Machine learning model based on plasma proteomics for identification of Parkinson's disease

This repository contains code used in a study aimed at developing and evaluating an ensemble classification machine learning model tasked with identifying Parkinson's disease in a mixed cohort of Parkinson's disease cases, neurologically healthy controls, and study participants with other neurodegenrative diseases. The ensemble model consists of 6 base models and makes decisions using plasma abundances of 11 proteins which were measured using Olink Proximity Extension Assay technology. Three cohorts were included in this study - the Main Cohort whose plasma protein meassurements were generated by the authors, and two independent cohorts designated as the External Validation Sets. The External Validation Sets consisted of study participants from the Parkinson's Disease Biomarker Program (PDBP) and the UK Biobank cohorts. The files in this repository contain the code for quality control and main analysis for this study. The files, numbered in order are described as follows: 01 - Data extraction and quality control for the study participants in the PDBP cohort 02 - Data extraction and quality control for study participants in the UK Biobank cohort 03 - Main analysis which includes: Quality control of the Main Cohort Splitting the Main Cohort into a Training Set and a Test Set Differential protein abundance analysis Feature selection - extraction of proteins that are predictive of Parkinson's disease status Machine learning model training using the Training Set Evaluation of machine learning model performance in the Test Set, PDBP and UK Biobank Shapley Additive Explanations (SHAP) 04 - Extraction of genetic variants in LRRK2 for control samples who were labelled as positive for Parkinson's disease by the machine learning model 05 - Network analysis 06 - A script to determine the combination of base models that maximizes model performance in an ensemble model 07 - A script to evaluate the contribution of each protein to the model and quantify risk using the SHAP framework 08 - A script to deploy the ensemble machine learning model on a web portal for public access