The AI revolution in drug discovery a status update for 2023

In this opinion piece, Philippe Moingeon (Servier, France) discusses the success of AI and machine learning (ML) in accelerating and optimizing drug discovery and design. Philippe discusses how AI algorithms can help tailor treatments personalized to patients, and his opinion regarding the future directions of AI in drug discovery.

A successful leveraging of AI and ML to accelerate drug discovery

AI and ML have been successfully utilized to support drug discovery, with evidence that they can significantly accelerate the design and optimization of novel drugs, making it more efficient, accurate and cost-effective. For example, a novel antibiotic termed Halicin, exhibiting a broad anti-microbial spectrum, has been discovered with AI . The first AI-designed drug candidates are now being evaluated in humans within one to 2 years, compared to 5 to 7 years for the usual discovery phase . Herein, I will discuss how AI is accelerating the drug discovery process by facilitating the identification of relevant therapeutic targets and the selection of optimal drug candidates, whilst also opening new frontiers in light of limitations that still need to be addressed.

Making the correct important decisions

AI and ML are useful for creating predictive models that inform critical decisions during drug design. To this end, comprehensive disease modeling is now commonly undertaken as the first step in the development of a novel drug. Disease models are based on extensive molecular profiling data generated with multi-omics technologies on large cohorts of patients, with two main objectives. The first objective is to represent disease heterogeneity by stratifying patients based on shared pathophysiological mechanisms documented in the form of dysregulated pathways in their blood or in target organs. Network computing methods are further implemented to create an interactome of dysregulated genes or proteins in order to predict those likely involved in disease causality. These targets can then be selected as highly relevant in the context of a comprehensive understanding of the physiopathology of the disease, which is a huge improvement provided by AI.

Another application of AI is the identification of appropriate drug candidates . AI algorithms can analyze chemical structures to predict drug-target interactions, identify potential drug candidates and optimize drug properties. This approach has been recently facilitated by the development of the AlphaFold algorithm, which can predict protein structure from primary amino-acid sequences, potentially significantly reducing the time and cost involved in traditional drug discovery methods. AI is also being utilized to predict and optimize drug properties, such as absorption, distribution, metabolism, excretion and toxicity. Selecting the optimal drug candidate with the desired properties interacting with the right target can increase the success rate of subsequent drug development, which is currently extremely low, with more than 93% of drug candidates failing to reach the market after negative results obtained during costly clinical studies.

AI algorithms can also help tailor treatments to biological specificities of individual patients. Ensuring that the right treatment is given to the right patient has the potential to improve patient outcomes and reduce the time and cost involved in clinical trials. This approach, informed by in silico models that we call computational precision medicine , has been successful in treating cancer, where patients with specific genetic mutations have responded well to targeted therapies.

Future directions

In light of these first successes, the pharmaceutical industry has now broadly adopted the use of AI and related technologies to support drug discovery. AI-powered industrialized platforms are being implemented to mine internal as well as public data sources in order to inform target identification, selection of therapeutic modalities and drug design. These high throughput computational modeling approaches are being associated with robotization of drug synthesis as well as with automated preclinical screening models. As of today, predictive algorithms are commonly utilized to design and optimize small chemical molecules as well as synthetic oligonucleotides. There is as well a strong interest in training artificial neural networks to predict functional properties of biologicals, such as monoclonal antibodies, proteins, peptides based on sequence and/or structure information. The prediction of stability and compatibility of biologicals with potential formulations is being considered to assess their developability as drug candidates.

AI also supports drug development, with a capacity to integrate and interpret massive and multi-dimensional data. AI becomes commonly utilized to facilitate the design, implementation and monitoring of clinical trials assessing novel drugs. AI-extracted insights can, for example, guide both site and patient selection based on specific requirements for the trial. Virtual patients created from the medical records of real patients are being utilized to generate virtual placebo groups, or even to predict drug efficacy and safety. Digital twins that model specific organs, such as a living human heart, have been successfully utilized to evaluate the performance of medical devices such as pacemakers or stents . For years to come, such in silico trial simulations will represent a complementary approach to clinical studies to better evaluate drug efficacy and safety. This mixed reality approach will reduce the time and cost involved in clinical trials, as fewer patients will need to be recruited, with a higher probability of success.

Remaining hurdles and challenges to be addressed

Despite the significant impact of AI in drug discovery, there are still some limitations that need to be addressed. One of the main limitations is that AI algorithms require vast amounts of high-quality data to be effective, but such data are often limited in the early stages of drug development. The need for standardized methods for data collection and analysis as well as legal restrictions for sharing and using personal health data, are other additional challenges. Also, AI algorithms are often regarded as "black boxes" and transparency and interpretability of their predictive outputs should be improved. Proper methods to ensure the robustness of AI algorithms and validate the outputs of predictive algorithms are also needed to facilitate acceptance of digital evidence by regulators.

Convergence of the intelligences as a key success factor

AI has already revolutionized drug discovery and is also poised to do so for drug development in the years to come, with an anticipated coming wave of innovative treatments . Nonetheless, many health professionals remain concerned about AI performing tasks traditionally carried out by humans. To fully capture the value of these novel technologies in the drug industry, it should be emphasized that AI is not a replacement for human expertise, but rather a tool to assist scientists in their work. Acculturation and training of human experts are thus essential to create the opportunity for synergies between humans and AI. ‍

Disclaimers:

The opinions expressed in this feature are those of the interviewee and do not necessarily reflect the views of Future Medicine AI Hub or Future Science Group.