Natera, Inc. announced the launch of its proprietary AI foundation model platform. These models and applications, which were developed in-house, are designed to drive innovation across therapeutic development, from early target discovery to real-time clinical decision support. The platform features a modular, multimodal architecture composed of three integrated layers: Data foundation layer: At its core is one of the largest longitudinal, multimodal oncology datasets ever compiled, purpose-built for AI training.

This includes de-identified data from more than 250,000 tumor exomes and over 1 million longitudinal plasma timepoints, enriched with clinical records, treatment histories, digital imaging, expression profiling and outcomes data. This high-fidelity dataset enables scalable, AI-ready model development. Core model layer: Leveraging over 1 billion parameters, the core AI foundation model is trained on de-identified Signatera??

and Altera?? datasets and integrates genomic, clinical, and imaging modalities. It is designed to power discovery in multiple domains, including biomarker development, patient stratification and therapeutic response prediction.

Application layer: Several AI-driven applications support clinical decision-making and drug development: Digital Patient Simulator: Virtually simulates patients for treatment optimization, such as suggesting next-line therapies and de-escalation opportunities. It can also predict patient outcomes. Real-Time Trial Matching: Uses molecular and clinical data to identify trial-eligible patients and accelerate recruitment.

This engine can also simulate virtual trials to optimize study design and reduce development risk. NeoPredict: An advanced algorithm that predicts individual immunotherapy responses based on tumor genomics and neoantigen presentation. Two recent pilot programs of the digital patient simulator demonstrated excellent performance, supporting its potential to enhance clinical decision making in oncology.

Results from the first pilot showed that Natera's algorithm can accurately recommend immunotherapy based on real-world electronic health record data. The second pilot showed outperformance of both tumor mutational burden and standalone pathology-based metrics in predicting immunotherapy response.