AI model boosts cancer gene prediction and patient prognosis
The artificial intelligence (AI) model GPT-4, known from its application in ChatGPT, shows impressive capabilities in biomedical research and can be used in many ways for simulations. A simulator developed at MedUni Vienna and based on GPT-4 shows increased accuracy in classifying the importance of genes in cancer cells, as well as in the prognosis of cancer patients. The results of the study were published in the journal Computers in Biology and Medicine.
Utilizing GPT-4 in Biomedical Research
Large language models such as GPT-4 have proven to be extremely useful in various fields, including biomedicine. A research team from MedUni Vienna's Institute of Artificial Intelligence and the CeMM Research Center for Molecular Medicine has shown that GPT-4 can be used effectively as a simulator for biological systems. This study tests the hypothesis that the stepwise simulation of biological and medical processes with GPT-4 leads to better results, which can be crucial for future applications in biomedical research.
SimulateGPT Methodology
Computer simulations of biological processes are essential for biomedical research but often require expertise and manual adjustments. The research team developed "SimulateGPT," a knowledge-based simulation method using structured inputs in GPT-4. This method, validated in various scenarios, including prediction of essential genes in cancer cells and progression-free survival of cancer patients, is designed for basic research and not clinical use.
Enhancing Simulation Accuracy
Language models like GPT-4 rely on text prompts to perform tasks. While models like ChatGPT/GPT-4 can handle simple questions, they may struggle with complex scenarios in biomedicine. By configuring GPT-4 with structured inputs and targeted instructions, the study demonstrated improved predictive accuracy in identifying essential genes in cancer cells and predicting patient outcomes compared to traditional responses.
Future Implications
The study illustrates the potential of large language models such as GPT-4 in developing advanced biomedical simulators. Text-based simulations offer flexibility and interpretability necessary for describing the complexity of living systems. To advance LLM-based biomedical simulators, integrating biological databases, mathematical modeling, and training new AI models with experimental data are suggested directions for future research.
