RT Journal Article
SR Electronic
T1 A computational knowledge engine for human neuroscience
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 701540
DO 10.1101/701540
A1 Beam, Elizabeth
A1 Potts, Christopher
A1 Poldrack, Russell A.
A1 Etkin, Amit
YR 2019
UL http://biorxiv.org/content/early/2019/07/14/701540.abstract
AB Functional neuroimaging has been a mainstay of human neuroscience for the past 25 years. The goal for this research has largely been to understand how activity across brain structures relates to mental constructs and computations. However, interpretation of fMRI data has often occurred within knowledge frameworks crafted by experts, which have the potential to reify historical trends and amplify the subjective biases that limit the replicability of findings. 1 In other words, we lack a comprehensive data-driven ontology for structure-function mapping in the human brain, through which we can also test the explanatory value of current dominant conceptual frameworks. Ontologies in other fields are popular tools for automated data synthesis,2, 3 yet relatively few attempts have been made to engineer ontologies in a data-driven manner.4 Here, we employ a computational approach to derive a data-driven ontology for neurobiological domains that synthesizes the texts and data of nearly 20,000 human neuroimaging articles. The data-driven ontology includes 6 domains, each defined by a circuit of brain structures and its associated mental functions. Several of these domains are omitted from the leading framework in neuroscience, while others uncover novel combinations of mental functions related to common brain circuitry. Crucially, the structure-function links in each domain better replicate across articles in held-out data than those mapped from the dominant frameworks in neuroscience and psychiatry. We further show that the data-driven ontology partitions the literature into modular subfields, for which the domains serve as generalizable archetypes of the structure-function patterns observed in single articles. The approach to computational ontology we present here is the most comprehensive functional characterization of human brain circuits quantifiable with fMRI. Moreover, our methods can be extended to synthesize other scientific literatures, yielding ontologies that are built up from the data of the field.