Different p value in models_linear.rds and significant_results.tsv

Downstream analysis and statistics MaAsLin
1

Hi!

I run Maaslin3 on my data with the code below

otu_table1 <- ps %>% subset_samples(!is.na(Comparison1)) %>% otu_table() %>% as.data.frame()
metadata1 <- ps %>% subset_samples(!is.na(Comparison1)) %>% sample_data() %>% data.frame()
f1 <- paste0('Comparison1 + ', paste(covariates, collapse = '+'))

#================
#================
# Fit models
fit_out1 <- maaslin3(input_data = otu_table1,
                    input_metadata = metadata1,
                    output = '../../result/',
                    formula = f1,
                    normalization = 'TSS',
                    transform = 'LOG',
                    augment = TRUE,
                    standardize = TRUE,
                    max_significance = 0.1,
                    min_abundance = 0.0001,
                    min_prevalence = 0.1,
                    median_comparison_abundance = TRUE,
                    median_comparison_prevalence = FALSE,
                    max_pngs = 100,
                    cores = 10,
                    save_models = TRUE)

And it run without error. Though in the results folder, the p values I read from models_linear.rds (1.545704e-04) for the significant correlation between a species Dorea_longicatena and a variable is not identical with their p value in significant_results.tsv (p individual =1.316814e-05). But the coefficients are the same. Is it because they are calculated in different ways?

Best,

Ben

2

Hi Ben,

The p-value in the rds file is from a comparison of the coefficient against 0, whereas the p-value in the significant results is from a comparison of the coefficient against the median since you have median_comparison_abundance = TRUE set. Does that explain what you’re seeing?

Will

3

Yes, that’s very clear! Thank you!