Studying the molecular development of the human brain presents unique challenges for selecting the best data analysis approach. The rare and valuable nature of human postmortem brain samples, especially for studies examining development, means that those studies have small sample sizes (n) but often include measurements (p) for a large number of genes or proteins for every sample. Thus, most of those data sets have a structure that is p >> n, which introduces the problem of sparsity. Here we present a guide to analyzing human brain development data by focusing on sparsity-based clustering methods developed for small sample sizes. We test different methods and identify an application of sparse K-means clustering called Robust Sparse K-means Clustering (RSKC) that does a good job revealing clusters of samples that reflect lifespan stages from birth to aging. The algorithm adaptively selects a subset of the genes or proteins that contributes to generating clusters of samples that are spread across the lifespan. This approach addresses a problem in current studies that were unable to identify postnatal clusters. The guide illustrates that careful selection of the clustering method is essential to reveal meaningful aspects of human brain development.
bioRxiv Subject Collection: Neuroscience