AI Drug Development: How AI Is Changing Medicine - Built In

Perhaps no industry has been transformed by artificial intelligence more than healthcare, especially drug development. The scientific community recognized this progress when it awarded the 2024 Nobel Prize in Chemistry to Google DeepMind for AlphaFold2 — an AI model that can predict protein structures. Tech companies have since been inspired to drive faster and more accurate drug development with AI, even designing supercomputers to perform more complex calculations on reams of healthcare data. 

Despite the growing popularity of AI-powered drug development, how exactly this process works and the host of issues it introduces are less widely understood among the general public. Here’s what to know about this revolutionary technological approach and why it continues to garner so much excitement amid some ongoing challenges. 

What to Know About AI in HealthcareAI in Healthcare: Uses, Examples and Benefits

What Is AI Drug Development? 

Before exploring how artificial intelligence is used in drug development, a basic overview of the process might be helpful. Overall, drug development can be broken down into five main stages, starting with a molecular compound that has medicinal potential and ending with a full-fledged drug that can treat a disease or condition: 

1. Drug Discovery: Researchers find new ways to treat diseases by creating molecular compounds or new technologies, as well as repurposing existing treatments. 
2. Preclinical Research: In lab settings, researchers test new compounds in animals and human cells to measure their toxicity, or the ability to cause harmful side effects. 
3. Clinical Drug Development: Researchers then test the compounds in actual people, conducting a series of clinical trials that use progressively larger sample sizes. 
4. FDA Review: If a compound is consistently safe and effective in clinical trials, researchers submit an application to the FDA to get approval for public use. 
5. Post-Market Monitoring: The FDA tracks all approved drugs and monitors reports of any beneficial or harmful effects in patients, adding safety measures as needed.  

More and more, artificial intelligence is being used to automate various aspects of drug development. For instance, researchers can use machine learning, deep learning or natural language processing to analyze vast amounts of health data, predict how certain compounds could affect the human body and identify ideal candidates for clinical trials, improving both the speed and accuracy of the process.  

According to Ryan Kelly, chief communications officer at biotech company Recursion, AI removes the guesswork from this process. Researchers don’t have to worry then about wasting time and money tediously testing millions of different compounds in search of the one that targets a particular disease. 

“It typically takes 10 years or more and costs $2 billion or more to bring a drug to market, and 90 percent of drugs fail in clinical development. AI drug discovery starts with data, not a hypothesis, and allows us to take an unbiased approach,” Kelly told Built In. “This allows us to move beyond the slow, trial-and-error chemistry process and quickly create differentiated compounds.”

How Does AI Drug Development Work? 

That said, building AI models that specialize in drug development is an entirely different task altogether. Each company has its own methods, tools, infrastructure and partnerships, but the process always starts with data. Researchers can compile data from sources like electronic health records, public databases or experiments. In Recursion’s case, Kelly notes how the company gathers high-quality data by conducting millions of wet-lab experiments each week. 

Next comes the training phase. AI models can be trained on labeled data through supervised learning or raw, unlabeled data through unsupervised learning. Researchers may also employ techniques like reinforcement learning, training a model to complete a task in a controlled environment through trial and error. Companies can even increase the amount of available compute to fuel more powerful AI models that can handle larger volumes of data, as demonstrated by Recursion’s supercomputer. 

Once a model has been thoroughly trained and tested, it can be used to generate molecular simulations, determine the most promising compounds, select compatible patients for clinical trials and more. Researchers can then use generative AI to accelerate these tasks, greatly cutting down the time it takes to discover and develop drugs. 

“With automated labs for gathering data, and a supercomputer to drive our modeling efforts, we’re able to speed up the process of discovery and development, and optimize every part of the process, all the way through clinical development,” Kelly said. “This is all with the goal in mind of bringing better medicines to patients.”

More on AI and Drug DevelopmentWhat Is Bioinformatics?

Companies Using AI in Drug Development  

These are some of the companies that have spearheaded efforts to integrate AI into drug development, promising to revolutionize the healthcare industry. 

Insilico Medicine

Insilico Medicine develops drugs with Pharma.ai, its generative AI platform that offers a suite of AI models and tools for targeting treatments for diseases, modeling disease behavior, conducting aging research and more. The company has created a framework to improve the accuracy of drug discovery and a longevity board to oversee the process in aging research, specifically. These efforts have begun to attract attention, with Eli Lilly agreeing to a $2.75 billion deal to help Insilico launch its drugs globally. 

AstraZeneca

AstraZeneca focuses on cancer treatment, cardiovascular disease and respiratory care, among other areas. It has spent years refining its AI drug development methods, with national healthcare systems having already adopted the company’s technology. To accelerate the process further, AstraZeneca agreed to share real-time clinical trial data with the Food and Drug Administration rather than waiting to report results to federal officials after it’s completed its analysis, potentially shaving months to years off drug development.  

