Will Machine Learning be Pfizer’s Answer to Curing Cancer?

Pharmaceutical giant Pfizer has partnered with IBM's Watson Drug Discovery program to accelerate their immuno-oncology drug discovery time. Perhaps IBM Watson's machine learning capabilities will uncover novel therapy combinations for Pfizer that could lead to a cancer cure.

           In 2016 US pharmaceutical giant Pfizer announced a partnership with IBM Watson to accelerate its drug discovery process for the immuno-oncology drug class. In forming this partnership Pfizer became one of the first organizations to deploy IBM’s Watson Drug Discovery program, which Pfizer customized to allow it to support the identification of new drug targets, combination therapies for research, and patient selection strategies in immuno-oncology [1].

            Recently immuno-oncology has become an increasingly popular field for the major biopharmaceutical companies given its potential promise in successfully treating cancer. The immuno-oncology market size is expected to increase to $19B in 2019 and to $34B by 2024 [2].  Immunotherapies offer a novel way of treating cancer by modifying a patient’s immune system to recognize and treat cancer cells. The therapies achieve their intended effect using a combination of vaccines and other types of molecules to attack a patient’s tumor and illicit the intended immune response. Cancer is one of the leading causes of death worldwide, and many in the medical field argue that it is one of the most complex diseases known. However, researchers believe that immuno-oncology will define the future of cancer treatment given the ability of immunotherapies to combine combination of therapies tailored to unique tumor characteristics [1].

            To understand the benefits that Watson Drug Discovery can have on Pfizer’s drug discovery process, it is important to understand how the US FDA defines the process for developing a new drug. The five steps of development are discovery and development, preclinical research, clinical research, FDA review, and FDA post-market safety monitoring. The Tufts Center for the Study of Drug Development estimates that the average cost of developing a new drug is approximately $2.55 billion at an average total process time of 10 years. The first phase, drug discovery, is quite challenging because at this stage thousands of compounds may be potential candidates for development as a medication, but only a small number will advance to further research [3].

            Drug discovery is an incredibly data-driven process with information published in millions of scientific papers among thousands of journals, thus it can be challenging for researchers to access all the data needed and make connections among the various sources to identify key insights that will impact their drug discovery work. By deploying machine learning to this vast amount of structured and unstructured data, including medical literature, images, and genetic and clinical trial information, Pfizer researchers can generate patterns from sparse data, predict relationships, and form hypothesis on which drugs may be promising to move into the development phase in far less time than before [4]. Ultimately Pfizer hopes that this process will help its researchers uncover potential new therapies in the immuno-oncology space.

            Pfizer entered the immunotherapy space back in 2014 when it announced a joint venture with Merck KGaA to continue to advance one of Pfizer’s immunotherapy products into Phase I trials for further indications. Pfizer executives were satisfied with this agreement because it allowed them to accelerate the timeframe of their development programs and move into the first wave of their own immuno-oncology based treatment regimens [5]. While continuing to form partnerships with biotech and biopharmaceutical companies to bring immuno-oncology drugs to market, Pfizer is successfully using machine learning to identify a diverse array of compounds that will address difficult to treat cancers. They currently have eight immuno-oncology compounds in clinical trials [6]. Given it has only been 2 years since Pfizer announced the Watson partnership, it is not surprising that they do not have more immuno-oncology drugs in the pipeline.

            Beyond using machine learning for the drug discovery phase, my recommendation is that Pfizer should also use machine learning to repurpose or expand the indication of existing drugs. Machine learning could be used to mine databases to explore the relationship between drug molecules and diseases or genes. This means that Pfizer would be able to sell a drug that is already on the market for a different disease or type of cancer. Additionally, given how competitive the immuno-oncology therapeutic class is, Pfizer should extend their use of machine learning to discover new drugs in areas such as rare diseases where pressure to recoup research and development spend is extremely high. The machine learning process would identify top drug candidates through lower amount of research effort and spend.

Open Questions:

  1. Given Pfizer’s recent past of acquiring assets already in clinical trials or partnering with biotechnology companies that have a product in clinical trials and using its brand to market and sell the products around the world, does it make sense for Pfizer to invest in this capability to reduce the drug discovery time?
  2. Would relying solely on machine learning for drug discovery cause Pfizer to fail to identify an important combination of therapeutics that may be the cure to certain types of cancer?


(Word count: 800)



[1] “IBM, Pfizer collaborate to utilize Watson for Drug Discovery to accelerate immuno-oncology research.” Pharmabiz.com, December 2016. http://www.pharmabiz.com/NewsDetails.aspx?aid=99032&sid=2. Accessed November 2018.

[2] A. Philippidis. “Top 15 Immuno-oncology Collaborations”. Genetic Engineering and Biotechnology News, September 2016. https://www.genengnews.com/topics/drug-discovery/top-15-immuno-oncology-collaborations/. Accessed November 2018.

[3] “How AI can speed up drug discovery.” Medium, April 2016. https://medium.com/syncedreview/how-ai-can-speed-up-drug-discovery-3c7f01654625. Accessed November 2018.

[4] “Artificial Intelligence: The Next Frontier in Drug Discovery.” Fortune Knowledge Group, 2018. https://public.dhe.ibm.com/common/ssi/ecm/17/en/17016517usen/watson-health-healthcare-and-life-sciences-white-paper-external-17016517usen-20180524.pdf. Accessed November 2018.

[5] “Merck KGa, Pfizer Launch Up-to-$2.85B Cancer Immunotherapy Alliance.” Genetic Engineering and Biotechnology News, November 2014. https://www.genengnews.com/topics/drug-discovery/merck-kgaa-pfizer-launch-up-to-2-85b-cancer-immunotherapy-alliance/81250602/. Accessed November 2018.

[6] “Pfizer 2017 Annual Review: The Power of Science.” 2018, https://www.pfizer.com/files/investors/financial_reports/annual_reports/2017/assets/pdf/pfizer-2017-annual-review.pdf. Accessed November 2018.





An Algorithm Could Be Your Next Favorite Artist


Using machine learning to bend the healthcare cost curve

Student comments on Will Machine Learning be Pfizer’s Answer to Curing Cancer?

  1. I think that machine learning can be very helpful for the discovery process of a drug, especially considering the length of the process and the amount of data involved. Pharma companies have realized that and they are increasingly trying to enter into partnerships to get access to these tools. Pfizer in particular announced another partnership in March of this year with XtalPi to work on drug discovery through quantum mechanics and machine learning. Please see the press release below for reference:


  2. “Would relying solely on machine learning for drug discovery cause Pfizer to fail to identify an important combination of therapeutics that may be the cure to certain types of cancer?”

    I think machine learning is a powerful tool in the hands of researchers in the field of drug discovery, however I do not believe that it will ever be able to be used alone. Data strewn across “millions of scientific papers among thousands of journals” is literally impossible for a team of humans at Pfizer to analyze effectively. I agree that the power of machine learning here is not just the ability to read all of these papers, but to generate patterns and predict relationships and hypotheses that must be evaluated by experienced human scientists. It would not at all surprise me if, as you said, machine learning aims in discovering meaningful relationships that existing drug molecules have with other diseases and genes, relationships that are currently invisible to us.

Leave a comment