Betty Tijms, PhD, VU University Medical Center Amsterdam, Amsterdam, Netherlands, discusses the value of random forest analysis in large scale proteomics studies. Random forest classification is built on decision trees that are each trained on different parts of the same training set. Each tree is then used to give a class prediction and the most common answer from their combined output becomes the model’s prediction. This powerful prediction method is able to capture complex dependency patterns much better than other models. Dr Tijms has used this in her work to identify subgroups of patients with Alzheimer’s disease based on cerebrospinal (CSF) protein expression profiles. Each decision tree in the forest considers a different, random selection of proteins to predict a given individual’s subgroup, before all the outputs are combined to give a single answer. Dr Tijms comments on some of the benefits of such an approach in this setting. Decision tree based statistical machine learning methods are gaining prominence in omics studies for their unmatched ability to detect non-linear relationships and give insights on the importance of particular variables. This interview took place at the AD/PD™ 2023 congress in Gothenburg, Sweden.
These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.