The Future of Drug Targeting and Clinical Development with Generative AI Tools – with Ramesh Durvasula of Eli Lilly

Riya Pahuja

Riya covers B2B applications of machine learning for Emerj - across North America and the EU. She has previously worked with the Times of India Group, and as a journalist covering data analytics and AI. She resides in Toronto.

Future of Drug Targeting – Lilly @2x-1-min

Lily is a pharmaceutical giant with a legacy dating back to its founding in 1876 by Colonel Eli Lilly. The company engages more than 9600 employees in research and development, with clinical research conducted in more than 55 countries. As of 2022, the company clocked a revenue of $28,541.4 million and made a net income of $6,244.8 million. 

In recent years, Lilly has made substantial investments in AI-driven technologies, leveraging the capabilities of machine learning and data analytics to enhance drug discovery, optimize clinical trials and improve patient care. These initiatives have streamlined research and development processes.

Emerj Senior Editor Matthew DeMello recently spoke with Ramesh Durvasula, SVP R&D IT at Lily, on the ‘AI in Business’ podcast to discuss the role of AI in drug discovery, the importance of data volume for training AI models for healthcare and the challenges in addressing rare diseases.

In the following analysis of their conversation, we examine two key insights:

  •  Addressing the scarcity of data for rare diseases: Collecting comprehensive data like digital biomarkers, demographic information and more to help AI algorithms discern patterns in patient conditions faster. 
  • Leveraging ‘digital twins’ in training digital co-pilots to encourage cross-enterprise collaboration: Using data-based replications of health systems for training conversational AI assistants in both public and proprietary data to provide tailored and informed assistance to individual employees.

Listen to the full episode below:

Guest:  Ramesh Durvasula, SVP R&D IT, Eli Lily

Expertise: Drug discovery, Informatics, Medicinal Chemistry

Brief Recognition: Ramesh Durvasula is responsible for the IT and informatics capabilities that accelerate the discovery and preclinical efforts of Lilly Research Labs. Before his time at Lilly, Ramesh was Executive Director of Discovery IT and Lab Automation at Bristol-Myers Squibb; over his 12 years at BMS, Ramesh focused on integrating physical lab and digital analytics systems into seamless capabilities. Ramesh earned his BA in Chemistry and Ph.D. in Biochemistry from the University of Virginia.

A Potential Solution to Address the Scarcity of Data for Rare Diseases

Ramesh acknowledges that AI holds substantial potential across the entire research and development process, extending further into manufacturing and commercial applications. 

While specifically focusing on R&D, he emphasizes the remarkable opportunities AI presents, particularly in drug discovery and clinical development. Within drug discovery, he points out that researchers face two essential questions for any disease:

  • Whether they understand the molecular basis of the disease
  • Once that understanding is clear, the team can identify therapeutic agents capable of addressing the disease’s molecular characteristics. 

Ramesh stresses that this approach applies to various diseases, such as Alzheimer’s, psoriasis and different types of cancer. He mentions that Eli Lily’s mission involves identifying the correct biological targets related to the molecular basis of diseases and then finding suitable therapeutic agents, which could encompass various types of molecules like small molecules, antibodies, RNAs and gene editing tools. 

He further discusses the application of AI in the context of rare disease research, highlighting two key challenges and potential opportunities. 

1. Challenges with Rare Disease Data: Ramesh notes that AI models benefit from extensive and high-quality datasets, which enable the development of more accurate and predictive algorithms. However, he points out that rare diseases present a unique challenge because the available data, including information on patients and disease agents, is minimal. 

In this context, he cautions against over-reliance on AI without considering the available data’s size, volume, variety and quality. The analogy here likens this situation to using advanced mapping algorithms on an ancient map, highlighting the importance of matching advanced algorithms with suitable data sources.

2. Opportunities in Collecting Comprehensive Data: Ramesh then transitions to discussing opportunities in rare disease research related to data collection. He suggests that one potential solution to address the scarcity of data for rare diseases is to collect a wider variety of data about the affected patients. This expanded dataset could encompass digital biomarkers, demographic information, geographic data, family histories and more. 

