Graph neural networks (GNNs) have received intense interest as a rapidly expanding class of machine learning models remarkably well-suited for materials applications. To date, a number of successful GNNs have been proposed and demonstrated for systems ranging from crystal stability to electronic property prediction and to surface chemistry and heterogeneous catalysis. However, a consistent benchmark of these models remains lacking, hindering the development and consistent evaluation of new models in the materials field. Here, we present a workflow and testing platform, MatDeepLearn, for quickly and reproducibly assessing and comparing GNNs and other machine learning models. We use this platform to optimize and evaluate a selection of top performing GNNs on several representative datasets in computational materials chemistry. From our investigations we note the importance of hyperparameter selection and find roughly similar performances for the top models once optimized. We identify several strengths in GNNs over conventional models in cases with compositionally diverse datasets and in its overall flexibility with respect to inputs, due to learned rather than defined representations. Meanwhile several weaknesses of GNNs are also observed including high data requirements, and suggestions for further improvement for applications in materials chemistry are proposed.

Theories of adult brain development, based on neuropsychological test results and structural neuroimaging, suggest differential rates of age‐related change in function across cortical and subcortical sub‐regions. However, it remains unclear if these trends also extend to the aging dopamine system. Here we examined cross‐sectional adult age differences in estimates of D2‐like receptor binding potential across several cortical and subcortical brain regions using PET imaging and the radiotracer [18F]Fallypride in two samples of healthy human adults (combined N = 132). After accounting for regional differences in overall radioligand binding, estimated percent difference in receptor binding potential by decade (linear effects) were highest in most temporal and frontal cortical regions (~6–16% per decade), moderate in parahippocampal gyrus, pregenual frontal cortex, fusiform gyrus, caudate, putamen, thalamus, and amygdala (~3–5%), and weakest in subcallosal frontal cortex, ventral striatum, pallidum, and hippocampus (~0–2%). Some regions showed linear effects of age while many showed curvilinear effects such that binding potential declined from young adulthood to middle age and then was relatively stable until old age. Overall, these data indicate that the rate and pattern of decline in D2 receptor availability is regionally heterogeneous. However, the differences across regions were challenging to organize within existing theories of brain development and did not show the same pattern of regional change that has been observed in gray matter volume, white matter integrity, or cognitive performance. This variation suggests that existing theories of adult brain development may need to be modified to better account for the spatial dynamics of dopaminergic system aging.

The evidence that dopamine function mediates the association between aging and cognition is one of the most cited findings in the cognitive neuroscience of aging. However, few and relatively small studies have directly examined these associations. Here we examined correlations among adult age, dopamine D2-like receptor (D2R) availability, and cognition in two cross-sectional studies of healthy human adults. Participants completed a short cognitive test battery and, on a separate day, a PET scan with either the high-affinity D2R tracer [18F]Fallypride (Study 1) or [11C]FLB457 (Study 2). Digit span, a measure of short-term memory maintenance and working memory, was the only cognitive test for which dopamine D2R availability partially mediated the age effect on cognition. In Study 1, age was negatively correlated with digit span. Striatal D2R availability was positively correlated with digit span controlling for age. The age effect on digit span was smaller when controlling for striatal D2R availability. Although other cognitive measures used here have individually been associated with age and D2R availability in prior studies, we found no consistent evidence for significant associations between low D2R availability and low cognitive performance on these measures. These results at best only partially supported the correlative triad of age, dopamine D2R availability, and cognition. While a wealth of other research in human and nonhuman animals demonstrates that dopamine makes critical contributions to cognition, the present studies suggest caution in interpreting PET findings as evidence that dopamine D2R loss is a primary cause of broad age-related declines in fluid cognition.

No abstract

We thank John Pearson, Duke University, for statistical consultation and advice as well as Kevin Seaman for web hosting consultation.Data: Pictures, scatterplots, and statistics for each ROI are available in an interactive app online at AbstractTheories of adult brain development, based on neuropsychological test results and structural neuroimaging, suggest differential rates of age-related change in function across cortical and subcortical sub-regions. However, it remains unclear if these trends also extend to the aging dopamine system. Here we examined cross-sectional adult age differences in estimates of D2-like receptor binding potential across several cortical and subcortical brain regions using PET imaging and the radiotracer [18F]fallypride in two samples of healthy human adults (combined N=132). After accounting for regional differences in overall radioligand binding, estimated percent difference in receptor binding potential by decade (linear effects) were highest in most temporal and frontal cortical regions (~6-16% per decade), moderate in parahippocampal gyrus, pregenual frontal cortex, fusiform gyrus, caudate, putamen, thalamus, and amygdala (~3-5%), and weakest in subcallosal frontal cortex, ventral striatum, pallidum, and hippocampus (~0-2%). Some regions showed linear effects of age while many showed curvilinear effects such that binding potential declined from young adulthood to middle age and then was relatively stable until old age. Overall, these data indicate that the rate and pattern of decline in D2 receptor availability is regionally heterogeneous. However, the differences across regions were challenging to organize within existing theories of brain development and did not show the same pattern of regional change that has been observed in gray matter volume, white matter integrity, or cognitive performance. This variation suggests that existing theories of adult brain development may need to be modified to better account for the spatial dynamics of dopaminergic system aging. LINEAR AND NONLINEAR EFFECT OF AGE ON BINDING POTENTIAL