Pfizer is using AI to help revolutionize the development of life-saving medications

The past hundred years have brought astonishing medical advancements. The ability to edit genes, build life-saving machines and create vaccines that help prevent or stop diseases in their tracks were previously unfathomable, but are realities of today. And excitingly, both scientific methods and the pace at which life-saving medical breakthroughs are developed continue to advance, partially thanks to the advent and use of artificial intelligence (AI) and machine learning (ML).

In general, the life sciences industry is shifting towards a higher adoption of digital tools across the entire continuum. While such tools may not necessarily be new, the scale at which they are being applied within healthcare has changed immensely, particularly since the onset of the COVID-19 pandemic. For instance, we saw increased adoption of telehealth visits and movement away from the traditional office visit, where appropriate.[i] Some estimates suggest that the pandemic accelerated these trends for patients.[i]

“AI has the potential to surpass all other strategies for early predictions. For example, advanced machine learning used at the earliest stages of small molecule research has the potential to further streamline the design and discovery of our investigational medicines," said Djork-Arné Clevert, Vice President Machine Learning, Pfizer Inc. “The technology has the ability to computationally screen large numbers of potential molecules and narrow in on those that should be further explored by chemists, without having to synthesize and test every such molecule,” he added.

And following the quicker identification of potentially successful molecules, ML is uniquely able to help predict how the molecules will interact with the target, predict where the medication will go in the body and predict how it may effectively treat the broader context of the disease. Because AI/ML algorithms have the ability to gather and analyze massive datasets, expanded use is expected to also help us learn about the still unknown components of biology that have yet to be understood or uncovered.

While the focus of such deep learning has typically been limited to areas that have large datasets available, few studies have explored application of this technique to scientific problems of practical interest that lack sufficiently large datasets, with difficulty in acquiring data in a timely fashion. One such area that has not yet been fully aided by AI/ML algorithmic ability is monoclonal antibodies due to time, material, and other resource constraints.

Recently, the Pfizer biomedicine AI team took on this challenge, addressing the constraints of small data in developing predictive models for antibody viscosity, a key developability attribute for monoclonal antibody-based therapeutics. In their work, Pfizer scientists demonstrated that deep learning-based models can generalize with high accuracy, even when trained on as few as a couple dozen datapoints.[i]

While these technologies do not replace the human element, AI and ML, paired with human insight and interpretation, will likely help ensure more accurate outcomes by elucidating targets for new medicines, better understanding how a disease develops in the body over time and helping ensure that medications are developed to intervene at the optimal time and in ways that are potentially meaningful to those living with a specific condition, as demonstrated by Pfizer’s ground-breaking work with antibody viscosity.

The fact that Pfizer has a decade-long history with practical use of these technologies provides a critical head start. The company’s access to expansive datasets that can be combined with available data around biology and the development of potentially better molecules is a significant advantage. And the early results speak for themselves — Pfizer used its super-computing tech to help fast-track the development of both vaccines and an authorized oral treatment for COVID-19.

Rod MacKenzie et al. COVID-19 must catalyse changes to clinical development. nature reviews drug discovery. Retrieved January 23, 2023
Julia Shaver, MD. The State of Telehealth Before and After the COVID-19 Pandemic US National Library of Medicine National Institutes of Health. Retrieved January 24, 2023
Brajesh K Rai, et al. Low-Data Interpretable Deep Learning Prediction of Antibody Viscosity using a Biophysically Meaningful Representation. Scientific Reports. Retrieved January 26, 2023