AI is transforming drug discovery—here’s how
Key Takeaways
Genomic profiles, imaging data, and chemical/drug databases can be paired with AI to discover new drugs.
Drug discovery by AI has the potential to save time and expense.
Although a small number of drugs discovered by AI have entered clinical testing, more complex predictions of efficacy or adverse effects have yet to be realized by the technology.
The discovery of new drugs became a priority during the COVID-19 pandemic. Nevertheless, conceptualizing a new drug and then bringing it to the clinic is lengthy, expensive, and complicated—and replete with many pitfalls.
There’s been an explosion in medical findings during the past 10 years, which, when coupled with breakthroughs in computational hardware and the advent of deep learning, open the door to leveraging artificial intelligence (AI) in drug discovery.
AI advancements continue to progress
“Due to recent progress, there is a great interest in the application of [AI] methods to improve various stages of drug discovery pipeline, including de novo molecular design and optimization, structure-based drug design, and pre-clinical and clinical development,” wrote the authors of a review published in Heliyon.[]
“Biomedical datasets, such as genomic profiles, imaging data, and chemical and drug databases, can be coupled with analytical methods, especially deep learning models, to coordinate the tools needed to discover useful drugs and their clinical applications,” they added.
AI can potentially help with various stages of drug discovery, including the following:
Identification of novel targets coupled with evidence of related diseases by leveraging the proteome
Using high-throughput screening of compound libraries vs such targets
Optimizing compounds for favorable drug properties and testing in preclinical and clinical trials, with the potential for eventual FDA approval
Generative models could be used to design novel synthetic molecules, and reinforcement learning could be utilized to optimize drug properties that point in the desired direction.
Graph neural networks (GNNs) refer to a form of machine learning algorithm that can be applied to drug discovery to predict drug-disease associations, drug-repurposing, and drug response.
GNNs rely on graph data, with these graphs representing associations between chemical compounds, proteins, and so forth.
Natural language processing (NLP) can be employed to identify drugs by sorting through the scientific literature, as well as automating FDA approval steps. For instance, AI can power through countless biology papers and find seemingly unrelated associations that, say, connect findings from a recent study with those from a decade-old forgotten study. Biologists can then assess this finding as a relevant target.
Related: ChatGPT: The clinician’s new tech assistant?And according to an oft-cited review published in Drug Discovery Today, in addition to drug discovery, AI can help with other aspects of clinical management.[]
“Involvement of AI in the development of a pharmaceutical product from the bench to the bedside can be imagined given that it can aid rational drug design; assist in decision making; determine the right therapy for a patient, including personalized medicines; and manage the clinical data generated and use it for future drug development,” stated the authors.
Rate of AI-fueled drug discovery
According to an analysis by the Boston Consulting Group (BCG), biotech companies utilizing an AI-first approach have more than 150 small-molecule drugs in discovery and more than 15 in clinical trials.
Overall, the AI pipeline is ballooning at an annual rate of nearly 40%.[]
BCG sees a critical role for AI in pharmaceutical development: “Given the transformative potential of AI, pharma companies need to plan for a future in which AI is routinely used in drug discovery. New players are scaling up fast and creating significant value, but the applications are diverse and pharma companies need to determine where and how AI can most add value for them. In practice, this means spending the time needed to understand the full impact that AI is having on R&D, which includes separating hype from actual achievement and recognizing the difference between individual software solutions and end-to-end AI-enabled drug discovery.”
Related: AI and healthcare: Who’s footing the bill?AI in action
Bill of Health, a blog out of Harvard Law School that focuses on health law, health policy, and bioethics, highlights several intriguing examples of AI drug discovery.[]
In July 2021, AlphaFold, which is an AI system developed by Google DeepMind, predicted the protein structures for 330,000 proteins. Among these proteins are the 20,000 in the human genome. To date, this system includes more than 200 million proteins, or nearly all documented proteins in the world.
In February 2020, the first-ever AI-discovered molecule entered phase I trials. Insilico Medicine achieved this feat in far less time and far less expensively than with conventional preclinical discovery.
In January 2023, using generative AI, AbSci formed and validated de novo antibodies in silico. And finally, in February 2023, the FDA awarded the first Orphan Drug Designation to a drug designed by AI, for which Insilico Medicine is planning phase 2 trials.
Current limitations
Several challenges still limit the current capabilities of AI in drug discovery, including growth, scale, diversity, and uncertainty of data. Current models train with small data sets, with a paucity of experimental validation. It’s also not yet possible to predict efficacy, adverse effects, and other complex properties of drugs, per the authors writing in Drug Discovery Today.
What this means for you
AI is revolutionizing various fields, including drug discovery. AI can be used to facilitate many stages of the process more quickly and cheaply. AI may also be able to aid with precision-medicine strategies and the management of clinical data used to develop future drugs. To date, a handful of drugs discovered by AI are in clinical trials, which is highly encouraging.