AI algorithm improves cancer gene predictions and patient prognosis

The GPT-4 artificial intelligence (AI) model, known for its use in ChatGPT, has demonstrated impressive capabilities in biomedical research and can be used in a variety of ways for modelling. Developed at MedUni Vienna and based on GPT-4, this simulator has shown improved accuracy in classifying the importance of genes in cancer cells and predicting cancer patients. The research findings were published in the journal Computers in Biology and Medicine.A study at MedUni Vienna has developed a large-scale AI model called GPT-4, which has shown improved accuracy in biomedical research.

Large-scale language models such as GPT-4 have proven extremely useful in a variety of fields, including biomedicine. A research team from MedUni Vienna\'s Institute of Artificial Intelligence and the CeMM Research Center for Molecular Medicine, led by Matthias Samwald and Christoph Bock, has shown that a large language model such as GPT-4 can be used effectively as a simulator for biological systems.

The study tests the hypothesis that the stepwise simulation of biological and medical processes with GPT-4 leads to better results. This is important for future application in biomedical research, as well as to understand these new models.

IT modeling of biological processes is an important tool for biomedical studies, but generally requires great knowledge and manual adjustments. The research team developed “SimulateGPT”, a knowledge-based modeling method through structured input data in GPT-4. This method has been tested and validated by experts in various scenarios such as mouse experiments, supporting sepsis treatment, predicting essential genes in cancer cells and progression-free survival of cancer patients. The method is intended for basic research and is not intended for clinical use.

Language models, such as GPT-4, are due to a text entry, the so-called  "indicates ", to carry out problems or solutions specific to problems. Current models such as ChatGPT/GPT-4 answer simple questions directly but struggle to address more complex scenarios common in biomedicine. In this study, the researchers configured GPT-4 with structured input and target instructions, allowing it to simulate specific scenarios in detail using text. The study showed that the GPT-4-based simulator achieved significantly better results, and the study's experiments demonstrated that biomedical experts preferred SimulateGPT's predictions over direct GPT-4 responses. Furthermore, compared with the traditional GPT-4 response, SimulateGPT improved the accuracy of identifying essential genes in cancer cells and predicting progression-free survival in cancer patients.