Fig. 1: Integrative network analysis pipeline to construct data-driven neuron-specific predictive networks of AD and predict key drivers associated with AD pathology.

a Discovery datasets include whole-genome genotype and RNAseq data of temporal cortex from the MAYO cohort and dorsolateral prefrontal cortex from the ROSMAP cohort in the AMP-AD consortium. Numbers in circles indicate the total number of subjects in each dataset with quality-controlled matched genotype and RNAseq data used in this study, whereas numbers in the table indicate the number of individuals of each phenotype used in the subsequent DE, co-expression module, and predictive network analyses. b Computational deconvolution of bulk-tissue RNAseq data into 5 single cell-type RNAseq sets per cohort dataset, followed by DE analysis and weighted gene co-expression network analysis in each cohort’s neuron-specific gene expression dataset. c mRNA expression and quantitative trait loci association analysis in each dataset provides a source of systematic perturbation for network reconstructions. d Construction of neuron-specific predictive network models and identification of key drivers (master regulators) from each dataset. e Prioritization of key driver targets from both datasets and experimental validation by shRNA-mediated gene knockdown in human iPSC-derived neurons. Venn diagrams on panels b–e indicate cross-validation at each step of the bioinformatics analyses performed independently in parallel for the MAYO and ROSMAP datasets, resulting in a single set of key driver targets. Statistical tests for each comparison are described in the text where relevant. Parts of this figure utilize graphics from Servier Medical Art (smart.servier.com) provided by Servier, licensed under a Creative Commons Attribution 3.0 Unported License. TCX temporal cortex, DLPFC dorsolateral prefrontal cortex, WGCNA weighted gene co-expression network analysis, eQTL expression quantitative trait locus. See also Supplementary Fig. 1.