Forschungsseminar: Vortrag von Artem Shmatko

Learning the natural history of human disease with generative transformers (an AI model) - Artem Shmatko, Doctoral Researcher, DKFZ German Cancer Research Center

Researchers have investigated how well modern AI models can understand text – for example, systems such as ChatGPT. The working group „Artificial Intelligence in Oncology“ (DKFZ Heidelberg), along with its PhD candidate Artem Shmatko, has further developed this technology: they adapted the GPT architecture (Generative Pretrained Transformer) so that it no longer processes language, but medical data from electronic health records (EHR).

The AI views a person's entire medical history as a chronological sequence of medical events, such as diagnoses or treatments. On this basis, the Delphi-2M model developed can predict the risk of more than 1,000 different diseases. In addition, the system recognises which diseases frequently occur together (known as comorbidities) and highlights where there are gaps or distortions in the underlying health data.

In short:
The Delphi-2M model analyses medical histories in a similar way to how other AI analyses texts, thereby helping to better understand disease risks and correlations between diseases.

The research seminar will focus on the following publication:
Shmatko, A., Jung, A.W., Gaurav, K. et al. Learning the natural history of human disease with generative transformers. Nature 647, 248–256 (2025). https://doi.org/10.1038/s41586-025-09529-3

Um die Forschungsseminare für alle angenehm zu gestalten, bitten wir Sie herzlich Folgendes zu beachten:

• Sollten Sie verspätet eintreffen, treten Sie bitte leise in den Raum, schliessen die Tür vorsichtig und nehmen leise Platz.
• Es ist in Ordnung, diskret eine Kleinigkeit zu essen, bitte jedoch ohne Lärm zu verursachen oder andere Teilnehmende abzulenken.
• Schenken Sie dem Gastredner, der Gastrednerin Ihre Aufmerksamkeit und vermeiden Sie Tätigkeiten am Laptop.

Vielen Dank für Ihr Mitwirken.

• 

  Übersicht Forschungsseminare

  Entdecken Sie weitere bevorstehende Forschungsseminare an unserer Hochschule.