The combination of new bioinformatics approaches with high throughput proteomics are increasingly important across many areas of biological research. In particular, discovering new biomarkers can help in the unveiling of molecular mechanism of disease and additionally has enormous potential to enhance clinical applications in early detection and prognostic classification. Although biomarkers discovery has been a central focus of proteomics over the past few decades, many discoveries fail to validate in independent patient cohorts and as a result are not useful in clinical applications. For these reasons discovery of reliable molecular markers of disease is still a daunting task. In this work we propose that in most diseases biomarker discovery is not likely to deliver a single protein marker capable of predicting, prognosticating, or tracking disease progression. Instead, biomarkers if they exist, are likely to be complex and dynamic. Here we suggest that proteins in circulating blood exosomes contain fingerprint proteoforms that can reflect the state of the releasing cells, such as degenerative neurons undergoing Parkinson's disease. We use the Random-forest machine learning algorithm to identify a group of peptides that are capable to efficiently classify patients versus control samples. Once identified, these biomarker peptides will form the core of a ‘tool box’ for ongoing and continuous classification of Parkinson’s disease blood samples for patient diagnosis, stratification, prognosis and therapy response.
In this work we used to different batches of purified exosomes: a first from a group of 62 Parkinson’s disease patients and case controls (spouses), while a second set of samples was a larger cohort of 132 samples. Analysis of these two cohorts have showed consistent signature and preliminary analysis of the identified peptides suggest that many of them belong to proteins involved in cytoskeleton integrity and endoplasmic reticulum stress.