AstraZeneca's AI Playbook: Jorge Reis-Filho With Michelle Lee

Author: Jorge Reis-Filho, Chief of AI for Science Innovation, AstraZeneca; Michelle Lee (moderator), Founder & CEO, Medra Collection: UNLOCK-2026-Sessions

Summary

Jorge Reis-Filho, Chief of AI for Science Innovation at AstraZeneca, describes the company's comprehensive strategy for AI deployment across drug discovery and clinical development in a fireside chat with Medra CEO Michelle Lee at UNLOCK 2026. Reis-Filho comes to this role from pathology — he retrained as a bioinformatician at the turn of the century, then fell into convolutional neural networks applied to pathology images in 2015, contributing to foundation models that now perform over 30 pathology tasks previously requiring specialist pathologists, at accuracy and reproducibility levels that exceed human performance. This direct experience of watching AI disrupt his own specialty shapes his conviction that pharma must proactively transform itself.

AstraZeneca's core thesis is that AI will change every step of drug discovery and clinical development, but the time horizons differ substantially by area. Reis-Filho's team is focused primarily on transforming clinical development — leveraging the company's 15 years of accrued translational data across what he claims is the most diverse oncology portfolio in the industry (8 modality categories: 5 targeting cancer cells directly — radiojugates, antibody drug conjugates (ADCs), DNA repair inhibitors, oncogenic driver inhibitors including EGFR and ER signaling, and epigenomics agents — plus 3 immune-system-targeting categories including checkpoint inhibitors, T-cell engagers, and cell therapy for solid malignancies). The strategy is to insource domain-specific frontier foundation models and fine-tune them on AstraZeneca's proprietary clinical trial data to address four problems: multimodal patient selection, phase 1/2 acceleration and derisking, increasing phase 3 probability of technical success (PTS), and combination therapy design.

The regulatory dimension is a major focus. In 2023, AstraZeneca submitted an AI-derived biomarker to the FDA that received breakthrough device designation. Since then the FDA has moved from being a bottleneck to an active partner, inviting AstraZeneca to seminars and co-iterating on standards. A key regulatory development: the European Society of Medical Oncology (ESMO) and FDA now treat mechanistic explainability as "desirable" rather than "mandatory" for AI-derived biomarkers, provided the model is validated in sufficiently large independent cohorts for its intended predictive task.