How AI and ML Drive iPSC Quality with Aspen Neuroscience's Thorsten Gorba, Ph.D.

Thank you, Erin. Thank you for inviting me and yeah, great pleasure to be here.

Yeah, certainly, Erin. So at Aspen we are developing autologous induced pluripotent stem cell-derived medicines for diseases with high unmet need, and our lead program is in a Phase 1/2 trial for moderate to advanced Parkinson's disease. And really importantly, so the autologous approach is really something that Aspen is specializing in, eliminates the need for immunosuppression, which can be really problematic in an elderly patient population like Parkinson's patients. And so for this particular treatment, we produce autologous dopaminergic precursor cells which are transplanted directly into the putamen of the patient's brain. The goal is for these cells to engraft and relieve the motor symptoms of Parkinson's disease. We have already demonstrated the safety of this therapy and have presented data at renowned international conferences, for example ISSCR. So what differentiates us from other developers, as you said in your introduction, is that we are very heavily leveraging genomics, bioinformatics, artificial intelligence, and machine learning to provide the highest quality cell therapy products for every patient, which is specifically important when you do an autologous therapy where you derive cells personalized from each patient. And yeah, further we are also automating our manufacturing processes to increase the production capacity and reduce the costs of the manufacturing process to reach more patients who could benefit from our cell therapies.

Yes, certainly. So at Aspen we are combining machine vision with AI/ML, with robotic cell processing and automated liquid handling, and for that we use the Celligent cell culture platform. That is a platform that is not widespread in the industry. It's something that we have co-developed. So the Celligent platform is really co-developed by Aspen and implemented in partnership with a company called Cell X Technologies, who are a young startup company, and we have really grown to great lengths to support them and bring them along and help us to develop the platform for our specific needs for an autologous process. And they are a spin-out from the Cleveland Clinic. So the first really important application is what important stem cell culture experts call weeding. This means that after the reprogramming, emerging iPSC colonies, they need space to grow to a large enough size and for the quality to be assessed so that they can then survive being mechanically picked and expanded after passage. Hence, for your budding iPSC colonies to grow and thrive, the fibroblasts surrounding them must be removed. And so traditionally this is done by skilled operators who sit at the laminar flow hood and have an open culture plate with the emerging iPSC colonies under a microscope, and then they use a pipette tip or similar to scratch away the cells that are in the way for the colonies to grow and expand. So, yeah, unfortunately this kind of treatment amounts to a culture shock with the cells sitting outside the incubator for a really long time, and in addition there's also significant risk of contamination due to the extensive manipulation in the open plate. So we have fully automated this process while keeping the cultures in the controlled environment of the Celligent platform all the time. So after automated imaging of the cultures, our AI/ML software identifies the emerging iPSC colonies, and it does so outperforming the human eye of the skilled operators. It can see colonies that the human operator would miss and defines the weeding path around them. So then the weeding is automatically executed by the Celligent robotic instrument, and when the colonies reach a sufficient size, the Aspen vision software will then assess their quality by morphology of the cultures and rank-order them for automated, gentle, shear-stress-induced but very effective colony pick passaging to establish high-quality clonal iPSC lines.

Yeah. So machine vision really is that it's like the self-driving car. So you have a computer system that is aided by a camera and is then, so enabled by AI or machine learning, able to understand what it sees and act accordingly. So that would be the definition of machine vision. Machine vision really only has benefit if it's combined with AI and machine learning. So what we do, the machine vision and the AI/ML software analysis guides us and informs the automated process execution, and this takes both the subjectivity in cell culture quality assessment and different manual skill levels of human operators out of the equation. So this results in a consistent and reproducible process and improves scalability in two distinct and very impactful ways. So firstly, the robotic culture automation that runs 24/7 can achieve a much higher patient lot throughput with a smaller cleanroom footprint than manual culture. And secondly, something that most people really don't think about is the use of machine learning reduces also the need for skilled operators who need to be trained up to a year before they reach a fully sufficient level, both from their judgment and from the manual skills. And this mitigates the often overlooked bottleneck of scaling human staff and keeping up with operator training and qualification, especially if you really move from thousands maybe to tens of thousands of patients. So it will be impossible really to find enough staff and keep them all trained. So we really have to automate not only the execution but also the judgment and assessment of the cultures.