How Is AI Speeding Pharma’s Journey Toward Precision Medicine?
Discover how AI is reshaping drug development by enabling faster precision medicine through smarter data use, patient targeting and clinical trial design.
Complete the form below to unlock access to ALL audio articles.
While generative artificial intelligence (AI) burst into the public consciousness a little more than two years ago with the release of ChatGPT, AI and machine learning (ML) have been part of drug research and development for far longer. AI and ML are now being applied to every stage of biopharmaceutical R&D including discovery, compound development, trial design, regulatory submission, supply-chain optimization and postmarket surveillance.
In the last two years, several published, peer-reviewed articles have referred to precision medicine as “futuristic” – a state that has not yet been fully realized – but drug discovery is rapidly becoming more patient-centered with each new application of AI.
Although ChatGPT and other generative AI engines built into smartphones, Internet search engines and social media platforms have captured public attention, generative AI is only one flavor of the technology. The pharma industry has been using predictive modeling for years. Investigators are turning to increasingly advanced algorithms to identify patient subgroups, forecast mechanisms of action, predict differential drug responses including toxicity and even adjust dosing strategies.
Popular use cases and therapeutic areas
In a 2023 report, the UK-based Wellcome Trust identified five key families of use cases for AI in drug discovery:
- Drug target identification and validation
- Small-molecule design and optimization by pinpointing “hit-like or lead-like small molecule compounds”
- Design and optimization of vaccines, particularly mRNA vaccines
- Design and optimization of antibody structures and properties
- Evaluation of safety and toxicity of promising compounds
According to the report, more than 80% of published articles on AI-enabled drug discovery in the preceding five years were related to understanding disease, target discovery and optimization of small-molecule compounds.
Of note, the organization said there has been a dearth of publicly available data on safety and toxicity to train AI models. Interviews with experts in the field included in the report mentioned the challenges of predicting safety and toxicity based on in silico data without sufficient supporting clinical validation.
The Wellcome report also cited a 2021 Nature article describing how AlphaFold, an AI/ML algorithm from Google sister company DeepMind, predicted the three-dimensional structure of human proteins with “atomic” accuracy.
About 70% of private-sector investments in AI for drug discovery between 2018 and 2022 were in the “commercially tractable” therapeutic areas of oncology, neurology and COVID-19, Wellcome added.
As noted in the American Journal of Managed Care (AJMC), numerous drug companies have already incorporated AI into drug R&D processes, particularly to improve target identification, molecule discovery and patient recruitment for trials.
“The integration of AI into the drug discovery process offers immense potential for accelerating drug development, reducing costs, and improving patient outcomes,” the October 2024 article concluded. “However, the successful implementation of AI requires addressing knowledge gaps, ensuring data quality, and navigating regulatory challenges.”
At least one major drug company has said that it is looking at AI for developing precision therapeutics in oncology, immunology and neuroscience, all popular therapeutic areas for AI application among Big Pharma. However, the Wellcome report cited COVID-19 response as a shining example of the power of this evolving technology.
In the early days of the pandemic in March 2020, AI was used to research monoclonal antibodies derived from convalescent plasma of COVID-19 patients. About 2,000 potential candidates were quickly narrowed down to 24, and the most promising compound, bamlanivimab, entered into clinical trials within three months.
The U.S. Food and Drug Administration (FDA) granted Emergency Use Authorization to bamlanivimab in November 2020, just eight months after research commenced. Though the FDA revoked the authorization in April 2021 after subsequent SARS-CoV-2 variants proved more resistant to the therapy, this episode highlighted just how rapidly drug-makers could develop narrowly targeted treatments.
Reality and promise
AI analyzes large genomic and multiomic datasets faster than ever before to help target therapies. A study published in Science in September 2023 explained how bioinformaticians have been able to build upon each other’s work to extend the capabilities of popular AI algorithms including AlphaFold to improve prediction of pathogenicity of missense variants in the human proteome.
A May 2024 review article published in the journal Fundamental Research suggested that AI, in the form of unsupervised machine learning, is a more efficient way of generating patient clusters and phenotype candidates than human training of datasets in the development of gene therapies. AI is also useful for optimizing drug dosing based on individuals’ genetic profiles, the international team of authors wrote.
A July 2023 article in Pharmaceutics noted that an AI technique called clustering – which groups similar datapoints to find subgroups of patient data, gene expression profiles, chemical structures and other pertinent information – is useful for identifying drug targets and stratifying patients. Other AI algorithms can sift through complex stores of data in search of anomalies.
“AI is being utilized to advance precision medicine approaches. By analyzing patient data, including genomics, proteomics, and clinical records, AI algorithms can identify patient subgroups, predict treatment responses, and assist in personalized treatment decision-making. AI also contributes to the development of biomarkers for disease diagnosis and prognosis,” the authors, academic researchers from India and Northern Ireland, wrote.
Why Automation is Integral to the EU-OPENSCREEN Compound Collection
This case study highlights how EU-OPENSCREEN implemented a fully automated workflow for compound dissolution, aliquoting, storage and plate reformatting, meeting the scale and speed of modern screening demands.View App Note / Case Study
“AI might revolutionize the pharmaceutical industry in the future to accelerate drug discovery and drug development,” they continued, offering a clear caveat with their word choices. “AI-enabled precise medicine could categorize patients, predict therapy responses, and customize medicines by analyzing genomes, proteomes, and clinical records.”
They saw promise in what they called “virtual screening” to accelerate identification of lead compounds by selecting therapeutic candidates with the necessary characteristics from massive chemical databases.
“Scientists may create innovative compounds with target-binding characteristics using deep learning and generative models, improving medication effectiveness and lowering adverse effects,” the Pharmaceutics article said. “AI improves clinical trial design, patient selection, and recruitment. AI algorithms will use electronic health records, biomarkers, and genetic profiles to find appropriate patients, lower trial costs, and speed up approval.”
But, as the AJMC article stated, AI performance is only as good as the data the algorithm is trained on — the old “garbage in, garbage out” maxim from computer science.
One potential tripwire is data bias. Writing in NPJ Digital Medicine, investigators from the Berlin Institute of Health at Charité and Harvard Medical School noted that AI data is often trained on datasets that have under-represent women, people of color, low-income groups and other historically disadvantaged demographics. “For example, an AI algorithm used for predicting future risk of breast cancer may suffer from a performance gap wherein black patients are more likely to be assigned as ‘low risk’ incorrectly,” they said.
Regulation
Regulation rarely keeps up with technological advancements, and AI is no different in this regard. While enabling legislation might be behind the times, regulators are at least attempting to stay ahead of the curve by offering guidance to constituents including drug developers.
For example, the European Medicines Agency (EMA) recently spent more than a year developing a “reflection paper” offering guidance on using AI/ML in pharmaceutical discovery, development, approval submissions, manufacturing and outcomes assessment. This guidance document spells out how existing European Union and national laws cover AI/ML across pharma lifecycles, including when technologies might be treated as medical devices.
The EMA pointed out the importance of paying attention to data quality and relevance in training algorithms. ML developers should take a “human-centric approach” to design be able to explain their methodologies to allow the agency to review and monitor “black box” models, the paper said.
“This requires not only that active measures are taken during data collection and modelling … but also that both user and patient reported outcome and experience measures are included in the evaluation of AI/ML tools when they interface with an individual user or patient,” the EMA explained.
