CPST-GAN: Conditional Probabilistic State Transition Generative Adversarial Network With the Biomedical Large...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Genetics

Background:

• Alzheimer's disease (AD) risk prediction is vital for early intervention but faces challenges in multiomics feature extraction and fusion.
• Existing methods often overlook the complex, multilevel evolutionary mechanisms underlying AD progression.
• Integrating genetic regulation with brain lesion development is key to understanding AD pathogenesis.

Purpose of the Study:

• To develop an advanced risk prediction model for Alzheimer's disease by integrating biomedical large foundation models and conditional generative adversarial networks (GAN).
• To mine the dynamic evolutionary patterns of AD by considering gene regulatory effects on brain lesions.
• To improve the accuracy and reliability of early AD risk prediction through enhanced feature fusion and evolutionary analysis.

Main Methods:

• Utilized biomedical large foundation models to construct high-quality imaging genetic features.
• Developed a conditional probabilistic state transition mathematical model to represent AD progression under genetic regulation.
• Proposed a conditional probabilistic state transition GAN (CPST-GAN) to mine dynamic evolutionary patterns by fusing brain imaging and genetic data.

Main Results:

• CPST-GAN effectively mines dynamic evolutionary patterns of Alzheimer's disease.
• The proposed algorithm demonstrates superior performance in AD risk prediction compared to existing methods.
• Experiments on public datasets validate the effectiveness of CPST-GAN in evolutionary pattern mining and risk prediction.

Conclusions:

• CPST-GAN offers a reliable intelligence algorithm for the early intervention of Alzheimer's disease.
• The study provides new insights into AD pathogenesis by considering gene-brain lesion regulatory effects.
• This research advances the integration of AI and multiomics data for neurodegenerative disease research.