70 Peptide antibody-binding enrichment Antibody-binding against peptide (seropositivity) was defined as described in.21 In brief, for each sample, null distributions per input level (number of reads per clone without IP) are generated. A two-parameter generalized Poisson model is fit to the null distribution, and the P-value to obtain the coverage level after IP for a given clone is estimated. Model parameters were estimated for each null distribution using maximum likelihood or directly interpolated as described in.11 A strict Bonferroni cut-off at PBonferroni < 0.05 was then used to define seropositivity. A total of 175,242 peptides were seropositive in at least one participant. Antibody-bound peptides and exploratory analysis Data analysis was performed in R v4.0.3 using the packages tidyverse, stats, vegan,30 corrplot, igraph,31 WGCNA,32 readxl, pheatmap, cairo and patchwork. Antibody-bound peptide selection Peptides to be used in the analysis were selected based on two filters. We chose peptides that had a prevalence at least of 5% and below 95% in either 1000IBD or LLD (excluding follow-up samples). For antibody-bound peptides with identical sequence, we chose the most prevalent antibody-bound peptide, resulting in 2,815 selected antibody-bound peptides. Principal component analysis We used 2,815 peptides to compute a PCA. Eigenvalues were used to produce a scree plot and eigenvectors to identify top peptides contributing to the first components. A K-means algorithm (k = 2) was performed on the dimensionally reduced dataset (PC1 and 2) to label observed clusters. This analysis was reproduced after removal of the 90 peptides belonging to CMV. Time and family distance analysis 322 LLD samples belonging to two different time points were used for a time consistency analysis. Jaccard distance was used as the dissimilarity metric between samples. The P-value of longitudinal effect of mean distance was estimated by computing the P-value of the mean pairwise difference of longitudinal samples in a null distribution of mean distances of pairwise differences of 2,000 label swaps. Interrogation of factors that might affect the degree of change in longitudinal samples was performed using pairwise distances from longitudinal samples as dependent variable and age and sex as covariates in a linear model. Antibody-bound peptides consistency was computed by averaging the number of changes in the enrichment profile of a peptide among all samples with longitudinal data points. To check whether antibody-bound peptide enrichment changes seen in follow-up are due to a different reactivity of the plates used for baseline and follow-up samples, we ran a Wilcoxon test comparing the number of enriched antibody-bound peptide of participants profiles from plates with follow-up samples vs. plates with no follow-up samples. We then selected samples belonging to the same family (Genome of the Netherlands Consortium, 2014) with three members (26 families). We computed pairwise distances (Jaccard) between family members (father to offspring, mother to offspring and father to mother). For each of the comparisons, we estimated a P-value comparing the mean distance with a random distribution of means between 2,000 permuted labels. Chapter 3
RkJQdWJsaXNoZXIy MjY0ODMw