Significance: Applying artificial intelligence tools to a highly complex question of immunology, we show that a deep neural network can learn to predict the patterns of chromatin opening across 81 stem and differentiated cells across the immune system, solely from the DNA sequence of regulatory regions. It does so by discovering ab initio the binding motifs for known master regulators, along with some unknown ones, and their combinatorial operation. These predictions validated biochemically, and a mouse-trained neural network predicts human enhancer/promoter activity much better than sequence comparisons would. Beyond serving as a trove of testable functional frameworks, this work is important in showing how massively complex integrated questions of immunology can be handled with such tools.

Complexity of cell-type composition has created much skepticism surrounding the interpretation of bulk tissue transcriptomic studies. Recent studies have shown that deconvolution algorithms can be applied to computationally estimate cell-type proportions from gene expression data of bulk blood samples, but their performance when applied to brain tissue is unclear. Here, we have generated an immunohistochemistry (IHC) dataset for five major cell-types from brain tissue of 70 individuals, who also have bulk cortical gene expression data. With the IHC data as the benchmark, this resource enables quantitative assessment of deconvolution algorithms for brain tissue. We apply existing deconvolution algorithms to brain tissue by using marker sets derived from human brain single cell and cell-sorted RNA-seq data. We show that these algorithms can indeed produce informative estimates of constituent cell-type proportions. In fact, neuronal subpopulation can also be estimated from bulk brain tissue samples. Further, we show that including the cell-type proportion estimates as confounding factors is important for reducing false associations between Alzheimer’s disease phenotypes and gene expression. Lastly, we demonstrate that using more accurate marker sets can substantially improve statistical power in detecting cell-type specific expression quantitative trait loci (eQTLs). Gene expression data generated from a tissue sample reflects an average gene expression profile across heterogeneous populations of cells. Because composition of constituent cell-types can vary across individuals (due to technical or biological factors), differential gene expression analysis requires estimating and adjusting for such cellular heterogeneity. While many deconvolution algorithms for estimating cellular composition from tissue gene expression data have been tested extensively in blood, their performance when applied to brain tissue is unclear. To address this gap, we generated an immunohistochemistry (IHC) dataset for five major cell-types from brain, in order to apply and then assess deconvolution algorithms for application to brain gene expression datasets. We show that these algorithms can indeed produce informative estimates of constituent cell-type proportions. Further, we show that adjusting for estimated cell-type proportions across individuals when conducting differential gene expression analysis is important in reducing false associations.

Brain connectivity arises from interactions across biophysical scales, ranging from molecular to cellular to anatomical to network level. To date, there has been little progress toward integrated analysis across these scales. To bridge this gap, from a unique cohort of 98 individuals, we collected antemortem neuroimaging and genetic data, as well as postmortem dendritic spine morphometric, proteomic and gene expression data from the superior frontal and inferior temporal gyri. Through the integration of the molecular and dendritic spine morphology data, we identified hundreds of proteins that explain interindividual differences in functional connectivity and structural covariation. These proteins are enriched for synaptic structures and functions, energy metabolism and RNA processing. By integrating data at the genetic, molecular, subcellular and tissue levels, we link specific biochemical changes at synapses to connectivity between brain regions. These results demonstrate the feasibility of integrating data from vastly different biophysical scales to provide a more comprehensive understanding of brain connectivity.