There are currently no results that match your search, please try again.
There are currently no results that match your search, please try again.
On 25th November, a diverse group of leaders came together at Bryden Wood in London to discuss the impact of AI on drug manufacturing. The group included those working in drug discovery and AI-first BioTech, the Medicine and Healthcare products Regulatory Agency, big pharma, big tech, research and academia, venture capital and AI solution providers. Here’s how the discussion unfolded.
A background question was to what extent is the hype around AI justified? Where are the transformations, where are the opportunities, the blockers, the risks and where may it have little impact.
The buzz of knowledge and insight in the room was both exciting and humbling. AI is a subject that we hear a lot about, grasp to some extent, but many of us have little expertise, yet here were people with deep insight and experience.
The pre-event input was illuminating and demonstrated both the breadth of the subject and the depth of knowledge. The discussions which came out of the feedback deviated and evolved beautifully to cover a wide range of important issues.
The group started by discussing the application of AI in the understanding of disease. The simple model presented did not reflect the depth and dimensions of work that is already underway in this area. With databases being curated of tissue samples, medical records and outcomes; there is work in the biological field, looking to deeply investigate cancer, detection, progression, treatment and prognosis. This database is government-funded and developed through an NHS collaboration. It is made available to organisations to use AI models to find insights, links, patterns and correlations which can and will save suffering and lives.
Although ethical issues had been raised in the feedback and in the introduction to the session, it had not been a central theme. However, the next direction of the conversation moved on to ethics. The ethics of privacy and consent in sharing personal medical data. The nuance being whether anonymity can be assured as combining multiple data points could permit identification.There are risks in this area, with ongoing developments to address these concerns.
The curation of this database raised the criticality of having good sources of clean data. Without the availability and quality of data the algorithms are impotent. Collating clean data is one of the biggest costs. Jumping ahead, it was stated that the algorithms are there, the frustration is the data:
• Individuals are reluctant to share their medical data which creates complexities. Different societies weigh the value of individual rights and privacy (deontology) versus social wellbeing benefit (utilitarianism) differently.
• The connection between individual permission to share medical data and finding cancer cures is often not clearly communicated.
• What is informed consent when there is very limited understanding of the benefits and risks in the public arena.
• The large pharmaceutical companies want to see return on investment; however, progress may be limited without their willingness to share data and not just invest funds.
• There is also an ethical-commercial angle here; do we want to give our data to large organisations with a profit motive and if so, how can the individual monetise their data?
It may be that some of the concerns about data security are overplayed. Certainly, the technology exists to keep transactions and data secure, for instance, Blockchain or encryption. Again, this needs to be part of the discussion on the ethical and regulatory playing field for these advances. There was a comment that for embryo research, there was a strong regulatory structure put in place to allow the technology to be developed but with an agreed and transparent ethical and regulatory framework. Is this a model that could be applied?
This discussion raised the issue of fear and mistrust that most significantly arises from a combination of limited understanding, or misunderstanding, and the way AI is presented in the media and online. Here the ideas of AI replacing humans and AI hallucinations were raised as specific fears. To the latter point, maybe the language used is unhelpful and maybe classifying these things as errors would be less provoking. Errors, limitations and the ongoing development to improve outputs is not discussed enough. Certainly, in this forum, as the issues were talked through the level of concern reduced.
There was a clear point raised that much of the news and hype around AI is about Large Language Models, e.g., ChatGPT etc. There was a feeling that most of the intense value of AI will not be in the large language model domain, but in complex, multifactorial understanding and problem-solving, e.g., medicine.
Around the issue of AI replacing humans, there are clearly areas where this is happening, how wide and deep this will be is unknown as is what new roles and opportunities there will be. As a technology it is in its infancy or adolescence, and the future is unclear. Perhaps what is missing are the conversations and engagement about the change.
In the realm of finding new treatments and new chemical and biological pathways, it was noted that AI can consider multiple different data points simultaneously. Unlike existing modelling methods, AI can integrate multiple scan parameters with a wide range of health data, allowing it to consider complex perspectives that would otherwise be impossible to evaluate together. At present, however, the push to deliver new treatments to patients and bring them to market forces us to simplify and deprioritise many of these deeper insights. This impacts the environmental effects of the manufacturing processes for decades, as well as falling short on patient focus and addressing dosage refinement. The lack of a willingness for companies to share their data prolongs this situation.
Deepening the understanding of the interaction between biology and medicines is a critical component to drug development and targeting. This should have significant impacts on treatments, patient outcomes, and both economic and environmental sustainability due to a significant reduction in wastage where medicines have limited or no healthcare value. Pathology was thought to be an area which is hugely under resourced. Jumping to a digital pathology model may be a solution.
If speed and efficiency of the clinical processes also follow there could be significant leaps forward. Well trained AI models should limit or remove the need for animal testing with its ethical and commercial impact. Safety and efficacy testing in vivo could be supercharged by in silico models meaning a quicker route to product regulatory submission. In fact, the potential of telescoping the clinical process was raised. This could, with the same or increased levels of safety and understanding, bring drug development times to months not years.
