AstraZeneca and BenevolentAI raise hopes of new IPF therapy

AstraZeneca and BenevolentAI have announced a new novel target for idiopathic pulmonary fibrosis (IPF) from their collaboration as the potential of their combined firepower becomes apparent.

The strategic collaboration began in 2019, with the first success – a novel target for chronic kidney disease – selected for AstraZeneca’s drug development portfolio entry in January. The latest success follows hot on the heels of a win for the two companies in the ‘Best Partnership Alliance’ category at the prestigious Scrip Awards earlier this month.

Founded in 2013, BenevolentAI is a leading clinical-stage AI drug discovery company based at Babraham Hall’s Minerva Building, with a mission to “accelerate the journey from data to medicines”. AstraZeneca’s global R&D facility is The Discovery Centre at Cambridge Biomedical Campus – though the team using the Benevolent platform is located on various sites, says Daniel Muthas, senior director in data science and bioinformatics at AstraZeneca.

Speaking with BenevolentAI CSO Anne Phelan, Daniel – who is based in Gothenburg – said: “Future milestones are very difficult to speculate about but this is a very exciting target and that could have a significant effect for patient care – we are very delighted at the discovery.”

IPF is a chronic and ultimately fatal disease that causes lung tissue to stiffen, leading to permanent lung scarring that makes it harder to breathe. As IPF progresses, patients often need oxygen and, in some cases, lung transplantation. With a median survival of around three years, the prognosis can be devastating, and there is a clear unmet need for better treatments.

Dr Phelan said: “The cause of IPF is largely unknown and the exact mechanisms involved in the progression of IPF remain elusive. Our collaboration uses advanced AI to enable expert scientists to navigate this challenging disease landscape, and discover novel targets with the potential to treat the underlying causes and prevent disease progression. This second important milestone in our collaboration with AstraZeneca is further evidence of how our platform can deliver tangible scientific results in the most complex therapeutic areas.”

The collaboration has created a co-working partnership that is now working on some very intricate biological challenges. BenevolentAI brings advanced AI models and expertise to the project, and AstraZeneca has deep scientific and clinical expertise, and extensive datasets available to be interrogated. By joining forces, the two companies can build – and interpret – huge new datasets that capture the whole picture of disease biology and allow scientists to more accurately find the right mechanisms to target in the relevant patient population.

The process starts by using BenevolentAI’s proprietary natural language processing (NLP) as well as other machine learning/AI techniques to integrate disease-specific data – including omics and AstraZeneca’s proprietary data – from a mixture of unstructured and structured data, into the BenevolentAI Knowledge Graph. A knowledge graph represents a network of contextualised scientific data such as genes, proteins, diseases and compounds, and the relationship between them.

Whereas scientists usually analyse data in silos, this holistic approach reasons through many data modalities, opening up a vast and diverse data landscape for exploration and experimentation without boundaries or bias. Ultimately, this gives “a full and unbiased overview of disparate datasets in order to draw connections that may not be obvious from any single source”.

BenevolentAI’s AI and machine learning models sit on top of the knowledge graph to find previously unexplored patterns and uncover a new understanding of disease biology. The partnership leverages this data-driven approach to empower AstraZeneca scientists to identify novel drug targets that siloed datasets and experts may not have otherwise found.

The way the teams interact is fascinating.

“We identified a target and the platform looks at it in the context of human biology, either as a cell type or a disease, plus the part it plays in signalling disease within the cells,” says Dr Phelan.

“So we run through the hypothesis generation workflows and we map that on to the mechanism. We identify a molecule that can either activate or inhibit the target, and rebalance the signalling casket to prevent disease progression or to restore health. The Knowledge Graph of a disease is mechanism-agnostic, it’s a really good graph of human biology plus patient-level data, and the structures of biology. There are hundreds of mechanisms and we have the capacity to look at any or all of them: we can prioritise any disease. It’s not specifically designed for IPF.

“We have spent years building our proprietary Knowledge Graph which contains all available scientific research, biomedical datasets and in-house experimental data, and for this IPF project, we enriched it with IPF specific data including omics and AstraZeneca’s proprietary data.”

Daniel adds that the ensuing “vast platform” is “very broad and unbiased omically”.

“Our experience of the graph is that we can interrogate it really well,” he says. “The algorithms provide us with target ideas, and then AstraZeneca and Benevolent sit down together to go through them and develop further experiments to validate them. It’s a very significant process.”

On how often these sit-downs take place, Daniel remarks: “It’s much more ongoing.”

Dr Phelan adds: “It’s a constant dialogue and exchange; a continual flow of information and thoughts.”

This network of exchanges is taking place across many sites, says Daniel.

“I have a global team, with team members working in Cambridge, including biologists and others, in other parts of the world. So yes, I’m very excited by it.”

Professor Maria Belvisi, SVP and head of research and early development, respiratory and immunology at AstraZeneca, commented: “At AstraZeneca we aim to target the underlying disease drivers in idiopathic pulmonary fibrosis to stop fibrosis in its tracks, promote tissue regeneration and enable people with IPF to live life without limits.

“Our ongoing collaboration with BenevolentAI has enabled us to leverage the world’s available scientific literature and our in-house experiments, all brought together through machine learning to identify previously unrecognised links. I’m proud that this collaboration has delivered the first AI-driven IPF target to AstraZeneca’s portfolio.”

- Click here to receive your free weekly digest of Cambridge Independent stories.