BrainGENIE: The Brain Gene Expression and Network Imputation Engine

Abstract

In vivo experimental analysis of human brain tissue poses substantial challenges and ethical concerns. To address this problem, we developed a computational method called the Brain Gene Expression and Network-Imputation Engine (BrainGENIE) that leverages peripheral-blood transcriptomes to predict brain tissue-specific gene-expression levels. Paired blood-brain transcriptomic data collected by the Genotype-Tissue Expression (GTEx) Project was used to train BrainGENIE models to predict gene-expression levels in ten distinct brain regions using whole-blood gene-expression profiles. The performance of BrainGENIE was compared to PrediXcan, a popular method for imputing gene expression levels from genotypes. BrainGENIE significantly predicted brain tissue-specific expression levels for 2947-11,816 genes (false-discovery rate-adjusted p < 0.05), including many transcripts that cannot be predicted significantly by a transcriptome-imputation method such as PrediXcan. BrainGENIE recapitulated measured diagnosis-related gene-expression changes in the brain for autism, bipolar disorder, and schizophrenia better than direct correlations from blood and predictions from PrediXcan. We developed a convenient software toolset for deploying BrainGENIE, and provide recommendations for how best to implement models. BrainGENIE complements and, in some ways, outperforms existing transcriptome-imputation tools, providing biologically meaningful predictions and opening new research avenues.

Conflict of interest statement

In the past year, Dr. Faraone received income, potential income, travel expenses continuing education support, and/or research support from Aardvark, Aardwolf, Akili, Atentiv, Corium, Genomind, Ironshore, Medice, Noven, Otsuka, Sandoz, Sky Therapeutics, Supernus, Tris, and Vallon. With his institution, he has US patent US20130217707 A1 for the use of sodium-hydrogen exchange inhibitors in the treatment of ADHD. In previous years, he received support from Alcobra, Arbor, Aveksham, Axsome, CogCubed, Eli Lilly, Enzymotec, Impact, Janssen, KemPharm, Lundbeck/Takeda, Shire/Takeda, McNeil, NeuroLifeSciences, Neurovance, Novartis, Pfizer, Rhodes, Shire, and Sunovion. He also receives royalties from books published by Guilford Press: Straight Talk about Your Child’s Mental Health; Oxford University Press: Schizophrenia: The Facts; and Elsevier: ADHD: Non-Pharmacologic Interventions. In addition, he is the program director of www.adhdinadults.com. In the past year, Dr. Glatt has received royalties from a book published by Oxford University Press: Schizophrenia: The Facts, and consulting fees from Cohen Veterans Bioscience. Dr. Cairns is supported by NHMRC project grants (1147644 and 1188493) and an NHMRC Senior Research Fellowship (1121474), and a University of Newcastle College of Health Medicine and Wellbeing, Gladys M Brawn Senior Fellowship.

© 2023. The Author(s).

Figures

Fig. 1. Schematic illustrating the process for training BrainGENIE using paired blood–brain transcriptome data from the GTEx dataset.

BrainGENIE is trained using paired blood and brain transcriptome profiles collected by GTEx (v8) from adult donors. A single 5-fold cross-validation is performed to estimate the predictive performance of BrainGENIE separately for each brain region. BrainGENIE uses top principal components of transcriptome-wide blood-based gene expression as features to predict brain-regional gene expression levels. The metric used for prediction performance was the coefficinent of determination (R2) to measure how well predicted per-gene expression levels captured observed gene expression levels in the validation folds. Model performance was summarized over the 5 validation folds to obtain an estimate of prediction performance for BrainGENIE.

Fig. 2. BrainGENIE recapitulates disorder-related changes in gene expression found in postmortem brain.

Concordance of case-control differential gene expression (DGE) signals obtained by BrainGENIE and S-PrediXcan compared to A DGE signals derived from postmortem cortical microarray meta-analyses for ASD, BD, and SCZ, B DGE signals derived from RNA-sequencing analysis for ASD, BD, and SCZ by the PsychENCODE Consortium, and C DGE signals obtained from postmortem prefrontal cortex RNA-sequencing analysis for SCZ by the CommonMind Consortium. ASD autism spectrum disorder, BD bipolar disorder, and SCZ schizophrenia. Symbols for significance thresholds: p < 0.05 (*), FDRp ≤ 1 × 10−5 (***), FDRp ≤ 1 × 10−10 (#), FDRp ≤ 1 × 10−20 (##).