The realisation of these potential benefits requires the developments to be done step-by-step and in conversation with the regulators. Those in the room understood the key role they have to create clarity and to evolve systems and approaches to support the opportunities. Early engagement with regulators was seen to be key and this is already starting to happen.
The MHRA have a sandbox which allows ideas to be explored and discussed outside of the main regulatory processes but this also must be accepted as a sensitive area. Although much of the conversation is about AI finding advantageous correlations, they can equally highlight links and understandings which expose new risks. How these are dealt with will have a significant impact on whether a culture of openness or protectiveness evolves, and is critical to the potential outcomes.
The group all agreed on a need for a different level and quality of collaboration. With disease understanding, treatment development, medicine manufacturing, and supply systems changing rapidly, the need for an iterative and wide collegiate response is critical.
As well as regulators, this includes a new intensity of academic collaboration. The ideas of transactional business as usual may simply not work when the whole system is moving so quickly. And there was a more general discussion about a systems approach. Individual improvements in discreet parts of the ecosystem will have very limited impacts overall and can, in fact, degrade the overall benefit. An example may be that an approach to drive down the cost of generic medicine may have negatively impacted the ability of the industry to develop and deliver new cost-effective medicines. There needs to be new funding models with less siloed thinking and more collegiate work across industry, regulators, research councils, and others.
Having discussed medicine and process development, the conversation moved onto manufacturing. The concept that the manufacturing universe is quite different to the development universe was postulated. AI is accelerating discovery and development, and the fundamental industry value driver is to get efficacious and financially attractive products to market. Manufacturing is a secondary value driver of incremental improving margins through cost reduction; the prizes are smaller and thus the investment attitude is different. Applications of AI are therefore case-by-case at present.
The group felt that manufacturing is not well-organised to meet the demands of the plethora of specialised medicines coming through. Legacy assets and systems represent a significant inertia to exploitation of opportunities, and, at the current time, companies are investing in legacy systems and not systems that will align with future requirements.
Perhaps, in the future, manufacturing will be distributed to hospitals and pharmacies to align with a drive to personalised and focused medicine. It's not clear whether a point of inflection in this direction is near.
Summing up the discussion was best done by the participants and brought together were some key themes to takeaway.
The session started with a question about whether the hype around AI was justified. On balance, the room saw the likely prospect of significant change; however, the possibility of the changes being much more discreet and focused, rather than transformational was raised. What was clear is that the hype is not useful as it drives distorted behaviour. Clarity and level-headedness about the significant opportunities - while highlighting and addressing genuine concern rather than scaremongering - would be a much better approach.
There is clearly a significant ethical dimension to the development and application of generative and agentic technology. To what extent do we allow systems and technology to replace human activity and decision making? Answers are not straightforward as AI systems start to become demonstrably better than human-centric ones. There is an equally significant dimension about the rights of privacy, and the use and monetisation of individuals' data. Is it right for large corporations to use people's data for their own profitability; should people licence their data and see financial return themselves and/ or should the companies be forced to open-source their models? There are issues about accountability that can become very blurred. Legal and ethical guardrails must be put in place.
The skills required for engaging, exploiting, and developing these technologies are lagging well behind the technology advancement. Our existing workforce lacks understanding, insight and ability at all levels, while the education and transition of the new workforce is also underprepared.
What may be the single largest need and opportunity is wider and deeper engagement. This event provided a snapshot of how understanding, cooperation, and significant value can be achieved through purposeful dialogue. The diversity of the group, the differing perspectives, the wide-ranging subjects, were all commented on as being valuable and a model for continuing the dialogue.
There was a clear sense that new technology requires a change in business models. Significant benefits won't be realised without a change in manufacturing networks, assets, and approaches to operations and investment. Legacy assets and ways of working were seen as inhibitors. Can we define the opportunity in terms that meet investment criteria?
Pre-clinical space presents a huge opportunity and the excitement is justified. We could entirely replace animal testing with something patient-centred, while also finding breakthroughs for many rare and debilitating diseases.
The importance of creating models that move from discovery – to development – to manufacturing is key, and linking sub-systems in larger systems offers huge potential.
The rate of tech disruption in finding solutions to multi-parameter problems is profound. How far this will percolate through to the whole of society or indeed to the healthcare system is unknown. Enabling it could help solve the dilemma of innovative versus affordable healthcare. But the work and investment required in the data sets for model training is not to be underestimated.
On reflection, like many technological advances in history, there is hype and fear as well as profound change and real issues. At Bryden Wood, our sense was that these are not aligned with reality. They are in the wrong places. As a delegate commented, we are not ambitious enough. We must grasp the potential of a new era of health and wellbeing and be highly mindful about the ethical framework and regulations.
Perhaps we should be less interested in the hype about large language models and the fear of losing our jobs and more interested in the new dawn unfolding before us.
Our Accelerate programme brings together competitors and key players to debate and discuss the future trajectory of a sector and how to collectively overcome barriers to technology acceleration. Get in touch to join the movement.
Related Ideas
.jpg)