# Papers & research

Methods, models, and clinical evaluations underwriting the platform.

Our research advances medical AI for healthcare — foundation models that learn from real patient data and are evaluated on real clinical endpoints. Published work from our team spans world models for longitudinal EHR, whole-genome encoders, molecular LLMs, and high-resolution vision–language models.

## Publications

**2026 · World model / longitudinal EHR**
[The patient is not a moving document: a world-model training paradigm for longitudinal EHR](https://arxiv.org/abs/2601.22128)
Introduces SMB-Structure, a world model for structured EHR that combines a Joint-Embedding Predictive Architecture (JEPA) with supervised next-token prediction. SFT grounds the model to reconstruct future patient states in token space; JEPA predicts those futures in latent space from the initial representation alone, forcing trajectory dynamics to be encoded before the next state is observed. Validated across 40,000 patients on long-horizon prediction tasks.

**2025 · Oncology & whole-genome sequencing**
[GenVarFormer: Predicting gene expression from long-range mutations in cancer](https://arxiv.org/abs/2509.25573)
A whole-genome-sequencing foundation model trained to predict the functional consequence of variants on gene expression. Distinguishes rare driver mutations from passenger mutations in the non-coding genome. State-of-the-art on downstream cancer tasks.

**2025 · Molecular language model**
[Patient-specific biomolecular instruction tuning of Graph-LLMs](https://arxiv.org/abs/2509.22853)
Links proteomic graph neural networks to language, creating a shared representation space between molecular and cellular foundation models. The approach generalizes to any graph-based representation at the cellular level.

**2025 · Electronic health records**
[Building the EHR foundation model via next-event prediction](https://arxiv.org/abs/2509.25591)
Reframes EHRs as timestamped chains of clinical events and fine-tunes large language models to predict the next event, improving temporal reasoning over disease trajectories. +4.6% AUROC over task-specific EHR models.

**2024 · Vision–language / medical imaging**
[Advancing high-resolution vision–language models in biomedicine](https://doi.org/10.48550/arXiv.2406.09454)
Foundational paper showcasing the strength of the Standard Model approach across high-resolution biomedical imagery and language. Establishes the vision–language backbone that later scale-specific papers build on.

## Related pages

- [Model hub](https://standardmodel.bio/model-hub.html) — the weights behind these papers
- [Install & quickstart](https://standardmodel.bio/install.html)
- [Team](https://standardmodel.bio/team.html)
- [Blog](https://blog.standardmodel.bio/) — announcements and technical writing
- Research collaborations: info@standardmodel.bio

© 2026 Standard Model Biomedicine · San Francisco, CA