for machine learning to learn it needs to see variability. For this reason, we run the process in such a way that we introduce variability while staying within proven acceptable limits of validated quality. This way the model can learn, which is very difficult to do in a commercial operation. This technique is called reinforcement learning. Here you are using a digital twin to speed up the learning of the model. Then you apply that learning to the actual process. You learn from the process and you eventually come up with a perfect model that can run a bioreactor autonomously. Most of the more valuable cures and therapeutics are going to come out of biologics, so the foundation of any biologics process course is a bioreactor and a bioreactors process is difficult to control. If you go out of control in the middle of production, it can be very expensive. That is one of the key technologies that we are focusing on and we have made quantifiable progress in the last six months. Now we are moving toward trials with customers in an r&d setting which allows us to now prove it and then take it to commercial manufacture.

The other big opportunity is in process analytics technology (PAT) which is an enabler of continuous manufacturing, and existing batch manufacturing. CMO’s and large pharma manufacturers have invested a lot in this technology. It is very expensive, but they haven't been able to get as much value out of it as they should have. The reason for that is because the data that the spectral analyzers like Raman spectroscopy and HPLC takes a lot of time to analyze and come to a meaningful conclusion. That makes it a perfect candidate for machine learning.

What objectives would you like Quartic to achieve over the next two years?

Quartic is creating solutions that boost quality and enable autonomous manufacturing. My first objective is to have as many of each of our applications commercially deployed and approved by the regulators. Our customers need that confidence. The other objective is to bring all of our isolated applications together into a product lifecycle management platform. The PLM platforms that exist today are old and based on legacy computing technologies. We want to implement a complete PLM program so that you can see the impact on the entire supply chain in real-time and predictively. Knowledge captured about the molecules from Phase 4 onwards can unlock a lot of value. This is only possible when you have complete application and we are doing this under what is called continuous process verification (CPV). We want to deploy CPV to as many pharmaceutical customers as possible in the next 12 to 18 months.