How AI affects biotech companies

This article is part of our Biotech Review of the Year - Issue 12 publication.

We help biotech companies to navigate the developing regulatory requirements of AI, adapting their contracting to accurately address AI risks and opportunities, and ensuring that they have internal policies to provide the necessary guardrails for employee use of AI in their roles. This article summarises our experience on the effect on, and use of, AI by biotech companies. 

Accelerating innovation

AI is having an increasingly significant impact on biotech companies, revolutionising various aspects of their operations. Set out below are some of the more exciting areas where AI could make a significant difference.

Enabling personalised medicine

Before we were all generating entertaining text and images using Generative AI, another branch of AI, namely machine learning (ML) models, was being used to enable the development of more personalised treatment strategies. By analysing genetic and clinical data, ML models can help identify the best treatment options for individual patients, increasing the likelihood of success and minimising side effects, thus making treatments more targeted. AI tools can also help interpret complex genetic information, identifying mutations, biomarkers and other insights that can drive personalised treatment plans.

We have already seen promising results from initiatives such as Exscientia’s EXALT-1 clinical trial in the oncology field. 

AI can also help advance the more pioneering fields of cell therapy and gene editing. ML models can predict the impact of genetic modifications and design optimal genetic sequences to enhance the accuracy of gene editing. This is crucial for complex therapies like cell and gene therapies.

Accelerating drug discovery & development

Traditional drug discovery is slow, expensive, and has high failure rates. AI is accelerating the process of identifying potential drug candidates.

AI algorithms can quickly analyse vast amounts of biological and clinical data to predict how different compounds will interact with targets and identify new drug candidates. Once a potential candidate is identified, AI can help optimise its structure, improving efficacy and reducing toxicity by simulating how the medicine will behave in the body. Multiple corporate and academic researchers are already using such tools to conduct in silico screening of candidates by analysing existing libraries, predicting which molecules are most likely to interact with specific disease-related proteins. Further, ML models can sift through genomic and proteomic data to identify novel therapeutic targets that have evaded human researchers. 

Improving clinical trials

AI can enhance clinical trial design and execution including:

1. patient recruitment, where AI algorithms analyse electronic health records to assess the impact of inclusion and exclusion criteria and identify suitable candidates for clinical trials more quickly and accurately;
2. predictive modelling, where AI predicts patient responses to treatment based on their medical history and genetic information, helping design more precise trials with smaller patient groups –potentially eliminating the need for placebo groups or dose escalation groups; 
3. optimising dosing and testing schedules;
4. identifying potential side effects and allowing more rapid identification of adverse events; and 
5. analysing trial data in real-time, enabling quicker decision-making and reducing time-to-market.

We have seen a number of deployments where trial participants undergo a whole genome sequence on enrolment and AI tools are used to analyse the emerging data in real time. This can lead to unexpected findings that deepen knowledge about the disease and therapeutic options. 

Low hanging fruit – assistance with regulated functions

While there is a lot of excitement about revolutionary applications of AI in drug design and discovery and in the development of personalised treatments, these may take some time to fully mature. 

In contrast, there are some easy (evolutionary rather than revolutionary) productivity wins involving the deployment of AI tools, including Generative AI, to assist with regulated functions, such as the management of:

• manufacturing and logistics;
• copy reviews and updates to product labelling;
• interactions with regulators (applications, variations, submissions, renewals, Q&A and even regulatory intelligence); 
• clinical trials, especially multi-jurisdictional trials; and
• pharmacovigilance.

These functions are all labour intensive and business critical. By way of example, due particularly to divergent regulatory requirements across jurisdictions and differing approval processes across regional business groups, it can take years to implement a global change to product labels. Additionally, the staff participating in such functions are often experienced and relatively expensive. Ideally, AI would perform more of the heavy lifting on mundane work, freeing up staff for them to spend more time on less routine tasks. AI could even accelerate these processes, such as by improving coordination and reconciling regulatory requirements.

Biotech clients are already trialling AI in many of these areas. They are confident that such deployments will significantly improve these functions and enable them to be nimbler. Further, such deployments will not be heavily regulated by the EU AI Act, reducing the regulatory burden associated with them (although deployment of AI in a regulated function such as pharmacovigilance will need to be appropriately validated and documented as part of the company’s audited quality management systems). 

