Pharma Companies Finding Their Way with AI

By Deborah Borfitz

July 15, 2025 | Massive progress has been enabled by individual applications of artificial intelligence (AI) and machine learning (ML) in drug discovery, AlphaFold being one of the most prominent, due to their ability to predict the type of molecules that should have a positive impact on patient care. Far less evidence exists yet to suggest that these capabilities have translated into “profound improvements in drug discovery efficiency or in our ability to prosecute complex targets with complex molecules,” according to Anthony Bradley, Ph.D., Assistant Professor in the Department of Chemistry at the University of Liverpool. 

As he sees it, solutions are being devised for the “easy-to-articulate” challenges rather than the most pressing and important ones that companies should be working on and investing in. It would be hard to argue that AI and ML have already had a “dramatic transformational impact” on drug discovery when the cost of identifying and developing new medicines is continuing to rise, Bradley points out. 

“Companies are often still doing drug discovery the way they used to, just with these tools added on rather than altering the [overall] process,” says Bradley. That viewpoint may or may not gain a consensus at the upcoming Discovery on Target conference in Boston where he’ll be moderating a sizable panel discussion on impactful applications of AI and ML in the drug discovery field. 

Some of the biggest players will be weighing in on the topic, including representatives from X-Chem, MilliporeSigma, Insilico Medicine, Chemotargets, NVIDIA, Psivant Therapeutics, and Amazon Web Services (AWS). Presently, big-pharma companies are committing resources to AI and ML without fully knowing how to innovate within their legacy organizational structure, Bradley says.  

People tend to be either “super-evangelists” or “super-skeptical” when it comes to AI/ML technologies and, as he sees it, “there is truth on both sides.” The big software providers are developing remarkable tools, but those tools do not guarantee an improvement in efficiency.  

Two-hat Solution

Bradley has spent much of his career endeavoring to create efficiencies in the drug discovery process from both the academic and industry sides. As a postdoctoral research fellow at UC San Diego, he worked at the Protein Data Bank building computational tools to compress big data and enable rapid analysis of protein structural data. 

He subsequently led a University of Oxford project developing computational and experimental automation for hit-to-lead in fragment-based drug discovery. In 2018, he joined a young AI-driven pharma company (Exscientia) that was then employing about 20 people. Bradley was a key part of its PKC-theta immune modulating drug target project that is now in phase 1 clinical trials with Bristol Myers Squibb. 

As Exscientia grew into a 500-person operation, Bradley assembled a physics-based modeling group that used molecular dynamics and quantum mechanics along with structure-based automation to improve the efficiency of the company’s drug discovery processes. He left Exscientia early last year to simultaneously build a London-based startup (DaltonTx) and a research lab at the University of Liverpool “to do the same things but on slightly different timescales.” 

The research lab is designed to answer some of the longer-term questions in the drug discovery space, while DaltonTx aims to quickly translate AI research into technological solutions addressing some of the harder drug discovery challenges. DaltonTx is currently in stealth mode but has a team of 15 that has grown over the past year, he reports. 

Finding Meaningful Signals

X-Chem, a prominent life sciences company, has DNA-encoded libraries and AI-based methods giving it super-high throughput data generation capabilities, says Bradley. The company has had a direct and demonstrably successful track record in generating novel hits. 

The industry has broadly found DNA-encoded libraries useful for large-scale screening to find novel chemical matter at the beginning of a project, Bradley continues. “The question is to what extent AI and ML can make use of that data because it can be quite noisy data.” X-Chem has developed a series of AI tools to get more from this data, he adds. 

MilliporeSigma (life science business of Merck KGaA) produces cloud-based software and has chemistry capabilities. Ashwini Ghogare, Ph.D., the company’s executive director and head of AI and automation for drug discovery, has been building the capability over the past three years and self-describes as an “intrapreneur.” 

