AstraZeneca, Pfizer and more are leveraging the computational power of AI to better design trials, predict the potential efficacy and safety profiles of their molecules and synthesize massive multi-omic information to gain a more complete understanding of challenging cancers.
In the grand scheme of cancer, personalized treatment is a relatively young approach. Over the past few decades, numerous advances have pushed the boundaries of what’s possible—most recently, artificial intelligence.
According to some records, precision therapies have their roots in the 1990s, when researchers first discovered that some drugs could be more effective in certain cancers carrying a specific genetic or biomarker profile. Before then, patients with cancer were managed through chemotherapy and radiotherapy—often indiscriminate chemical and radioactive assaults on the body that led to severe side effects.
“Cancer is a deeply individualized disease,” Mohan Uttarwar, CEO and co-founder of 1Cell.Ai, a California-based biotech that leverages AI technology to improve cancer diagnostics and, ultimately, care, told BioSpace in an email. “We believe every oncology solution should be precision-based.”
The advent of precision oncology was made possible by the parallel rise of various technologies, such as DNA sequencing, sophisticated computational models that helped scientists identify molecules that could distinguish one specific subset of cancer from another, and accurately stratify patients according to these biomarkers.
Today, there is one particular emerging technology that, if maximized and deployed to its full potential, could push the field of precision oncology even further forward, opening up better diagnostic and therapeutic horizons: Artificial intelligence (AI).
In precision oncology, and in drug development more broadly, AI is used primarily as a means to analyze massive sets of data beyond human ability, identify pertinent patterns and make well-grounded predictions.
“AI is increasingly central to pharmaceutical R&D,” Ofer Sharon, CEO of OncoHost, told BioSpace in an email. “AI enables companies to move from intuition-driven to data-driven drug development.”
Arun Krishna, head of U.S. oncology at AstraZeneca, agreed. “The drug discovery process has rapidly changed with AI,” he told BioSpace at the 2025 American Society of Clinical Oncology (ASCO) Annual Meeting last month.
Drug Discovery and Patient Selection
Companies can leverage AI to dig through entire cancer genomes and find specific mutations that are ripe for targeting, Uttarwar said. Alternatively, they can also use these analytic models to find new and yet-unused biomarkers to better identify patients that are likely to respond to treatment, resulting in better, more streamlined studies—“critical for clinical trial success,” he said.
Conversely, drugmakers could also harness AI technologies to design better therapies like antibody-drug conjugates, he added.
For instance, companies can leverage their vast biochemical datasets and train their analytic models to predict the potency a specific molecule is likely to have, or what toxicities can be expected. AI can also be used to anticipate potential drug-drug interactions that could not only affect the clinical efficacy of a drug candidate but also pose safety risks and worsen the patient experience.
Typically, collecting these data requires heavy, labor-intensive and time-consuming experimentation, often involving animal models, in turn raising ethical concerns. AI could potentially minimize or outright remove these barriers. The FDA recently recognized the value of AI, announcing in April that it would begin phasing out animal testing for monoclonal antibodies and other therapies in favor of “AI-based computational models” and human organoid lab models.
With this shift, “we can get safer treatments to patients faster and more reliably, while also reducing R&D costs and drug prices,” FDA Commissioner Marty Makary said at the time.
Several biopharma companies have already started incorporating AI models into their processes. AstraZeneca, for one, has put more than $1 billion on the line in a series of AI partnerships over the past few years. For Krishna, predictive AI in drug discovery is “the holy grail.” This is what we should be getting right, he told BioSpace.
The drug discovery stage is where AI can, hopefully, identify potentially useful molecules much faster than the months or years the process would take in wet labs, Krishna said. “Now it can be 30 days or less.”
AI can also be useful in patient selection. Last year, AstraZeneca turned to AI to better stratify lung cancer patients in an effort to build a better case for one of its most closely watched assets.
In September 2024, seeking to explain a Phase III lung cancer failure for the Daiichi Sankyo-partnered antibody-drug conjugate Dato-DXd, the pharma employed an AI model to determine which patients are more likely to respond to treatment. The effort yielded an AI-derived biomarker, called TROP2-QCS, which deals with expression levels of TROP2 in patients. Dato-DXd is now approved in breast cancer and marketed as Datroway.
Subsequent exploratory analyses showed that patients with the TROP2-QCS biomarker saw a 43% drop in the risk of disease progression or death when treated with Dato-DXd versus docetaxel. Meanwhile, the effect of Dato-DXd was notably weaker in those without the biomarker—risk reduction was only 25% versus docetaxel. This exercise, though preliminary, demonstrated how companies can use AI to better direct their drug development efforts.
Meanwhile, Pfizer is “building the next generation of tools to use across the preclinical and clinical development spectrum,” Jared Christensen, current vice president and head of internal medicine and infectious disease statistics, told BioSpace in a previous interview. Pfizer also has a predictive ML research hub, which is responsible for creating novel predictive models and tools.