He believes that by broadening the types of data collected, including population-related information that is often overlooked, AI algorithms could excel in extracting meaningful signals from what might initially appear as noise.

 Ramesh also underscores that AI algorithms are adept at discerning patterns within extensive and diverse datasets, making this a promising avenue for enhancing rare disease research.

He continues to discuss the continuous quest for improved methods to analyze data and methodologies for extracting insights from available data. He emphasizes the organization’s primary focus on developing medicines that enhance people’s lives. He is committed to exploring various approaches, whether AI, less advanced, or more cutting-edge technologies. The goal is to identify the most innovative and effective ways to advance their research.

The speaker also touches upon translational medicine, highlighting his interest in understanding how clinical trial data they collect can be used to support more in-depth and focused early research efforts. He desires to bridge the gap between clinical trial data and early-stage research activities, such as experiments conducted in a petri dish. This approach implies a strategic method to utilize data from clinical trials to inform and improve how they model and conduct future clinical trials.

Leveraging Digital Twins and Training Co-Pilots to Encourage Cross-Enterprise Collaboration 

Ramesh then transitions to highlighting his organization’s keen interest in harnessing AI capabilities to boost the productivity and value of their employees, ranging from scientists and researchers to clinicians and individuals in sales and marketing. They envision a future where “digital co-pilots” play a central role in assisting every position within the company.

These digital co-pilots are described as advanced AI-driven assistants capable of handling various tasks, including answering questions, monitoring experiments and aggregating and summarizing data. 

Introducing these digital co-pilots aims to empower their workforce, especially scientists and employees, by relieving them of routine or basic tasks. This introduction, in turn, allows these professionals to dedicate more time and energy to more intricate and strategic aspects of their work.

By leveraging LLMs and other bespoke models, co-pilots use historical data to assist human agents in solving their problems. When Microsoft launched Microsoft 365 Co-pilot, it defined co-pilot as something that combines the power of LLMs with individual data in the Microsoft Graph and the Microsoft 365 apps to help people in their work.

Ramesh’s vision extends beyond the essential functions of digital co-pilots. He expresses excitement about the second aspect of his vision, which involves these co-pilots continuously learning and improving over time. What is unique about the process is that these AI assistants would not rely solely on publicly available data. Instead, they would draw knowledge and insights from the organization’s historical data (Eli LillyLily), often referred to as proprietary data. 

The goal is for these co-pilots to become progressively more intelligent and more insightful by analyzing the decisions made by Lilly’s staff. This way, each co-pilot becomes a repository of organizational wisdom and expertise, enabling it to offer tailored and informed assistance to individual employees. 

Empowering workflows across the enterprise in such a way emphasizes the power of AI in leveraging internal knowledge and experience to augment decision-making and problem-solving within the company.

Lastly, Ramesh discusses a future vision where numerous opportunities are anticipated in digital twin technology. He mentions that the term “digital twin” is particularly relevant in this context, and various types of digital twins will be deployed. It’s not limited to a single kind of digital twin, such as one that allows a complete simulation. 

NASA defines a digital twin as a virtual representation of an entity that serves as a real-time digital counterpart of a physical object. Theoretically, a change made to the physical object automatically leads to a change in the digital object and vice versa. 

To clarify his point here, Ramesh uses the example of a digital twin for an airplane engine, where every engine component is replicated, and a complete simulation of the engine’s behavior can be run.

In Lily’s case, he acknowledges that it may not necessarily entail a complete simulation of a patient. Instead, their approach will involve creating models that act as proxies for diseases or mimic the decision-making processes of professionals like physicians, chemists, or biologists. These digital twin models will be integrated into the co-pilot systems intended to assist researchers within Eli Lilly.

Stay Ahead of the AI Curve

Discover the critical AI trends and applications that separate winners from losers in the future of business.

Sign up for the 'AI Advantage' newsletter: