NeuroPredictome: A Data-Driven Predictome Linking Neuroimaging to Phenotype

Neuroimaging studies generally provide evidence only on a narrow aspects of the human brain function, suffering from small sample sizes and are hard to reproduce. These factors severely limit synthesis of neuroimaging findings and our ability to reach a global view of human brain organization, mapping and decoding. In this paper, we present a novel prediction based framework called Neuropredictome that allows identification of statistically significant linkages between phenotypes and neuroimaging features on UK-Biobank data. We evaluate phenotype linkage to brain fMRI activity on 4926 variables pertaining to the health, physiology, psychology, social and economic state for 19,831 subjects. We corroborate our identified regions of the brain with previous work by providing a novel quantitative evaluation of how well our results align with existing meta-analyses of 14,371 published neuroimaging research articles. Our analysis is presented as a public resource at https://neuropredictome.com providing an interpretable view of human brain organization and decoding, to assist in hypothesis generation and evaluating future studies.