Bildnachweis: BioM.
Artificial intelligence (AI) is here to stay. This technology and its use in drug development, respectively their protagonists, were awarded Nobel prizes in 2024. But AI is not new and around for at least half a century. Is AI truly revolutionizing biopharma, or are we just caught up in the hype?
With the increasing availability of high-performance and cloud computing, global networks and solutions like generative AI with its large-scale models, it opens up entirely new dimensions. So do genetic engineering and gene editing, systems biology, various RNA technologies, et cetera, but AI has also entered our personal settings. For most of us, AI is essential for managing our daily routine, our commute to and from work, our social interactions, and many other things. We all more or less become users and experts at different levels. As we need AI and become dependent on it, AI needs data respectively our data as private individuals or as a professional or part of a user group or (social) community. So, we feed the AI lindworm; hence it becomes bigger, mightier, more powerful, but also more intimidating.
AI along the entire biopharma value chain
Especially generative AI is expected to be a game changer on top of the hype cycle with high hopes to improve diagnostic technologies and accelerate drug development. While annual savings might have a potential of up to USD 70 bn according to Bekryl (2018), McKinsey foresees an annual revenue gain of USD 60 bn to 110 bn (2024), almost equally divided into marketing, research and drug discovery, and clinical development.
AI in target identification and lead optimization
However, AI can and will be used along the entire biopharmaceutical value chain. It starts with the identification and validation of novel but also known targets to select and optimize the best lead candidate. Large language models (some of them quite popular from our private life, such as ChatGPT) have established themselves as natural chatbots to query large protein, genome, or other data banks. The British company Exscientia PLC fused with US-based Recursion Pharmaceuticals to leverage this technology and even better and more precisely predict drug design based on creative AI models. With NVIDIA being another partner in this relationship, they formed an ‘AI-first biotech/techbio’ consortium to offer their services to many outside big pharma companies.
Data quality and validation
Nevertheless, chatbots also need to be trained with proper and sufficient data (like the lindworm), otherwise they might ‘hallucinate’, inventing and misinterpreting information. This is sometimes difficult to realize, but any result needs to be validated, surfacing the underlying evidence. Otherwise, the power of AI will turn against the user and may misguide pre-clinical and clinical strategies.
AI in drug repurposing
Another promising use case is the field of drug repurposing, which refers to the use of approved drugs in different indications. If successful, any development will be much faster and cost-efficient compared to investigational new drugs. One example is dexmedetomidine, which was initially developed and approved as sedative and pain reliver, repurposed by the FDA in 2022 for the treatment of schizophrenia and bipolar disorder.
AI in small molecule design and functionality
AI furthermore raises high hopes in the design and functionality of small molecules which can influence binding affinity, target specificity, off-target effects, safety profile, toxicity, immunogenicity, adsorption, metabolism, solubility, et cetera. All publicly available information from sources like PubChem and PubMed together with corporate and scientific insights can be used to identify novel drugs or drug combinations. AI expedites the understanding of such quantitative structure-activity relationship (QSAR) and to predict precisely and fast the target-lead interaction. This can be applied to the modification and improvement of already existing proteins and molecules or the de-novo generation of such, entirely in-silico. The applied generative AI models work similar to large language models like ChatGPT, and use chemical structures as its basis and foundation (simplified molecular-input line-entry system; SMILES). Such in-silico applications can also support wet lab activities and experiments, named ‘design-make-test-analyses’ (DMTA) cycle. Those approaches will open up a multi-dimensional chemical space of drug innovation. The AI biotech company ‘Insilico Medicine’ has designed a novel drug for the treatment of fibrosis using generative AI. The AI-supported lead optimization was 15 times faster than the conventional way and approximately ten times cheaper. However, the candidates still need to prove themselves in the different clinical phases.
AI in clinical trial design
The initial design of a pivotal clinical trial is also of critical importance and can be simulated through ‘digital twins’, if sufficient and representative data are available. The availability of
real-world data (RWD) also allows the integration of virtual cohorts as controls which will not only increase the confidence in the drug performance but also reduce the time and costs of the drug development. During the clinical trial, AI-assisted monitoring allows the use of smart sensor technologies to increase the availability of real-time data and decreases the drop-out rate that otherwise can jeopardize the timely and overall success of a clinical lead candidate.
The future of AI in biopharma
The AI lindworm is still in its infancy, but the number of strategic partnerships between pharma and techbio companies increases: Genentech acquired Prescient Design, and BioNTech incurred InstaDeep to secure their AI-assisted and data-driven drug-development capabilities. Sanofi took a broader approach with their ‘All-in-on-AI’ strategy, but a unifying ecosystem of drug developers and AI-technology providers is building up. There is still some mutual shyness which also includes investors on both sides as they still need to get used to higher risks and also potentially higher revenue streams. The hype is over, and there are now high hopes which still need to be substantiated in the number of ‘AI-designed and guided’ drugs that make it to the market. We need a little bit more patience – but they will come.
About the author:
Prof Dr Ralf Huss is the Managing Director of BioM. The physician and pathologist is a well-known expert in the field of biotechnology and has over 30 years of experience in research, development and company management.
