NDD Risk Variant Clustering

Analysis of genome-wide association study (GWAS) significant single-nucleotide polymorphisms (SNPs) across Alzheimer's disease (AD), Parkinson's disease (PD), amyotrophic lateral sclerosis (ALD), frontotemporal dementia (FTD) and Lewy body dementia (LBD).

Processing

1. Preprocessing

• ROSMAP/MayoRNAseq/MSBB: Concatenate individual chromosome vcf files and convert to PLINK binary files.
• ADNI: Merge individual chromosome PLINK binary files.
• DementiaSeq: Separate FTD, LBD, and ALS PLINK binary files.
• AMP-PD/ADSP: No processing needed to get PLINK binary files.
• All cohorts: Attach phenotype and sex data to .fam files.

2. QC

• Individual cohort QC run through GenoTools (https://github.com/dvitale199/GenoTools).
• ROSMAP/MayoRNAseq/MSBB, ADNI, AMP-PD are the cohorts that undergo genetic sex confirmation. Other cohorts did not have high-quality 23rd chromosome data available.

3. LiftOver

• From hg19 to hg38.
• ROSMAP/MayoRNAseq/MSBB, AD stats, PD stats, ALS stats and FTD stats.
• Using code adapted from https://genome.sph.umich.edu/wiki/LiftOver#Lift_genome_positions.

4. Merge

• Merge cohorts across common SNPs.
• Make phenotype/cohort information file.
• Use annovar to convert from chr:bp:ref:alt ro rsID (https://annovar.openbioinformatics.org/en/latest/).
• Run relatedness and duplicate prune as well as ancestry prune to ensure all samples are of EUR ancestry (again using GenoTools).
• Extract GWAS significance (i.e. p < 5e-08) variants from summary stats.
• Annotate gene-associated SNPs. When a SNP did not have a gene association, annotate the nearest gene.
• Sample 1000 cases from each disease.

5. Adjustment

• GenoML munging and PC adjustment (https://github.com/GenoML/genoml2).
  
  

Analysis

1. Get UMAP Parameters

• Runs via swarm. Returns a .txt file listing the top parameter combinations based on the Disease ~ Cluster regressions.

2. Cluster Consistency

• Runs via swarm. Gets iterative cluster membership for multi-disease and disease-specific clusters

3. Analysis

• Queues the swarm jobs for Get UMAP Parameters and Cluster Consistency.
• Run full analysis. This needs to be separately run with disease set to None for multi-disease analysis and disease set to 'ad', 'pd', 'als', 'ftd', and 'lbd' for disease-specific analyses.
• Note that for single disease analyses disease_regression() should be skipped (Disease ~ Cluster regression will fail when only one disease is present).
  
  

Streamlit

https://nih-card-ndd-risk-variant-clustering-app-25rr5g.streamlitapp.com/