Pfizer Doubles Down on AI/ML to Bring Transformative Medicines to Patients

Pfizer’s Cambridge, Massachusetts website/courtesy of Pfizer

Artificial intelligence and machine learning (AI/ML) are key to enabling drug discovery and development, and Pfizer is leading the biopharma industry in the next wave of innovation. The company is rapidly scaling up and recruiting talent for a joint effort intended to get transformative medicines to patients faster.

The mandate is “uncompromising and extremely high-quality science,” said Sandeep Menon, chief scientific officer, AI digital sciences, SVP and head of early clinical development. BioSpace.

The vision is threefold: uncover disease biology with AI; use these insights to design the right molecules; determine the right patient population for clinical trial success.

“We’re building the next generation of tools to use across the preclinical and clinical development spectrum,” said Jared Christensen, vice president and head of early clinical development, clinical AI/ML and quantitative sciences.

Pfizer is building an “ML Research Hub” aimed at creating new predictive models and tools in what it called “a key investment.”

This team, led by Enoch Huang, vice president, machine learning and computer sciences, will collaborate with experts across the company to ensure successful application of AI/ML by designing, deploying and maintaining modern tools and techniques. This will uncover insights related to disease pathophysiology and generate relevant, testable hypotheses. The ML Research efforts will be led by Djork-Arné Clevert, who recently joined the company.

“AI/ML is woven into the fabric of drug discovery at Pfizer,” said Huang. “A sign of success is when our project teams or design chemists looking at compounds use machine learning without knowing they are using machine learning. This is what happens behind the scenes.”

“However, we need to apply AI/ML beyond drug design, and start with the patient in mind,” Huang continued. “We see enormous potential to exploit public and proprietary data sets using ML methods to better understand disease pathophysiology, potentially leading to breakthrough efficacy for patients that meaningfully change their lives.”

Numerous therapeutic applications

The innovation resulting from the collaboration will be therapeutic agnostic, Christensen shared.

“We’re going to start in areas where we already have a foothold,” he said. Pfizer’s core therapeutic focus areas are internal medicine, inflammation and immunology, oncology, vaccines and rare diseases.

The lift will be lighter in oncology, where there has already been significant progress in precision medicine. Pfizer plans to build on these gains to better understand patient populations and stratifications, Christensen noted.

“We are looking for data-rich indications to train the models. The opportunity before us is to inform and influence target prioritization and patient stratification with AI/ML, just as we have done in chemistry,” he said.

In internal medicine, Christensen emphasized heart failure, diabetes, and nonalcoholic steatohepatitis where there are large populations and more data is accumulating every day. The same can be said for inflammatory and immunological diseases such as rheumatoid arthritis, Crohn’s disease and ulcerative colitis.

Pfizer intends to leverage this data, along with relevant biomarkers and next-generation sequencing datasets, to better understand where its drugs can have the most impact.

“I firmly believe that diseases that we now call one thing will continue to subdivide based on biomarkers and clinical phenotypes,” Christensen said. “I believe that kind of revolution is going to continue to come to other diseases, similar to what we’ve seen in oncology. We try to catch and ride that wave.”

It might not be too long before the wave crests either.

These and other clinical use cases will help drive methodology development within the ML Research Hub. Subha Madhavan was recently recruited as head of clinical AI/ML and data sciences within early clinical development to help define core requirements of drug programs that will leverage the Hub’s innovative methods to accelerate development.

These efforts will use historical clinical trial data, biomarker data, and real-world evidence such as from electronic medical records to precisely define patient populations to inform study design.

It’s ultimately about improving the likelihood of technical and regulatory success of Pfizer’s clinical trials, Madhavan said.

“Within clinical AI/ML, we are truly driving a paradigm shift in precision medicine. Our focus is on using multimodal data to inform trial design, first-in-human studies [and] our sign of clinical activity studies.”

Pfizer applies advanced methods such as classical and deep learning to molecular data sets compiled from its own clinical trials and published studies “to identify the patient subpopulations that may respond better to a certain treatment,” she explained.

“I am very optimistic about our ability to leverage multimodal, high-dimensional data sets and also rapidly develop algorithms to predict a variety of outcomes for patients.” She predicts that the impact of many of these new innovative tools will reach patients within the next three to five years.

Madhavan was attracted to Pfizer by the company’s “light-speed thinking” and “cross-functional” approach to drug development.

“Pfizer is a company that, despite being one of the ‘big pharmas,’ can turn around very quickly, as demonstrated by the COVID vaccine and anti-viral programs in response to the global pandemic,” she said. “The culture has transformed into one where we can take advantage of it [cross-functional] teams and bring innovation to multiple therapy areas.”

“We take a disciplined product development approach to define business value, key stakeholders, core functionality and usage for each AI/ML model to help align and accelerate our portfolio,” she added.

Applying the latest AI/ML

Pfizer is also applying AI/ML to digital medicine in the Pfizer Innovation Research (PfIRe Lab). Here, researchers are developing algorithms for wearable devices to help scientists monitor symptoms, assess health and better understand how treatments work.

Wearable devices provide researchers and doctors with a “complete and continuous picture of the patient’s experience” during the assessment period, Menon said, rather than relying on the patient’s memory during a single office visit.

Advances abound in AI/ML, but Pfizer is particularly interested in those that can help it reach patients with breakthrough medicines. As Menon said, “we’re not just using AI as a fancy term or shiny object. It’s all about tangible and actionable solutions to key research questions.”

Christensen highlighted explainable AI as an area that can help build science around disease.

“We’re looking for new computer models that are less black box and more open to understanding what’s going on under the hood,” he said.

When it comes to understanding the molecular basis of disease pathophysiology, Huang pointed to a powerful ML architecture called Transformer, which was developed with language models in mind. Transformer is the basis for Google Translate.

“It can help us understand the biomedical literature through natural language processing, which is an area of ​​great interest to us at Pfizer,” he said.

Madhavan said knowledge graphs can help link genes to diseases and drugs and help identify new biomarkers associated with certain disease pathways. Knowledge graphs can also draw connections between patient phenotypes and enable researchers to develop more effective treatments for these patient groups.

Bilingual data scientists wanted

As Pfizer builds out its AI/ML-focused teams, the company is “rapidly recruiting” data scientists.

A dynamic mission requires a dynamic mindset that can sort through complex data to make the correct scientific and therapeutic connections.

“There are a lot of data analysts out there who have incredible skill, but we need to connect that skill with people who understand the science and are willing to take a scientific lens to these hypotheses,” Christensen said. And with a strategic vision that spans so many divisions and specialties, “a communicative and collaborative mindset,” is another key attribute.

Madhavan noted that Pfizer is looking for bilingual data scientists with a deep understanding of both data science and clinical science. The desired candidate will have “work experience where they have actually deployed their quantitative skills to answer key clinical and/or biological questions,” she said.

In the ML Research Hub, Huang is also looking for ambidextrous scientists who not only have ML research and data engineering expertise, but also understand chemistry and molecular/cellular biology.

As every scientist knows, the key to any experiment is reproducibility. “If it doesn’t happen in the real world, I think you’re just going to do more harm than good,” Menon said.

He emphasized the importance of responsible AI. “A lot of times AI is used as a buzzword. It’s basic statistical modeling and mathematical modeling, but if it’s not done by experts who are experts in the science, it’s going to be a weapon that backfires.”

Christensen outlined the opportunity for potential team members: “It’s early days, but this is a great opportunity for data scientists who want to build foundational and enduring systems to help us make data-driven decisions to innovate across the entire drug development paradigm.”

Those interested in joining the Pfizer team can find more information here.