Pfizer Accelerates Drug Discovery with AI: The Future of R&D

Which Company's Case Study?

This case study focuses on Pfizer, one of the world's leading pharmaceutical and biotechnology companies. The company is globally known for developing vaccines and treatments for COVID-19, but behind the scenes, it is also aggressively promoting digital transformation leveraging AI. In particular, Pfizer aims for AI-driven innovation in the drug discovery process, which typically requires over a decade and enormous costs to bring a single new drug to market.

Challenges They Aimed to Solve

Traditional drug discovery processes faced several significant challenges. Firstly, the process of discovering new drug candidates and verifying their efficacy and safety was extremely long and inefficient. Secondly, the sheer volume of information researchers needed to analyze – including scientific papers, patents, and clinical trial data from around the world – was exploding, pushing manual information gathering and insight extraction to its limits. These challenges led to prolonged development times and a low success rate (the probability of a drug progressing to clinical trials and being approved is only about 12%).

How AI Was Used

Pfizer has implemented AI and Machine Learning (ML) at every stage of drug discovery, notably through collaborative initiatives like 'PACT (Pfizer-Amazon Collaboration Team)' with Amazon Web Services (AWS). Specifically, generative AI is used to design new molecular structures, and Natural Language Processing (NLP) is employed to analyze vast medical literature, allowing researchers to quickly find the information they need. For instance, prototype development leveraging AWS services has resulted in reducing scientists' search time by up to 16,000 hours annually. Furthermore, Pfizer uses machine learning models to predict compound properties and efficiently narrow down promising new drug candidates. These efforts contributed to accelerating actual drug discovery processes, such as the development of the COVID-19 treatment 'Paxlovid,' at a record pace.

Estimated Architecture

The following is an estimated architecture based on publicly available information and common configurations. It does not definitively represent the actual internal structure.

• ▸Input Data: Worldwide scientific paper databases (e.g., PubMed), patent information, clinical trial data, genomic data, internal R&D data, and experimental data.
• ▸AI Processing/Model or Search Layer: Leverages AI/ML platforms such as AWS's Amazon SageMaker and Amazon Bedrock. Natural language processing models for analyzing papers and research data, generative AI models for creating new compounds, and machine learning models for predicting compound efficacy are operational. Retrieval-Augmented Generation (RAG) technology achieves highly accurate information retrieval and summarization by combining internal data with the latest external literature.
• ▸Output to Business Systems: Analytical results and insights are provided through an integrated data search platform for researchers (Scientific Data Cloud). Lists of promising compound candidates and simulation results are visualized on dashboards to support researchers' hypothesis testing and decision-making.
• ▸Monitoring/Governance: Continuous monitoring of AI model accuracy and fairness. Given the sensitive data involved in drug development, AWS security services are utilized to ensure strict data governance and compliance.

Implementation Effects and Key Takeaways

• ▸Accelerated R&D: Significant reduction in drug discovery process duration through AI-driven information search efficiency and early identification of promising compound candidates. There are reports of reducing scientists' search time by up to 16,000 hours annually.
• ▸Cost Reduction and Increased Success Rate: AI screening of compounds with high failure risk in early development stages reduces wasteful R&D investment. Furthermore, cloud utilization is reported to have cut infrastructure costs by 55%.
• ▸Cultivation of a Data-Driven Research Culture: By building platforms like 'Scientific Data Cloud,' Pfizer has created an environment where researchers across the company can easily access vast amounts of data and benefit from AI. This promotes data-driven decision-making.
• ▸A key point for readers to note is that Pfizer is not merely adopting specific AI tools but rather building an end-to-end ecosystem, leveraging a cloud platform as its foundation – through collaborations like with AWS – to handle everything from data aggregation to AI model development and delivery to researchers.

What Japanese Companies Can Learn

Pfizer's case study offers valuable insights not only for the pharmaceutical industry but for many Japanese companies that handle vast amounts of specialized knowledge and R&D data. The mechanism of organizing and leveraging 'intellectual assets' such as scattered papers, reports, and experimental data within a company using AI can be applied across various fields, including material development in manufacturing, market analysis in financial institutions, and contract review in legal departments. Beyond solving specific problems, the perspective of developing a company-wide data infrastructure and creating an environment where everyone can utilize AI will be a crucial point for promoting DX.

Today's AI News

Reference Links

Find Related AI Adoption Case Studies