Anna Lagos: How effective have these early predictions been?
Slave Petrovsky: For about 1,000 diseases studied, we achieved a predictive performance of about 70%, meaning that in 70% of cases, when the AI predicts that an individual will not develop a disease, it is correct. In around 121 diseases, the accuracy reaches 90%, allowing for much earlier interventions.
Anna Lagos: How does artificial intelligence help you identify new therapeutic targets from genomic data? And what challenges do you face when integrating large volumes of genomic data with AI?
Guillermo del Angel: A major challenge is ensuring genetic diversity in our data. Milton’s model was initially trained on 500,000 UK Biobank participants, which provides incredible insights, but we aim to make it more applicable to global populations. That’s why AstraZeneca is seeking collaborations with local researchers around the world. Another challenge is that although Milton is a powerful research tool, it is not yet ready for clinical practice.
Slave Petrovsky: To add, the exciting part about using AI in genomics today is that we finally have the right data to feed these advanced models. The discovery of therapeutic targets was traditionally based on preclinical models, such as cell culture or animal studies. But having access to large-scale human data allows us to develop treatments directly based on human biology, significantly improving applicability.
Guillermo del Angel: Additionally, in the case of Alzheimer’s, for example, Milton has identified specific circulating biomarkers (Neurofilament L, ApoE, and GFAP) that, although not measured in routine checkups, are strong predictors of Alzheimer’s years before symptoms appear. This opens the possibility of early interventions.
Anna Lagos: What are the main ethical considerations when using AI for disease prediction, and how does AstraZeneca address these concerns?
Slave Petrovsky: We take privacy and diversity very seriously. Participants in the UK Biobank, for example, are anonymous volunteers who altruistically contribute their data to medical research, often without direct personal benefit. Our responsibility is to protect your data and make sure we are using it ethically.
Guillermo del Angel: Yes, and globally, we publish all our data and methods transparently, so that others can build on them and adapt them locally. We believe this will support preventive health systems around the world, where early diagnosis can have an even greater impact due to limited resources.
Slave Petrovsky: Furthermore, 95% of the world’s genomic data comes from individuals of recent European ancestry, which does not reflect global diversity. That’s why, at AstraZeneca, we are intentional about collaborating with other less-studied regions to improve equity in genomic medicine.
Anna Lagos: Looking ahead, how do you think tools like Milton will transform the healthcare landscape over the next five years?
Slave Petrovsky: These AI tools will only get better as we add more data, such as image data and wearable devices. This will improve predictions for diseases where Milton currently has limitations. Ideally, we will reach a point where we consistently achieve 90% accuracy for a wide range of diseases, making predictions very actionable. Over the next five years, as more health systems recognize the value of early detection, we expect to see a more prominent role for AI tools like Milton in global health.
#MILTON #AstraZenecas #predicts #diseases #symptom
