The diagnosis of AD and PD are still challenging tasks in clinical practice, partly due to the lack of accessible and accurate blood biomarkers. Here, different classification algorithms were applied to transcriptomics data to identify a panel of genes and an optimal ML model that has the potential as a prediction model for ND.
Using a diverse variety of feature selection and ML approaches the best-performing models were identified for AD and PD respectively. The best-performing model for PD was RF with all genes included (accuracy = 0.702, ROC AUC = 0.743, prAUC = 0.762). The best AD model using a RF model with a 159 gene panel performed better than PD model (accuracy = 0.810, ROC AUC = 0.889, prAUC = 0.919). Many previous AD studies using ML to identify biomarkers have utilised small datasets12. A recent study by Lee and Lee14 tested various feature selection and classification…