At a more prosaic level, most biotech companies have already adopted a proprietary version of one of the Generative AI tools (such as Gemini or ChatGPT) and are using it across the company to improve productivity and to increase staff familiarisation with AI technology. 

Interestingly, a number of regulators such as the MHRA have already started using AI to help manage applications for clinical trial authorisations and MAs to speed up these processes. Both regulators and regulated entities can benefit from AI.

Potential AI deployments in general business functions such as HR and Finance (whilst not specific to biotech companies) may also be an important source of efficiency gains.

The impact of EU AI Act

The EU AI Act establishes a regulatory framework categorising AI systems based on risk levels, with some systems being deemed high risk, including some used in medicine and healthcare. The good news for biotech companies is that many of the “easy win” deployments of AI discussed in this article do not fall into the heavily regulated categories under the EU AI Act. A significant reason for this is that such use cases do not constitute medical devices or in vitro diagnostic medical devices (IVDs). The more revolutionary deployments discussed in this article would be regulated under the EU AI Act because they constitute medical devices or IVDs, but there are still exceptions. For instance, an AI for early-stage drug discovery prior to pre-clinical testing is not a medical device or IVD, as its purpose is mere hypothesis generation and, although increasingly important, the use of the AI tool represents just one aspect of many other (human-led) processes and stages in drug discovery.

However, the principles under the EU AI Act are already becoming pervasive. For example, in a recent Reflection Paper, the European Medicines Agency (which does not regulate medical devices or IVDs) encourages compliance with data science competence (and the principles set out in the EU AI Act) when using AI to help develop novel medicines and conduct clinical trials¹. This is the case even where such deployments are not directly regulated by the EU AI Act. The takeaway here is that it may soon become best practice or in line with sector guidance to comply with EU AI Act principles – including those around transparency, robustness and data quality – even if the relevant system is not directly regulated by the Act.

Further, increasingly novel medicines require the use of a companion medical device or an IVD. Also increasingly, such devices are AI systems under the EU AI Act, or incorporate an AI system as a safety component. In such cases, the device will be regulated by the EU AI Act and such devices will need to undergo a conformity assessment for compliance with the EU AI Act as well as the Medical Device Regulation or IVD Regulation. The takeaway for biotech companies should be to ensure that the regulatory approvals of these associated devices are on the critical path for the development of novel medicines, given the need to align approvals for both the AI system and the medicine. 

Biotech companies should note, however, that existing, established law still also governs their use of AI. For example, the GDPR and medical confidentiality rules (which vary by jurisdiction) regulate AI systems analysing genetic or health data. These impose strict requirements for an appropriate legal basis for the data processing, while data anonymisation or pseudonymisation is often required to reduce privacy risks. GDPR compliance extends to ensuring AI systems’ transparency, enabling individuals to understand how their data is processed and, where applicable, contest decisions made by AI systems. Finally, some countries (including the UK) have strict rules regarding storage of samples for genetic testing.

Practical Steps

Even where the proposed deployment of an AI system is not regulated as such under the EU AI Act, clients are already adopting guidelines and governance frameworks to assist the use of AI and to prevent possible misuses. These include systems to:

1. ensure that each AI system is transparent and that its decisions can be explained;
2. prioritise privacy and data protection in the design and deployment of AI systems;
3. clearly set out accountability structures for AI decisions and outcomes such that there is a human-in-the-loop before a final decision is made;
4. design AI systems to be human-centric and support human decision-making;
5. ensure AI systems are safe, reliable, and resistant to attacks;
6. ensure AI systems comply with existing laws and are adaptable to future changes;
7. keep a record of the provenance of all data sets used to train the AI system; and
8. regularly audit AI systems for fairness and bias. This can include active steps to mitigate bias.

Many of our clients use software evaluation tools that help them to objectively measure the quality and effectiveness of large language model-based applications. These tools are designed to assess model accuracy, examine biases and check for inconsistencies and inaccuracies. 

While challenges remain, the potential benefits of AI are immense, promising to revolutionise the life sciences sector and improve patient outcomes.

Footnotes

^{1    See, in particular, the conclusion on page 12-13 of the Reflection Paper}