Andrew Busey, Co-Founder, Form Bio, discusses why vertical AI is set to reshape the cell and gene therapy sector.
The once-fictional artificial intelligence (AI) depicted in literature and movies has transitioned into a tangible and easily accessible tool in our modern world. General AI tools such as ChatGPT have acted as a catalyst, igniting a wave of AI innovation akin to the early days of the internet era. Although the breadth of these applications is undeniably impressive, it merely scratches the surface of AI’s potential. Crucially, vertical AI has the potential to perform tasks that were previously inconceivable within a human lifetime, creating ecosystems that enhance user experiences to a degree that will revolutionise industry capabilities and drive growth. The cell and gene therapy industry is no exception to this digital transformation. Investments in AI and related technologies are driving this seismic shift where modest improvements in early-stage drug development success rates enabled by AI and machine learning could lead to an additional 50 novel therapies over 10 years, a more than $50 billion opportunity. Here’s what this means for pharmaceutical leaders and patients.
The current state of vertical AI integration in cell and gene therapy development
Vertical AI was born out of the need to integrate AI systems that would convert its extensive data and expertise into meaningful outputs that deliver targeted solutions to specific customer problems. It’s true that both general and vertical AI tools can make businesses more efficient but only vertical AI will enable a reduction in costs, improve operational efficiencies and productivity, and produce high-quality cell and gene therapy programs. Trained on industry-specific data provided by subject matter experts, the vertical AI concept trades the breadth of general AI applications for depth, offering unique benefits not available with tools developed for broad applications.
The largest integration of AI tools has happened in the small molecule drug development world, with computational approaches facilitating target identification, drug design, candidate optimisation, and other subtasks along the drug discovery and development pipeline. With human clinical trials with AI-designed drugs now ongoing, it’s only a matter of time until the FDA approves its first AI-designed drug. Furthermore, the insights gleaned from the initial clinical trial data emphasise the importance of establishing pragmatic anticipations regarding the extent of improvement that AI will offer so you can create realistic goals.
These successes and learnings provide a solid foundation for AI’s recent shift into more complex and newer drug development applications, such as monoclonal antibodies and cell and gene therapies. These therapeutics are complex and one of the major bottlenecks in their development is avoiding downstream impurities during manufacturing, which can cause major safety concerns. As biologic drug development and manufacturing is a highly intricate area, there is an incredible amount of data and variability and, thus, a growing portfolio of AI solutions to enhance various steps in the development pipeline.
The need for cell and gene therapy development standardisation
Presently, a total of 3,866 cell and gene therapies are in various stages of development, spanning from pre-clinical research to pre-registration phases. It is imperative for the industry to communicate the precise composition and attributes of these therapies to regulatory bodies. Thorough product characterisation analysis plays a pivotal role in this process, ensuring that we can provide a comprehensive and well-documented account of our therapies. This meticulous approach not only facilitates regulatory compliance but also establishes a clear and traceable record of the therapeutic development journey that will help determine which patient will benefit the most.
The need for transparency in validating in silico and in vitro work
Ensuring the biological validation of AI applications is paramount, as it serves as a linchpin in their effective utilisation by reducing skepticism of AI use and improves models. The absence of industry-wide standards significantly hampers the swift integration of innovative AI solutions. While clinical decision support (CDS) improvements are beneficial, they alone fall short of addressing the challenges. Overcoming reluctance to embrace post-IND sequence modifications is imperative. Equally vital is the need for transparent validation processes encompassing both in silico and in vitro methodologies. These measures collectively fortify the foundation upon which AI-powered advancements in healthcare and life sciences can confidently build their future.
Overcoming data scarcity problems in cell and gene therapy development programmes
In cell and gene therapy development, companies face data scarcity problems. The data needed for analysis takes months to collect and has a high cost. Also, this type of therapeutic is so new that only a handful of cell and gene therapies have been evaluated and approved by regulators. Consequently, the task-specific training datasets are small and likely not fully representative of the types of therapies that are being developed.
Yet, over the past year, there has been significant progress in addressing these data challenges through strategic partnerships that have led to increased yield, safety, and efficiency of preclinical biomanufacturing saving downstream time and cost. By augmenting the task-specific datasets with insights derived from LLMs trained on vast quantities of genomic data it resulted in the curation of more than 70 gene therapy-specific datasets, a substantial collection within this domain, which continues to expand. Combining these task-specific datasets with the knowledge acquired from custom-made LLMs enables us to model several biological phenomena with unprecedented accuracy. These models serve as a foundation for vertical AI and lead to the safest, most effective, and manufacturable options for our biopharma customers reducing time and cost to get their products to market.
Practical steps to integrate vertical AI in cell and gene therapy development programmes
Gaining access to data serves as a fundamental requirement for deploying vertical AI solutions. Yet, an equally critical component is to possess a deep understanding of industry-specific customer workflows. This insight is essential in identifying the tasks most suitable for AI augmentation. It’s widely recognised that these workflows and user networks play a pivotal role in fostering effective end-user adoption and long-term retention.
For such major transformations, what do cell and gene therapy companies need to do to prepare and adopt?
- Identify a small pilot project
The general rule of thumb to handle major change is to start small. Find a good pilot project where you have some data, and gradually scale up your initiatives as you gather valuable insights and refine your approach. This incremental process helps mitigate risks and ensures a smoother transition to larger-scale implementations.
- Map out the end-user workflow
A commonly used method is based on a technique that economists use to analyse AI, which is also used by Workhelix to help enterprises assess their AI opportunities.
(a) break down the user experience into specific tasks
(b) prioritise tasks from most commonly used to least commonly used
(c) evaluate the most common tasks to understand if they would be amenable to AI assistance
(d) assess value (ie. cost versus benefit)
- Maintain data advantage
The concept of gaining a competitive advantage through a data flywheel, where increased usage leads to more data, which in turn improves the model, resulting in even more usage, holds particularly true in the cell and gene therapy industry with highly specialised and challenging-to-obtain data.
Resources need to be employed to improve the efficiency of the AI itself and keep developing the processes and technicalities to deliver comprehensive solutions to evolving customer needs.
- Train employees to have confidence in AI
For CEOs, AI is the number one priority, but 73% of employees believe that AI introduces new risks1. Privacy, hallucinations, bias, and being replaced by AI are just some of their concerns. It’s essential to create a cultural shift and emphasize that AI is meant to augment a human decision or action in a way that improves speed, quality or both not replace it. Start by setting clear AI use guidelines and democratise access to AI skills training.
While general AI capabilities are undeniably fascinating and broadly applicable, vertical AI holds the power to reshape and empower the cell and gene therapy industry. The specialised nature of vertical AI allows it to address complex, cell and gene therapy-specific challenges that were previously impossible to address. This shift towards vertical AI will facilitate the creation of superior final therapy products, cost savings, enhanced operational efficiencies, increased productivity, and ultimately, make it possible for these life-saving treatments to reach patients faster.
About the author
Andrew Busey is Co-Founder & Co-CEO of Form Bio & Chief Product Officer & Co-Founder, Colossal Biosciences. In the past 25 years of his career, Busey has pioneered some of the internet industry’s most important technologies — including work on Mosaic, the first web browser (now part of Microsoft Internet Explorer); creating iChat, the first web-based chat system and one of the first instant messaging applications, invented chat-with-a-customer-service-rep and more. Busey has a degree in computer science from Duke University and an MBA from The Wharton School at the University of Pennsylvania.