Nimbus Therapeutics

Nimbus Therapeutics combines internal knowledge and an extensive network of partnerships to produce what it calls “small molecule medicines.” Its methodology includes running molecular simulations and employing machine learning tools to predict treatment effectiveness and better target diseases. Nimbus is another company partnering with Eli Lilly, which has committed $1.3 billion in milestone payments if Nimbus successfully develops an oral treatment for metabolic diseases, primarily obesity. 

Recursion

Recursion made waves in 2024 when it teamed up with Nvidia to build the pharmaceutical industry’s largest supercomputer, which processed immense volumes of data to train even more powerful AI models. This initiative has culminated in LOWE — an AI agent that oversees complex workflows on the Recursion OS platform. As a result, biologists and chemists can collaborate using a single interface and natural-language commands, more quickly visualizing data and testing algorithmic predictions.  

Tempus 

Tempus was founded with the mission of collecting healthcare data to create personalized cancer treatments. Having compiled a massive database over the past decade, the company can leverage AI to predict how patients might respond to different treatments, determine the ideal participants for clinical trials and support doctors with an AI-powered clinical assistant. Tempus’ research on using biomarkers to inform its oncological medicines and treatments has earned it partnerships with Merck and Daiichi Sankyo. 

10x Science

10x Science is a rising startup that aims to evaluate new drug treatments recommended by AI tools. Researchers from Stanford University, UC Berkeley, UC San Francisco, MIT and the University of Wisconsin worked together to build AI agents that can analyze protein structures and categorize treatments based on their molecular makeup, making it faster to discern which treatments to move forward with. 

What Challenges Does AI Drug Development Face?  

Although drug development is positioned as one of AI’s most promising areas of innovation, the technique still poses a range of problems. Data quality remains a major challenge, with researchers often struggling to find enough healthcare data to train their AI models, let alone data free of missing or incorrect values. As a result, models may not capture the “complexity of human biology,” which can contribute to costly failures during clinical trials, according to Recursion’s Kelly. 

The ethical dilemma of giving AI models access to highly personal information has also divided the healthcare world. According to a 2026 survey by the American Medical Association, more than 80 percent of physicians now use augmented intelligence tools in their professional lives. Yet 86 percent believe the broader adoption of AI can’t happen without data privacy, and 88 percent believe stronger safety measures are essential. 

Policymakers and everyday Americans appear to want robust guardrails across the board as well, pushing for data center moratoriums, companion chatbot rules and privacy regulations, among other tech-related laws. If people don’t trust AI to make decisions that could impact their well-being, tech leaders may find themselves navigating an even more intricate web of regulations to get AI-developed drugs to market. 

These obstacles aren’t immovable, though. Researchers can generate synthetic data to address data shortages and ease any public distrust by adopting explainable AI — an approach that prioritizes transparency around AI models’ decisions. Such measures can improve the AI drug development process and pave the way for AI to take on a larger role in revamping how the healthcare industry operates in the years ahead.

More Concerns About AI in HealthcareAI Is America’s New Doctor. But Can You Trust It With Your Health?

How AI Drug Development Could Transform Healthcare 

AI drug development is becoming more of a reality, with the studies to back it up. Researchers at Harvard University and Boston’s Beth Israel Deaconess Medical Center found that OpenAI’s o1-preview model accurately diagnosed 67.1 percent of ER patients, surpassing two physicians, who diagnosed 55.3 percent and 50.0 percent of patients correctly. Meanwhile, scientists at the University of South Florida used a model called PanPep to predict whether human cells would activate an immune response to foreign substances in the body. 

None of this research suggests AI is anywhere close to replacing doctors or medical professionals outright, but it’s enough to amplify the hype around the technology. After all, pharmaceutical powerhouses like Johnson & Johnson are beginning to experiment with AI drug development, and Meta CEO Mark Zuckerberg’s nonprofit Biohub received $500 million in funds as it aims to eliminate all human diseases with the help of simulations. 

Tech titans aren’t the only ones benefitting from AI drug development, either. A cheaper, faster process for bringing drugs to market means researchers globally can achieve pharmaceutical breakthroughs, not just those in the “handful of resource-rich nations” that have traditionally spurred scientific innovation, according to Insilico Medicine.  Of course, partnerships like those between Eli Lilly, Insilico and Nimbus will continue to be crucial for bringing AI-developed drugs to market, hinting at a more collaborative healthcare ecosystem. 

With democratized drug development on the horizon, companies have reason to hope for a refined drug development process, and patients around the world can look forward to higher-quality, more affordable medicines as companies like Recursion double down on efforts to fine-tune their methods. 

“What’s exciting is that we are now seeing the first real ‘proof points’ that an AI-led approach to drug discovery works,” Kelly said. “Ultimately, we’re focused on not just single successful drug stories, but creating a repeatable product for developing and delivering drugs, de-risking every decision with data and having greater confidence from the earliest stages that the drugs we are developing will be transformative for patients.”