AstraZeneca director says AI should be a “thought associate” in drug discovery

AstraZeneca’s Krishna Bulusu offered on the ELRIG Drug Discovery convention in London on 11 March. Credit score: SOPA Pictures through Getty Pictures.

Synthetic intelligence (AI) is reworking drug discovery, however its implementation should be considerate and strategic, in accordance with Krishna Bulusu, senior director in oncology information science at AstraZeneca.

Talking on the ELRIG Drug Discovery assembly in London on 11 March 2025, Bulusu outlined how AI can enhance effectivity, scale back prices, and assist personalise drugs, however cautioned that its success is determined by information integration, mannequin explainability, and context-specific predictions.

“Accelerating drug discovery doesn’t simply imply doing the identical factor very, very quick. It signifies that we’re additionally going to do various things, and we’re going to do issues extra effectively,” Bulusu mentioned. He burdened that AI shouldn’t be utilized indiscriminately however must be used to reply well-defined scientific questions.

One key instance of how AstraZeneca is utilizing AI function in drug discovery is the corporate’s collaboration with organic simulation firm Turbine. The partnership, introduced in January 2024, makes use of Turbine’s simulated most cancers cell expertise to mannequin drug resistance mechanisms in haematological cancers.

By integrating public and proprietary information, the AI-driven mannequin generates thousands and thousands of simulations, predicting drug interactions and figuring out potential mixture therapies. “Now that’s highly effective, proper?” Bulusu requested. “As a result of the scalability side – if I’ve to go to the lab and do that, it’ll value me some huge cash.”

In contrast to conventional lab-based analysis, which could be time-consuming and costly, AI simulations permit researchers to discover advanced organic pathways at a fraction of the time. The mannequin supplies quantitative insights into how completely different inhibitors impression mobile pathways, providing a brand new technique to generate hypotheses and refine drug improvement methods.

“From an enormous pharma perspective, that is nice as a result of I’ve a portfolio of medicine, and also you’re telling me the place to place my portfolio medication, both as regular remedy or in combos. That’s why we work very intently with Turbine on this,” Bulusu added.

The concept of AI-driven organic simulation is gaining traction throughout the business. Demis Hassabis, CEO of DeepMind and co-creator of AlphaFold, has beforehand shared his imaginative and prescient for AI-powered digital cells: “My dream is to ultimately have digital cells, like a simulation of a digital cell. We’re perhaps ten years away from that,” Hassabis mentioned on the Monetary Occasions Pharma and Biotech Summit in November 2024.

Nevertheless, Bulusu underscored that AI’s success in drug discovery is not only about technological development but in addition about guaranteeing that fashions are interpretable, and predictions are related. “The worth for AI, at the very least with the bigger organisations prior to now, is when a non-data scientist understands and appreciates them. And this doesn’t occur until you’re working collectively.”

Investigating the scope of AI in drug discovery

Regardless of AI’s potential, vital challenges stay. Bulusu pointed to gaps in longitudinal affected person information, biases in AI fashions, and the necessity for higher early illness detection.

“AI must develop into a thought associate. And for that to occur, the arrogance and belief in what we’re doing as information scientists wants to come back by,” he mentioned.

To handle these challenges, Bulusu emphasised the necessity for improved information assortment and integration, notably capturing the full affected person journey from preclinical to medical phases. “We’re very dangerous as a group at producing longitudinal information,” he mentioned. Longitudinal information must be colletcted earlier than preclinical and medical analysis, however that is at present occurring in reverse. That is essential, as a result of Bulusu mentioned whereas “a affected person’s journey is essential to seize, we’re simply taking snapshots of it.”

Trying forward, Bulusu emphasised the significance of beginning with the fitting scientific query earlier than deciding on an AI mannequin.

“From an AI perspective, worth and impression is and can at all times be pushed by beginning with the fitting query after which asking, what’s the fitting mannequin to reply that query. It’s by no means the opposite method round,” Bulusu concluded.