The Role AI Can Play in Pharmaceutical Discovery

The pharmaceutical industry has long been at the forefront of scientific innovation, driving advancements in medicine that aim to improve the quality of human health and lifespan.

The race to develop vaccines and treatments for COVID-19 shows that innovation in pharmaceuticals can have an enormous impact on public health.

However, drug discovery is expensive, time-consuming, and fraught with uncertainty. It is estimated to cost approximately $2.5 billion to bring a new drug to market (PDF), according to a report by UK-based health research foundation The Wellcome Trust. Scientific and technical challenges mean that the probability of discovering a new drug and bringing it to clinical trial is around 35%, and the likelihood of successfully progressing from Phase 1 trials to receiving regulatory approval is just 9-14%, with the process taking an average of 12-15 years. This is a major barrier to innovation, and market forces typically focus on areas with the potential for large commercial returns.

This opens the way for artificial intelligence (AI) to profoundly affect the pharmaceutical industry by accelerating the drug discovery process, reducing costs, and increasing the likelihood of success.

The Potential for AI to Advance Drug Discovery

Over the past decade, AI has made major advancements in machine learning (ML), deep learning, neural networks, and generative AI. The potential to apply these techniques to drug discovery is gaining attention from the pharmaceutical industry, technology companies, investors, and biomedical research financiers.

AI could transform the economics of drug discovery and innovation, allowing scientists to discover new medicines to treat or prevent a wider range of conditions and patients than is currently possible.

Major AI Use Cases

There are opportunities to incorporate AI into nearly every drug and vaccine discovery stage. There are vast amounts of data that algorithms can synthesize. This would not replace the role of experienced scientists, as it requires medicinal chemists to interpret the output of the models and frees up time to focus on higher-value tasks.

Target identification and validation: AI algorithms can analyze vast amounts of data, including genomics and clinical data, to identify potential drug targets more efficiently than traditional methods. This reduces the time required for drug discovery research and increases the chances of identifying successful targets.
Drug design: AI can help design new drug candidates by predicting chemical structures and properties and optimizing existing drug molecules for improved efficacy and safety. Automated systems can analyze thousands of chemical compounds for their potential as drug candidates, reducing how long it takes to identify promising leads.
High-throughput screening: AI-powered robotics and image analysis can speed up the screening of compounds.
Clinical trial optimization: AI can help streamline clinical trials by identifying suitable patient groups, predicting responses, and optimizing trial protocols. This reduces costs and makes it more likely that trials succeed.
Drug repurposing: AI can analyze vast datasets of drug interactions, disease pathways, and patient data to identify existing drugs that can potentially be used to treat other diseases. This approach can save time and resources compared to developing new drugs.
Drug safety: AI can continuously monitor patient data and adverse events to identify potential safety concerns early in the drug development. This can support the development of safer drugs and reduce the recalls once launched onto the market.

The medical industry is moving ahead of academics in its efforts to deploy AI, led by biotechnology companies building their R&D workflows around AI tools and pharmaceutical companies adopting AI in drug discovery.

Companies such as Absci and Antiverse are advancing AI-driven ‘de novo’, or new, antibody design. The algorithms used to design antibody sequences are trained on universal antibody properties to indicate what a functional antibody that binds to a disease target might look like. They use this data to develop a new design, like examining a lock to design a new key to open it.

This could reduce the time it takes to bring new drug candidates to trial by more than half while increasing the probability of success, according to Absci. The company validated its antibodies against more than 100,000 antibodies and found its hit rate to be 5-30 times greater than biological baselines.

Pharmaceutical companies are working with AI-led drug discovery companies to advance their development processes. For instance, AstraZeneca and Merck have partnered with BenevolentAI, while Sanofi has a $1.2 billion strategic research collaboration with Insilico Medicine. Merck is also working with Exscientia. Merck has identified three potential clinical development drug candidates with potential oncology, neurology, and immunology.

Challenges and Ethical Considerations

While there is promise in applying AI modeling in drug discovery, it has yet to be demonstrated at scale across populations and diseases. Challenges and ethical considerations include algorithm bias, the need for high-quality source data, and regulatory hurdles. The use of AI in healthcare also raises concerns about privacy, data security in training models, and the potential for bias in decision-making.

To reach the full potential of AI in addressing global health issues, there is a need for a better understanding of its current applications and limitations and the barriers the industry faces, Wellcome states in its report. The sector is developing rapidly but unevenly, with over 80% of publications in the last five years focused on applying AI to understanding disease, target discovery, and small molecule optimization. Financing from private investors is still skewed towards the most commercially viable areas, with around 70% of AI-related investments in the last five years being made in oncology, neurology, and COVID-19.

To address these challenges, the pharmaceutical industry must collaborate with regulatory bodies, ensure transparency in AI algorithms, and prioritize data privacy and security. Ethical frameworks are required to guide the responsible use of AI and ensure the benefits are accessible to all patients. Initiatives must be developed to support the application of AI in research of less commercially attractive conditions and provide access to researchers in lower-income countries.

This will help address barriers to adoption, such as trust in AI algorithms, the validity of their conclusions, and concerns about the implications for scientific research and broader society.

There are efforts underway to remove these barriers. For instance, the World Economic Forum and the University of Oxford have established the AI Governance Research Group to improve understanding of the development of AI and manage risks across various settings, including in medical research. The Wellcome-Sanger African Genome Variation Project is working to provide a basic framework for generating high-quality genomic datasets in sub-Saharan Africa. And the H3D Foundation, which aims to support African researchers in drug discovery and development, provides courses on the use of AI in discovering drugs to treat locally relevant infectious diseases. In the US, the National Institutes of Health (NIH) offers grants to standardize datasets for use in machine learning.

The Bottom Line

The role of AI in pharmaceutical discovery is poised to transform the industry by accelerating research processes, reducing costs, and increasing the likelihood of success in developing new drugs. Responsible and transparent adoption of AI has the potential to lead to breakthroughs in drug discovery.

At the same time, barriers risk concentrating the benefits of AI to already data-rich and commercially attractive treatment areas. Concerted action will be needed to shape the application of AI to allow populations worldwide to benefit from bringing new treatments to market more efficiently than ever before.