Constrained neuro fuzzy inference methodology for explainable personalised modelling with applications on gene...

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in healthcare

Background:

• Interpretable machine learning models are crucial for understanding gene expression data and disease mechanisms.
• Early diagnosis in mental health is challenging due to data heterogeneity, necessitating personalized approaches.
• Existing methods often lack personalized interpretability for gene expression datasets.

Purpose of the Study:

• To propose a novel methodology, Personalized Constrained Neuro Fuzzy Inference (PCNFI), for interpretable rule learning from high-dimensional data.
• To enhance personalized interpretability and classification performance in mental health gene expression datasets.
• To identify potential biomarkers for early disease differentiation.

Main Methods:

• Development of the Personalized Constrained Neuro Fuzzy Inference (PCNFI) methodology.
• Application of PCNFI to schizophrenia and bipolar disorder gene expression datasets.
• Comparative performance analysis against standard machine learning methods and benchmark datasets (including cancer).

Main Results:

• PCNFI achieved enhanced interpretability and superior classification performance on mental health datasets.
• The method demonstrated comparable performance to existing benchmarks on a cancer dataset.
• Identified ATRX and TSPAN2 genes as potential biomarkers for differentiating individuals at ultra-high risk, with bipolar disorder, and healthy controls.

Conclusions:

• PCNFI offers a promising approach for interpretable analysis of high-dimensional biological data.
• The identified genes (ATRX, TSPAN2) highlight the role of cognitive and impulsivity differences in mental health.
• PCNFI has potential applications in diagnosis, prognosis, and personalized treatment planning.