Presentation Information
[2I6-OS-7b-01]Variational Energy Landscape Analysis Reveals Collective Trends and Individual Heterogeneity in Mental Health During and After the COVID-19 State of Emergency
〇Masato TSUTSUMI1,2, Taisei KUBO3, Takahiro KATO3, Naoki HONDA2 (1. University of Tsukuba, 2. Nagoya Univerisy, 3. Hokkaido University)
Keywords:
Machine Learning,Energy Landscape
Mental disorders should be understood not as static states, but as dynamic systems involving transitions between states and stability. Pathological states such as major depressive disorder can be modeled as stable states from which escape is difficult. However, conventional cross-sectional assessments cannot identify the dynamic individual differences underlying identical scores. Elucidating this underlying structure is crucial for prognosis prediction and individualized interventions.
This study employs energy landscape analysis to quantify the structure and transition characteristics of mental states. The conventional Pairwise Maximum Entropy Model (pMEM) required large datasets for estimation, limiting its use to group-level analysis. Therefore, this study adopted Variational Expectation–Maximization for pMEM (VEM-MEM), based on a hierarchical Bayesian framework, enabling individual-level energy landscape estimation from limited data.
This method was applied to longitudinal data from depression screening assessments (PHQ-9) collected during and after the COVID-19 state of emergency declaration. The results successfully visualized not only population-level trends but also personal heterogeneity in the stability and transition structure of mental states.
This study employs energy landscape analysis to quantify the structure and transition characteristics of mental states. The conventional Pairwise Maximum Entropy Model (pMEM) required large datasets for estimation, limiting its use to group-level analysis. Therefore, this study adopted Variational Expectation–Maximization for pMEM (VEM-MEM), based on a hierarchical Bayesian framework, enabling individual-level energy landscape estimation from limited data.
This method was applied to longitudinal data from depression screening assessments (PHQ-9) collected during and after the COVID-19 state of emergency declaration. The results successfully visualized not only population-level trends but also personal heterogeneity in the stability and transition structure of mental states.