1. Gene therapy vector design
The clinical success of novel gene therapies is determined by a combination of target specificity, delivery efficiency, and transgene incorporation. However, these performance attributes are difficult to predict for genetic medicines in vivo in humans.
AI is proving instrumental in the work of Sean Bedingfield of Eli Lilly’s Genetic Medicine Team and recent guest on TTP’s Invent Life Sciences podcast. He sees increasing use of AI in addition to traditional chemo-informatics to design high-quality product candidates, be they viral vector sequences or the lipid nanoparticles (LNP) used to carry mRNA molecules into cells. He also sees significant value in AI for pulling disparate datasets together. He highlighted the growing importance of formal characterisation and quality control in providing a solid experimental foundation for AI-generated insight.
2. Modelling, cell metabolism, media optimisation
Optimizing the media in which advanced therapies are manufactured can drive dramatic improvements in yield, therapy quality and, importantly, therapy potency. Despite this, the industry still largely relies on off-the-shelf media that assume uniform cell needs.
Tolemy Bio is one company working to change this. The company is using a model-driven approach to optimising growth media with a proprietary computational platform. AI can overcome the challenges of combinatorial and biological complexity and non-linear cellular responses, to enable the design of media optimised for the manufacturing needs of particular cell and gene therapies. As AI is increasingly used to combine datasets and reinforce better biological models, this approach is likely to get progressively stronger, which could be a significant boost to therapy potency.
3. Digital twins
reproducibility of drug products. So, currently process development needs to be undertaken by highly skilled staff and incurs significant development costs. In addition, scarcity of donor material creates a high barrier to experimental process development.
In this context, digital twins (DT) could have significant value for preclinical companies to explore their manufacturing process, whether the process will scale, and what bottlenecks they might face, in order to help them win funding and prioritise development
At TTP. we are currently developing tools and techniques to take advantage of the capabilities of generative AI to accelerate the early stages of DT development, to create what we call ‘seed’ DTs that can form the basis for the full DT under expert supervision. We are experimenting with this approach for bioreactor process design and process control to help to improve the product quality of advanced therapies and enable manufacturing scale up.
4. Machine vision
Imaging has long been and continues to be AI’s poster child. In the advanced therapies industry, imaging and ML algorithms can be used follow and categorize cells at all stages of development and manufacturing, providing the foundation for manufacturing automation and scale-up.
Cellino Biotech’s autonomous biomanufacturing platform uses “a combination of label-free imaging, high-speed laser editing, and machine learning to automate the production of stem cell-based therapies”. The combination of imaging and computer algorithms can overcome the limitations of manual cell therapy production, such as high variability and cost.
Together with Massachusetts General Brigham’s Gene and Cell Therapy Institute, Cellino is now building its first on-site foundry to manufacture clinical-grade patient-specific IPCs using its “Nebula” platform, Nebula compresses the company’s bioprocess technology into cassette-based systems to enable scalable deployment.
Similarly, Aspen Neuroscience is collaborating with Cell X Technologies to add tools and methods to the company’s Xceligent™ platform to support the Aspen’s autonomous manufacturing goals for iPSC-derived neurons. The platform uses machine learning and deep learning computer vision algorithms trained by experts to identify and select desirable cell differentiation states to control the manufacturing process.
5. Manufacturing and product release
With the high complexity of manufacturing processes and of the drug products themselves, quality control is of growing importance for advanced therapies, a theme we touched upon in a previous TTP Insight focusing on mRNA vaccine production.
Ken Harris, Head of Strategy and AI at OmniaBio and recent guest on TTP’s podcast, says the CDMO has a strong focus on AI to increase the efficiency of advanced therapy manufacturing. He sees AI as a partner to automation, with a particular focus on QC. “Our goal there is to basically get a 4-5-fold output in the same infrastructure and therefore get the cost down.”
Ken was previously at Amazon and hopes to bring AI to improving the logistics of therapy manufacturing and delivery. OmniaBio is also interested in developing AI-trained classifiers to stratify cells entering the manufacturing process, adapting the process accordingly and following through to clinical response.
Even more exciting, he said on the podcast, is the development of data cleanrooms, which enable companies across the industry to apply AI learning across the data to improve product quality without revealing or sharing underlying data, says Harris. “We can actually compute against all of the data without anyone knowing or seeing what my data is or looks like. And that's just revolutionary.”
Accelerating advanced therapies
These examples show that implementing AI and digital in the context of manufacturing automation and analytics is now key to improving product quality, increasing operational efficiency and enabling the manufacturing scale-up of advanced therapies.
Our mission here at TTP is to accelerate the progress of cutting-edge advanced therapies to the clinic. Reach out to find out how our customised technology solutions could accelerate your product.