Novartis is also active on the AI scene. Last year, the Swiss pharma put down $65 million up-front in a partnership with Flagship Pioneering’s Generate:Biomedicines, looking to leverage the startup’s AI platform, which “infers generalizable principles of biology” to design novel medicines. Details of the collaboration are scant: It is not known what targets the partners will prioritize, nor how they plan on leveraging AI.
Beyond the drug development pipeline, OncoHost’s Sharon, who led the personalized medicine session at the Biomed Israel 2025 conference in Tel Aviv last month, said AI could also help develop “advanced companion diagnostics,” which he claimed will “match patients to treatments with greater precision.”
Futures and Frontiers
Excitement over AI’s potential is palpable, especially as generative AI—a subset of the field that goes beyond analysis and predictions, instead dealing with the creation and generation of wholly new models based on existing information—breaks into the mainstream.
In August 2023, Insilico Medicine brought the first drug fully created with generative AI into Phase II clinical trials. Last Fall, the biotech reported positive results from that trial in idiopathic pulmonary fibrosis. The following month, Generate:Biomedicines collected $273 million in series C financing for its pipeline of preclinical and clinical protein therapeutics targeting everything from asthma to advanced solid tumors.
Though still a very immature area, generative AI, according to Sharon, has the potential to “redefine the pace and scope of innovation” in drug development, affecting a variety of domains such as de novo molecule design and sophisticated synthetic biology, in turn “unlocking new therapeutic possibilities once considered out of reach.”
Uttarwar agreed: “As generative AI matures, its ability to simulate biological interactions and propose entirely new therapeutic molecules will dramatically reduce the time and cost involved in drug discovery.”
Generative AI, with its high-powered computational capabilities, can also allow cancer drug developers to access what Uttarwar called “multi-omics datasets.” Both Uttarwar and Sharon agreed that the next frontier for precision oncology is multi-omics, a field of study that draws and synthesizes information from various -omes, including not just the genome but also the transcriptome, the proteome and the metabolome—referring to the entire set of RNA transcripts, proteins and metabolites expressed or otherwise present in a cell or tissue at any given time.
Such an approach could provide a more comprehensive picture of a disease beyond singular dysregulated genes or signaling pathways.
“Genomic alterations tell part of the story, but proteins reflect what’s actually happening in real time within the tumor microenvironment,” Sharon explained. “AI plays a vital role here by integrating complex datasets across omics layers, uncovering patterns invisible to the human eye, and generating actionable insights.”
Just as the broader field of generative AI has seen rapid, explosive growth in recent months, so too will the application of the technology in medicine—and in cancer more specifically, according to Sharon. The increasing integration of AI into the field will be facilitated by what he expects will be the “growing regulatory acceptance of AI-defined biomarkers” as well as an intentional push from the industry to train models and incorporate AI tools into their workflows and trial designs.
The Data Dilemma
Despite its potential, there is the dilemma that AI is still very much an emerging science, experts who spoke with BioSpace agreed.
“AI models are only as good as the data they’re trained on,” Sharon said. “Inconsistent or biased datasets can limit generalizability” of these AI models.
Uttarwar agreed, noting that harmony of information is of particular importance. “Genomic, proteomic, and imaging data often come from different sources, formats and protocols,” Uttarwar explained. “If not properly harmonized, they can introduce noise into AI models.”
Therefore, better data management is needed across the drug development pipeline. Doctors, scientists and technologists have to better organize and standardize their information to make sure their AI models are trained properly. Otherwise, “the insights generated aren’t actionable or trustworthy,” Uttarwar said.
Aside from its AI models, 1Cell.Ai is also working on solutions to the data dilemma. The company offers a data management platform called iCore, which ensures compliance with U.S. and European Union health data collection regulations, according to Uttarwar. iCore—which was built in collaboration with a handful of academic institutions including Stanford and MD Anderson—“allows us to harmonize diverse datasets like genomics, proteomics, and imaging to power AI models,” Uttarwar said.
On a deeper level, however, Sharon said the fundamental challenge to AI in drug development is one of trust. There is a “need for transparency in AI decision-making,” he contended, which in turn can foster trust among those who are on the interpreting end of these models—clinicians and regulators.
Drugmakers and AI model developers have shown some openness through “more collaborative data-sharing initiatives, development of explainable AI models, and regulatory frameworks that guide validation and deployment,” Sharon explained. But there is much left to do.
“Ultimately, success will depend on how well AI tools integrate into clinical and operational workflows, not just how advanced the algorithms are,” Sharon added.
For the field as a whole, Sharon believes this year will be pivotal. “We expect 2025 to mark a turning point, with the first AI-discovered or AI-designed therapeutic [oncology] candidates entering first-in-human trials, signaling a paradigm shift in how therapies are developed.”
