A Graph-Driven Machine Learning Framework for Biomedical Prediction and Spectral Network Modeling

Abstract

The increasing intricacy of biomedical systems has necessitated the development of integrated
computational models that merge predictive precision with structural clarity. This article outlines the
creation and validation of a machine learning-based clinical decision support system for the early
evaluation of heart attack risk, enhanced by a conceptual network architecture relevant to broader
biological contexts, including microbiomes. A clinical dataset comprising nine cardiovascular markers
was used to assess three supervised algorithms: K-Nearest Neighbors, Naive Bayes, and Decision Tree.
The decision tree achieved an accuracy of 98%, validating its efficacy for structured health data. A
Python-based interface was developed, facilitating both manual and PDF data input for real-time clinical
use.
In addition to categorisation, each patient was represented as a node in a similarity network, facilitating
the conversion of flat data into a topological structure. The outputs of machine learning were interpreted
as node labels, serving as the foundation for subsequent applications in microbiome-host interaction
networks and gene co-expression research. This method facilitates the application of spectral graph
techniques, including Laplacian eigenvalue analysis and matrix functionals (e.g., exp(A), cosh(A)), to
investigate structural disturbances in biological systems.
This twin contribution—an accurate clinical prediction tool and a transferable graph-based modelling
framework—facilitates transdisciplinary applications in systems biology and computational
epidemiology. This study advances digital health initiatives by integrating machine learning with
topological reasoning, providing a reproducible basis for predictive modelling in biologically intricate,
network-structured fields.