Deep learning for brains? Different linear and nonlinear scaling in UK Biobank brain images vs. machine-learning datasets, bioRxiv, 2019-09-06

AbstractIn recent years, deep learning has unlocked unprecedented success in various domains, especially in image, text, and speech processing. These breakthroughs may hold promise for neuroscience and especially for brain-imaging investigators who start to analyze thousands of participants. However, deep learning is only beneficial if the data have nonlinear relationships and if they are exploitable at currently available sample sizes. We systematically profiled the performance of deep models, kernel models, and linear models as a function of sample size on UK Biobank brain images against established machine learning references. On MNIST and Zalando Fashion, prediction accuracy consistently improved when escalating from linear models to shallow-nonlinear models, and further improved when switching to deep-nonlinear models. The more observations were available for model training, the greater the performance gain we saw. In contrast, using structural or functional brain scans, simple linear models performed on par with more complex, highly parameterized models in agesex prediction across increasing sample sizes. In fact, linear models kept improving as the sample size approached ∼10,000 participants. Our results indicate that the increase in performance of linear models with additional data does not saturate at the limit of current feasibility. Yet, nonlinearities of common brain scans remain largely inaccessible to both kernel and deep learning methods at any examined scale.

biorxiv bioinformatics 100-200-users 2019

Bayesian analysis of GWAS summary data reveals differential signatures of natural selection across human complex traits and functional genomic categories, bioRxiv, 2019-09-01

AbstractUnderstanding how natural selection has shaped the genetic architecture of complex traits and diseases is of importance in medical and evolutionary genetics. Bayesian methods have been developed using individual-level data to estimate multiple features of genetic architecture, including signatures of natural selection. Here, we present an enhanced method (SBayesS) that only requires GWAS summary statistics and incorporates functional genomic annotations. We analysed GWAS data with large sample sizes for 155 complex traits and detected pervasive signatures of negative selection with diverse estimates of SNP-based heritability and polygenicity. Projecting these estimates onto a map of genetic architecture obtained from evolutionary simulations revealed relatively strong natural selection on genetic variants associated with cardiorespiratory and cognitive traits and relatively small number of mutational targets for diseases. Averaging across traits, the joint distribution of SNP effect size and MAF varied across functional genomic regions (likely to be a consequence of natural selection), with enrichment in both the number of associated variants and the magnitude of effect sizes in regions such as transcriptional start sites, coding regions and 5’- and 3’-UTRs.

biorxiv genetics 100-200-users 2019

Robustness and applicability of functional genomics tools on scRNA-seq data, bioRxiv, 2019-09-01

AbstractMany tools have been developed to extract functional and mechanistic insight from bulk transcriptome profiling data. With the advent of single-cell RNA sequencing (scRNA-seq), it is in principle possible to do such an analysis for single cells. However, scRNA-seq data has characteristics such as drop-out events, low library sizes and a comparatively large number of samplescells. It is thus not clear if functional genomics tools established for bulk sequencing can be applied to scRNA-seq in a meaningful way. To address this question, we performed benchmark studies on in silico and in vitro single-cell RNA-seq data. We included the bulk-RNA tools PROGENy, GO enrichment and DoRothEA that estimate pathway and transcription factor (TF) activities, respectively, and compared them against the tools AUCell and metaVIPER, designed for scRNA-seq. For the in silico study we simulated single cells from TFpathway perturbation bulk RNA-seq experiments. Our simulation strategy guarantees that the information of the original perturbation is preserved while resembling the characteristics of scRNA-seq data. We complemented the in silico data with in vitro scRNA-seq data upon CRISPR-mediated knock-out. Our benchmarks on both the simulated and real data revealed comparable performance to the original bulk data. Additionally, we showed that the TF and pathway activities preserve cell-type specific variability by analysing a mixture sample sequenced with 13 scRNA-seq different protocols. Our analyses suggest that bulk functional genomics tools can be applied to scRNA-seq data, outperforming dedicated single cell tools. Furthermore we provide a benchmark for further methods development by the community.

biorxiv bioinformatics 100-200-users 2019

 

Created with the audiences framework by Jedidiah Carlson

Powered by Hugo