Written by:
Senior Director, Head of Biosciences Renal, BioPharmaceuticals R&D, AstraZeneca
Executive Director, Data Sciences and Quantitative Biology, Discovery Sciences, R&D, AstraZeneca
Lead Product Manager, Benevolent AI
Data science and artificial intelligence (AI) have the potential to transform the way we discover and develop medicines. We know that selecting the right target with a strong link to the drivers of disease remains the most important decision we make in the drug discovery process, and data science and AI is having a positive impact here.
Working side-by-side in joint multidisciplinary teams, with leading expertise in drug discovery and disease understanding from AstraZeneca, and with machine learning models and AI from BenevolentAI, we are evolving drug discovery for some of the world’s most complex biological challenges. Together we’ve adopted an innovative, data-driven approach to target discovery, using BenevolentAI’s powerful AI models and tools, to provide AstraZeneca scientists with new insights into disease biology in order to select novel targets that specifically treat the cause of disease.
We have access to more data than ever before but the value of this data can only be realised if we are able to analyse, interpret and apply it. We’ve had success in utilising human OMICs data to identify new targets but we wanted to take it to the next level. Data has historically been analysed in silos and we knew that the ability to better integrate datasets via a knowledge graph had the potential to be game changing.
Our collaboration focuses on two diseases with significant unmet need: chronic kidney disease (CKD) and idiopathic pulmonary fibrosis (IPF). These are also two complex multifactorial diseases where drug discovery is particularly challenging.
We started out by building disease-specific knowledge graphs for IPF and CKD. A knowledge graph is a framework that integrates a large number of patient relevant datasets and complementary types of data modalities in a meaningful way. Machine learning and AI applications can then query this data to uncover previously unknown patterns and make novel target predictions. AI tools then enhance target assessment by surfacing the most relevant data to help scientist make data-driven decisions over which targets to prioritise. The information is stored in the knowledge graph in a scalable manner. Therefore, as our data and knowledge grows and evolves, so will our graphs, which means every newly designed experiment will benefit from everything learned before.
CKD include a number of different aetiologies, driven by multiple pathways and varying underlying causes, unfortunately affecting one in ten people globally. This complexity means it is challenging to understand the underlying drives of disease and identify novel targets. The aim, in the future, is to identify treatments that will halt or even reverse disease progression.
The cause of IPF is idiopathic or unknown. One of the challenges therefore in target discovery is knowing which biological mechanism to focus on. Being able to distinguish which mechanisms are driving a disease from those that are consequences of a disease can help to increase efficacy. Our aim is to treat the causative mechanisms of the disease as opposed to just the downstream symptoms so that we can prevent the progression of the disease.
The collaboration has already yielded a validated AI-generated target for CKD that has entered the AstraZeneca portfolio. For IPF we’ve identified potential targets and are currently experimentally validating them using novel experimental procedures like CRISPR.
The work we’ve developed together has furthered our understanding of these diseases. The use of knowledge graphs, AI tools and machine learning models have put the spotlight on both well-established biology as well as more novel mechanisms. AI is no longer just a promise in drug discovery; we are transforming how new medicines are discovered.