ARTIFICIAL INTELLIGENCE-ASSISTED GENOMIC PREDICTION MODELS FOR EARLY DIAGNOSIS OF COMPLEX HUMAN DISEASES

Abstract

Complex neurodegenerative diseases are difficult to diagnose early on because of the heterogeneity of molecular alterations and overlapping clinical stages. Transcriptomic profiling based on blood is a less invasive approach to detecting disease-related genomic activity. This study developed and evaluated artificial intelligence-assisted genomic prediction models for classifying Alzheimer’s disease, mild cognitive impairment, and cognitively normal controls using peripheral blood gene expression data. Normalized microarray expression data from 329 samples were preprocessed, and missing values were handled using median imputation. Analysis of variance was applied to select the top 1000 discriminative features. Logistic regression, support vector machine, and random forest models were trained using a stratified 80:20 split, and performance was evaluated using accuracy, macro-averaged precision, recall, F1-score, confusion matrix, and multiclass ROC analysis. Functional enrichment analysis was conducted to assess biological relevance. Class-balanced logistic regression achieved the best performance, with an accuracy of 69.7%, macro precision of 0.6917, macro recall of 0.6966, and macro F1-score of 0.6873. ROC analysis showed strong discrimination, with AUC values of 0.89 for controls, 0.83 for Alzheimer’s disease, and 0.76 for mild cognitive impairment. Enrichment results highlighted pathways related to ribosomal activity, oxidative phosphorylation, mitochondrial function, and neurodegeneration. These findings indicate that transcriptomic signatures can support interpretable artificial intelligence-assisted prediction of early disease states.