The evolution of AI in drug discovery: learning from history’s mistakes (Part 1)

Sujeegar Jeevanandam, a veteran with 13 years of experience in life sciences R&D solutions, shares key insights into the evolving role of artificial intelligence in drug discovery. Drawing from his extensive work with pharmaceutical and biotechnology companies across North America, Sujeegar offers a unique perspective on how AI adoption might mirror previous technological shifts in the industry.

Historical parallels: the electronic lab notebook revolution

Reflecting on the industry fervour for AI, Sujeegar draws an illuminating parallel between today’s AI revolution and the electronic lab notebook (ELN) transformation of the early 2000s. During the ELN transition, organisations eagerly adopted various platforms, often implementing multiple solutions simultaneously without proper governance. This initial enthusiasm led to challenges: organisations found themselves managing numerous systems, struggling with licensing costs and searching for specialised developers to maintain these platforms.

The comparison is particularly pertinent, revealing a pattern in how the life sciences industry adopts new technologies. An initial period of intense enthusiasm and rapid adoption is later followed by challenges that necessitate a more strategic approach. Sujeegar suggests that AI adoption is following a similar trajectory, with organisations currently in the early, eager phase of adoption.

The future of AI in drug discovery

Looking ahead five to ten years, Sujeegar envisions machine learning models becoming just as integral to research and development as ELNs are today. However, he emphasises that this integration has not happened yet – machine learning models are not yet part of scientists’ routine workflows. Instead, they remain at the experimentation stage, with scientists interested in their potential but not yet fully incorporating them into daily operations.

A significant transformation Sujeegar anticipates is the shift from wet lab to dry lab experiments. He predicts that in the future, scientists will conduct extensive decision-making processes using AI models at their desks before entering the laboratory. However, he clarifies that wet labs won’t disappear entirely. Instead, their role will evolve to focus on validating simulation results rather than being the primary experimental platform. This represents a fundamental shift in the drug discovery paradigm, where in silico experiments could take projects almost to the endpoint, with wet lab validation serving as the final confirmation step.

Challenges and obstacles to AI adoption

Trust and acceptance: Many scientists remain skepticical about AI models, similar to the initial resistance faced by ELNs. Rather than focusing on AI’s potential benefits, they often emphasise its differences from traditional methods.

• Education and mindset: The industry must actively  educate and inspire scientists to recognise the value of AI tools. Concerns about job security and the broader implications of AI adoption must be addressed.
• Black box problem: Unlike the transition from paper to electronic records, AI adoption faces additional challenges due to the ‘black box’ nature of many models. Scientists struggle with the lack of clear causality in AI-driven decisions, making it difficult to trust outcomes, especially when explanations seem artificial.
• Academic-industry gap: Traditional academic training in manual lab environments can create resistance to adopting AI-driven approaches in professional settings.

The discussion will continue in Part 2 of the series, set to be published on 24 March.