A generalizable computational framework for integrating heterogeneous biomarkers into interpretable scalar risk...

A central challenge in biomedical informatics remains the integration of heterogeneous biomarker data into unified, interpretable, and clinically meaningful representations. In this study, we present a computational framework for the construction of unified scalar representations derived from multiple blood-based biomarkers, with the objective of capturing latent structures associated with mental health status. As a proof-of-principle, we instantiate this framework through the Mental Disorder Risk Index (MDRI), a scalar and dimensionless representation that integrates multiple biomarkers through an optimization-based and severity-sensitive aggregation strategy. The proposed framework includes linear and nonlinear formulations, with parameter estimation performed through a combination of deterministic and meta-heuristic optimization strategies. Model performance is evaluated using established metrics, mainly the Matthews correlation coefficient. The framework is applied to a dataset composed of eight biomarkers and respective mental health scores. The results indicate that stable scalar representations can be obtained within the normalized biomarker space, with robustness observed in several optimization trajectories, confirming the stability of the underlying latent structure of the biomarkers. Sensitivity analysis reveals heterogeneous contributions of the biomarkers, indicating that the proposed index captures non-trivial multivariate relationships. These findings establish the computational viability of the construction of unified biomarker-based indices and provide a structured basis for the development of interpretable, scalable, and generalizable representations of complex biomedical data.