Adrian Tombling, partner and patent attorney, Withers & Rogers explores the future of GenAI and drug discovery.
Withers & Rogers
The role of AI in drug discovery is well established, particularly when it comes to accelerating research activity and improving the chances of success in developing an effective therapy. However, recent technological advances are bringing new challenges for pharmaceutical and biotech companies, and in some areas is requiring a change in approach.
Whilst there are virtually no drugs on the market that have been badged ‘discovered by AI’, the reality is that many drugs currently in clinical trials have, at least in part, relied on the use of advanced AI models. Insilico Medicine’s AI-generated candidate, Rentosertib, produced strong results recently in peer-reviewed phase 2a trials for use in improving the lung function of patients with idiopathic pulmonary fibrosis. With a success rate in phase 1 clinical trials of over 80%, it seems likely that more AI-generated drugs will reach the market in the next few years.
Trained on huge quantities of data from diverse sources, large-scale AI models can analyse vast amounts of data, including successful and unsuccessful clinical trial data, quickly and efficiently, in order to identify potential areas of interest. For example, specific patient subgroups may be identified that demonstrated better responses, lower side effects or reduced disease progression. AI is currently often used to support repurposing activity as the active agent is already known to be compatible with human use; however, advanced AI models are also widely used for target identification, drug design and predicting patient outcomes prior to the start of clinical trials. Aiming to improve clinical trial outcomes, the company CureMatch Inc is seeking patent protection for a computer system that selects an optimal cohort of patients for a clinical trial by predicting the impact of a drug on patient biomarkers.
Deep learning models
Building on the success of early AI-enabled drug discovery models, the Massachusetts Institute of Technology (MIT) has recently launched a new deep learning model, which can predict how tightly drugs will bind to proteins. Boltz-2 is much faster and more accurate than previous approaches. Predicting how well the molecules will bind together is critical to the drug discovery process.
In the UK, the recently launched OpenBind consortium, backed by £8 million from the Government’s Sovereign AI Fund, is using ‘experimental’ technology to develop the world’s largest collection of data on how drugs interact with proteins. Specifically, the consortium is aiming to create an unrivalled AI-powered resource for drug developers by providing access to high-quality data on drug-protein interactions. Having access to large amounts of high-quality data relating to drug-protein interactions is essential in training AI models that are to be used to predict drug-protein interactions.
Accelerating processes and better outcomes
Advanced AI models trained on large volumes of patient data are used by pharmaceutical companies to advance therapeutic candidates faster, more cost-efficiently, and with fewer failures. For example, a suite of AI-powered models developed by Massachusetts-based Valo Health are currently being used in partnership with Novo Nordisk to support the development of next generation therapeutics for cardiometabolic diseases.
With more bespoke large-scale computational models in development, pharmaceutical companies are preparing for a second wave of AI-powered opportunities, which could result in many more drugs coming to market. For some companies and organisations significant untapped commercial potential could still be hidden in stored patient data, as well as in clinical trial data. Such data can be of value in training advanced AI models that may then be used to identify therapeutic agents for treating unmet clinical needs more quickly and accurately than ever before.
However, challenges lie ahead. The regulatory landscape is still evolving and the US Food and Drug Administration (FDA) recently published draft guidance for the use of AI in drug discovery, which recommends use of a risk-based credibility assessment. The guidance also flags potential regulatory issues in relation to ‘companion diagnostics’ (CDx) due to the role these AI-powered medical devices play in providing information to ensure the safe and effective use of a therapeutic product. Furthermore, the EU AI Act classifies any AI method intended for medical purposes as ‘high-risk’. This means that such methods must comply with strict requirements including risk management, transparency, detailed documentation, human oversight and more.
When working collaboratively, for example with database right holders, AI drug discovery companies, and pharmaceutical companies, it is important to establish clear contractual terms and conditions, especially when it comes to the ownership of any existing IP that is being shared and any new IP that might be generated as a result of the collaboration.
Under UK and European patent law, patent protection can only be secured for innovations where there is a human inventor. For this reason, patent applications for AI model outputs may need to be prepared with care. It is also sometimes wrongly assumed by developers of AI models that the software itself is not patentable, even though both the UK Intellectual Property Office (UKIPO) and the European Patent Office (EPO) have been clear that in many cases it can meet the requirement for patentable subject matter. For example, Fujitsu was able to obtain a patent for a computational method of efficiently computing a stable binding complex of a target molecule and a drug candidate. In addition, the company BenevolentAI was recently granted patent protection for a computational method of identifying, using a machine learning model, an optimal drug molecule that targets disease-linked genes. Working in partnership with an IP professional can help to ensure that the commercial potential of AI drug discovery models and their outputs are protected.
Growing use of AI models in drug discovery means more patent applications are being filed to protect training methods or data-driven processes. Recognising this trend, it may be advisable for pharmaceutical companies and their research partners to conduct freedom-to-operate searches to establish if there is any existing IP out there that might act as an obstacle to their desired activities.
AI has already transformed drug discovery processes, and whilst challenges lie ahead, the pharmaceutical industry knows that the best is yet to come.