Biology subjects

Spalletta, G.

Publications and source records attributed to Spalletta, G..

4 recordsLinked to original sources

bioRxiv2018

Deep Learning for Quality Control of Subcortical Brain 3D Shape Models

We present several deep learning models for assessing the morphometric fidelity of deep grey matter region models extracted from brain MRI. We test three different convolutional neural net architectures (VGGNet, ResNet and Inception) over 2D maps of geometric features. Further, we present a novel geometry feature augmentation technique based on parametric spherical mapping. Finally, we present an approach for model decision visualization, allowing human raters to see the areas of subcortical shapes most likely to be deemed of failing quality by the machine. Our training data is comprised of 5200 subjects from the ENIGMA Schizophrenia MRI cohorts, and our test dataset contains 1500 subjects from the ENIGMA Major Depressive Disorder cohorts. Our final models reduce human rater time by 46-70%. ResNet outperforms VGGNet and Inception for all of our predictive tasks.

neuroscience↗

bioRxiv2018

Gene-Based Analysis in HRC Imputed Genome Wide Association Data Identifies Three Novel Genes For Alzheimer’s Disease

A novel POLARIS gene-based analysis approach was employed to compute gene-based polygenic risk score (PRS) for all individuals in the latest HRC imputed GERAD (N cases=3,332 and N controls=9,832) data using the International Genomics of Alzheimers Project summary statistics (N cases=13,676 and N controls=27,322, excluding GERAD subjects) to identify the SNPs and weight their risk alleles for the PRS score. SNPs were assigned to known, protein coding genes using GENCODE (v19). SNPs are assigned using both 1) no window around the gene and 2) a window of 35kb upstream and 10kb downstream to include transcriptional regulatory elements. The overall association of a gene is determined using a logistic regression model, adjusting for population covariates.\n\nThree novel gene-wide significant genes were determined from the POLARIS gene-based analysis using a gene window; PPARGC1A, RORA and ZNF423. The ZNF423 gene resides in an Alzheimers disease (AD)-specific protein network which also includes other AD-related genes. The PPARGC1A gene has been linked to energy metabolism and the generation of amyloid beta plaques and the RORA has strong links with genes which are differentially expressed in the hippocampus. We also demonstrate no enrichment for genes in either loss of function intolerant or conserved noncoding sequence regions.

genetics↗

bioRxiv2018

Elevated plasma ceramide levels in post-menopausal women

Circulating ceramide levels are abnormally elevated in age-dependent pathologies such as cardiovascular diseases, obesity and Alzheimers disease. Nevertheless, the potential impact of age on plasma ceramide levels has not yet been systematically examined. In the present study, we quantified a focused panel of plasma ceramides and dihydroceramides in a cohort of 164 subjects (84 women) 19 to 80 years of age. After adjusting for potential confounders, multivariable linear regression analysis revealed a positive association between age and ceramide (d18:1/24:0) ({beta} (SE) = 5.67 (2.38); p = .0198) and ceramide (d18:1/24:1) ({beta} (SE) = 2.88 (.61); p < .001) in women, and between age and ceramide (d18:1/24:1) in men ({beta} (SE) = 1.86 (.77); p = .0179). In women of all ages, but not men, plasma ceramide (d18:1/24:1) was negatively correlated with plasma estradiol (r = -0.294; p = .007). Finally, in vitro experiments in human cancer cells expressing estrogen receptors showed that incubation with estradiol (10 nM, 24 h) significantly decreased ceramide accumulation. Together, the results suggest that aging is associated with an increase in circulating ceramide levels, which in post-menopausal women may be at least partially dependent on lower estradiol levels.

biochemistry↗

bioRxiv2017

Machine Learning for Large-Scale Quality Control of 3D Shape Models in Neuroimaging

As very large studies of complex neuroimaging phenotypes become more common, human quality assessment of MRI-derived data remains one of the last major bottlenecks. Few attempts have so far been made to address this issue with machine learning. In this work, we optimize predictive models of quality for meshes representing deep brain structure shapes. We use standard vertex-wise and global shape features computed homologously across 19 cohorts and over 7500 human-rated subjects, training kernelized Support Vector Machine and Gradient Boosted Decision Trees classifiers to detect meshes of failing quality. Our models generalize across datasets and diseases, reducing human workload by 30-70%, or equivalently hundreds of human rater hours for datasets of comparable size, with recall rates approaching inter-rater reliability.

neuroscience↗