Genomic Structural Equation Modelling and Mendelian Randomization of Loneliness and Psychiatric/Substance Use Phenotypes

This documentation briefly outlines the stages of analysis and displays some of the results.

All scripts used for data analysis can be found in Analysis and Results.

This project uses the GenomicSEM package and scripts from Schoeler 2021.

These analyses were carried out in R version 4.0.3 and GenomicSEM version 0.0.2.

Getting Started

• Refer to the GenomicSEM wiki
• Install necessary R Packages
• Download necessary software tools, e.g., PLINK
• Download summary statistics files shown in the analysis scripts

Analysis Scripts:

Quality Control and Data Preparation Procedures

Linkage Disequilibrium Score Regression (conducted via GenomicSEM) and Correlation Heatmap

Exploratory Factor Analysis, Confirmatory Factor Analysis, Genomic Structural Equation Modelling

• The multivariate Genomic Structural Equation Model (used for the multivariate GWAS and Mendelian Randomization) and the Path Diagram F1 = Neurodevelopmental and Mood Disorders (NMD); F2 = Substance Use Traits (SUT); F3 = Disorders with Psychotic Features (DPF)

• The supplemental figure without path constraints (negative associations between LONE and F2/SUT and LONE and F3/DPF) - the * indicates no statistical significance at 0.05

Multivariate GWAS (conducted via GenomicSEM using the UCL Myriad Cluster)

GWAS	N (sample)	included SNPs	number of LD-independent genome-wide SNPs	smallest p-value	SNPs (shared)	SNPs (non-shared)
F1/NMD	378372	1968630	22	5.6e-14	5	17
Loneliness	445024	1968630	18	3.8e-17	18	0

Bidirectional Mendelian Randomization

• Mendelian Randomization Report and Plots

• Mendelian Randomization Analysis Script

• this also includes sensitivity tests

• Loneliness to F1/NMD

outcome	exposure	method	nsnp	b	se	pval
F1	loneliness	MR Egger	11	0.5443916	0.2789841	8.28x10-02
F1	loneliness	Inverse variance weighted	11	0.4956359	0.0547951	1.49x10-19
F1	loneliness	Weighted mode	11	0.5238836	0.1343346	2.96x10-03
F1	loneliness	Weighted median	11	0.4810642	0.0803782	2.16x10-09

• Forest plot

• NMD/F1 (Qsnp filtered) to Loneliness

outcome	exposure	method	nsnp	b	se	pval
loneliness	F1_Qsnp	MR Egger	10	0.4439023	0.3011940	1.79x10-01
loneliness	F1_Qsnp	Inverse variance weighted	10	0.3393446	0.0654521	2.16x10-07
loneliness	F1_Qsnp	Weighted mode	10	0.3857237	0.0880167	2.61x10-03
loneliness	F1_Qsnp	Weighted median	10	0.3826074	0.0639450	3.60x10-09

• Forest Plot