Tempus has been building novel biological foundation models and agentic workflows by leveraging its more than 500 petabytes of rich, molecularly grounded data—more than 45 million total de-identified patient journeys, 1.5 million with sequenced data, and over 400,000 records in cancer with full genomic, transcriptomic, imaging, and clinical data. These efforts transform this data into unified patient representations, unlocking actionable insights to speed precision medicine efforts in both the clinic and drug development.

Tempus' latest multimodal, transformer-based model was trained on 2.5 million longitudinal records encapsulating more than 250 million pages of clinical notes, 450,000 digitized medical images, and 500,000 genomic and transcriptomic sequences. By aggregating modalities derived from billion-parameter foundation models, it is designed to address thousands of prediction objectives anchored in overall survival (OS) and progression-free survival (PFS), without requiring additional data or model fine-tuning. This architecture significantly reduces the time and data required to produce hundreds of clinically relevant insights for clinical trial design, patient risk prediction, and novel multimodal diagnostics.

In a zero-shot setting without any further training, Tempus leveraged this patient trajectory model and a series of agentic workflows to unlock insights in a cohort of more than 1.2 million de-identified records with robust multimodal data.

As a primary proof of concept, Tempus' model was utilized to analyze EGFR-mutant NSCLC patients treated with osimertinib, the current frontline standard of care third-generation EGFR-inhibitor. Tempus assessed whether the model could accurately stratify response to the standard of care treatment of osimertinib in patients with known clinically actionable biomarkers like EGFR. This tests the model's ability to learn a composite of features that predict response outside of known biological and clinical features to identify patients more likely to experience poor response to current therapies.

Without any pre-specified training, the model demonstrated:

  • Predictive Accuracy: A C-index of 0.802 for overall survival (p value < 0.001).
  • Significant Survival Stratification: Hazard Ratio of 4.536 (95% CI: [3.289, 6.255]) between high- and low-risk subgroups.

The model was evaluated against a range of mutation profiles (including more than 30 EGFR-specific features including L858R, Exon 19 del); and the results still produced prognostic value independent of molecular and clinical subgroups, significantly stratifying the overall survival of TP53(+) patients (HR=5.96), as well as progression-free survival in patients without CNS metastasis (HR=1.94). This outcome is just one example of how Tempus' model can produce novel clinical insights, as similar results have been observed for predicting overall survival in other patient populations.

Tempus' multimodal patient trajectory models have also demonstrated substantial utility for drug development activities. For example, the models have successfully predicted outcomes of patient cohorts that mirrored pre-established clinical trials. In NSCLC, Tempus was able to assess cohorts mirroring those of practice-changing clinical trials (KEYNOTE-189, FLAURA-2, and DESTINY), and the multimodal patient trajectory model was able to zero-shot outperform standard approaches, such as Cox-PH modeling. These outcomes indicate substantial utility for biopharmaceutical developers, including the potential to assess features that may drastically impact clinical trial performance before a trial begins.

"Foundation models, combined with agentic workflows, will help unlock the full potential of precision medicine — enabling rapid hypothesis testing and dramatically compressing the time from discovery to clinical application," said Eric Lefkofsky, Founder and CEO of Tempus. "The fact that our general purpose models are already outperforming highly tuned smaller models bodes well for the ability of our novel biological multimodal foundation models to improve clinical trial design and biomarker development, empowering physicians and life science companies in the pursuit of new diagnostics and novel medicines."