Verily’s multimodal AI model marks a new era in precision medicine with integrated health and genomic data

Verily has developed the first multimodal foundation model combining electronic health records with genomic data, enhancing disease risk prediction, with support from NVIDIA and the US NIH’s All of Us Research Program, signalling a potential leap forward in personalised healthcare.

Verily Health says it has built the first multimodal foundation model to combine electronic health records with genomic data, in a collaboration with NVIDIA and the US National Institutes of Health’s All of Us Research Program. The effort is an early sign of how artificial intelligence could move beyond single data sources to create a fuller picture of disease risk and patient health.

The company’s argument is that health records, while valuable, only capture part of a person’s story. By adding genomic information, the model can start to distinguish inherited risk from changes that emerge later in life, which Verily says is especially useful for conditions such as type 2 diabetes. The All of Us programme, which is designed to support precision medicine research, provides the kind of large, linked dataset needed for this work.

According to Verily, the combined model improved its ability to identify people at high risk of type 2 diabetes, while also reducing false alerts. The company also says the system could estimate risk over a five- to 10-year horizon, rather than focusing only on short-term prediction. On the computing side, Verily says it used NVIDIA’s hardware and software tools to speed up pre-training, cutting the time needed to iterate on the model.

The broader significance, Verily says, is that the same approach could eventually pull in other sources of health data, including wearable devices, clinical notes, imaging and patient-reported outcomes. The company has made the model, called Forecast 1.0, available to researchers through GitHub and a secure research environment, positioning it as a template for future work in biomarker discovery, risk stratification and precision medicine programmes.