Laboratory Medicine & Pathology

High-throughput technologies ranging from high-content microscopy screening to automated mass spectrometry pipelines and peptide microarray assays enable the generation of big medical data sets in a brief period of time with the requirement of very small analyte volumes. As a result, extremely large multiomics data sets can be generated which are ideally suited for statistical learning procedures using well established classifiers such as random forests or multivariate logistic regression models. To extract meaningful information from the data, a major challenge comprises the dimensionality reduction and

unsupervised pattern recognition via machine learning algorithms including Fisher discriminant analysis, shrunken centroid analysis or clustering. Another problem to be tackled is represented by multivariate feature collinearities and ways have to be identified to remove this ambiguity. Yet, the reward for this mathematical rigor is the generation of reliable classifiers which can quickly stratify individual patients into disease/treatment subpopulations with acceptable sensitivity and specificity from multiomics data sets. Above all, technology suppliers such as opentrons and Oxford Nanopore Technologies promise the automated generation of multiomics data sets for a reasonable price.