Moving forward, the exciting possibility is that more meaningful signals will be extracted from high-throughput data, something Insilico Medicine is especially skilled at, says Bradley. “Insilico is at the cutting edge of applying AI to drug discovery and has called increasingly for improved benchmarks... to demonstrate efficiency, and they have some nice, published examples of delivering high-quality candidates for projects.” 

Petrina Kamya, Ph.D., vice president of Insilico Medicine and president of Insilico Medicine Canada, is the company’s global head of AI platforms. Toronto and Montreal both have a lot of AI talent, Bradley points out, as does London.  

Problem Discussions

Chemotargets, a computationally oriented biotech company, is also producing software intended to improve the efficiency of drug discovery. The company is focused on how to ensure molecules have good intellectual property protections and understanding known knowledge about them and ways of predicting their safety, in addition to running some AI-driven drug discovery projects itself, says Bradley.  

Another biotech, Psivant Therapeutics, is designing novel small molecule therapeutics using AI and what is known about quantum mechanics and Newtonian physics to understand how proteins interact with each other, he says. Woody Sherman, Ph.D., the company’s chief information officer, previously worked for drug development software company Schrödinger and drug discovery company Silicon Therapeutics that was acquired by Roivant Sciences in 2021. After a year of developing a computational platform (QUAISAR) to run at scale across dozens of drug discovery projects, Psivant was born. 

NVIDIA is of course the dominant force in the hardware used for deep learning and is now driving the development of novel algorithms as well. Notably, Bradley says, it is building NVIDIA Inference Microservices, or NIMS, providing optimized “blueprints” for accelerating the deployment of AI models. It is also investing in an assortment of companies, primarily in the AI and data center spaces that are important to individual tool development.  

The NVIDIA GPU Technology Conference in March featured an increasing amount of dialogue about life science problems, says Bradley. The discussions extended to NVIDIA's BioNeMo platform for AI-powered drug discovery, including its integration with Sapio Sciences' drug research software, during the conference and keynote by CEO Jensen Huang. 

Embracing the Cloud 

All but one of the top 20 pharma companies are building their drug discovery capabilities using the cloud computing platform of Amazon Web Services (AWS), and Insilico Medicine also sells their software through the AWS Marketplace. “Scaling data and scaling compute are both obviously crucial for AI, and the great thing from what we’ve observed is that the conversation has moved on from people being very afraid of the security implications of the cloud,” says Bradley. 

Today, there is growing recognition within many pharma companies that the cloud is in fact more secure than building everything on-premises and resolving security issues on their own, Bradley says. Given the pace of change, it is also a necessary mindset. “There is no way you’d be able to keep pace with the growth in available hardware.” 

AWS, despite its popularity, is not the only cloud computing option. There’s also Google Cloud Platform (GCP), Microsoft Azure, and big Chinese companies such as Alibaba and Baidu. NVIDIA also offers some cloud computing resources, making it both a customer and competitor of AWS, says Bradley. 

What’s exciting about Azure is that it has a significant focus on chemistry and other scientific applications, he says. “It is growing a really exciting application layer.” GCP has a family of products to help with large-scale data searches as well as Tensor Processing Units enabling accelerated AI/ML computations.

Restructuring Question

Answers about how to incorporate AI and ML into the drug discovery processes of big-pharma companies likely won’t come easy, says Bradley. “One of the biggest challenges is going to be how to apply [those technologies] within their organization, including whether or not they need to restructure completely or not.” 

Companies are typically organized by therapeutic area (e.g., oncology and respiratory), each with their own department. For AI drug discovery purposes, this is not the ideal setup, Bradley says. “What you often end up with are two AI groups... doing roughly the same thing, or you try to have some organization in the middle, which can bring its own challenges.” The drug discovery process of a startup like Insilico Medicine, in contrast, is structured around AI. 

Novo Nordisk has publicly talked about building a new AI and digital innovation unit within the company, he reports. GSK has brought in Kim Branson, a world-renowned AI/ML expert, to build its computational capabilities. Roche, meanwhile, has disclosed it has a “lab in a loop” mechanism bringing generative AI to drug discovery and development, and is establishing innovation hubs.