How to calculate hypergeometric test for each row in data frame with Bonferroni correction

Question

I am calculating, hopefully right, the hypergeometric test per row in a data frame in R.

Where column 1 is names of genes (microRNAs), the column "Total_mRNAs" is how many mRNA exist in total in the genome so that doesn't change. Column "Total_targets_targets" is how many mRNAs each microRNA can target if all the mRNAs are present. However, for this example only "subset_mRNAs" are present (that number is also always the same) and among these I know how many mRNAs each microRNA can target "subset_targets".

In order to determine if targets for each microRNA are enriched compared to the background (total mRNAs and total microRNAs targeting them) I am performing a hypergeometric test per row like this:

phyper(targets-in-subset, targets-in-bkgd, failure-in-bkgd, sample-size-subset, lower.tail= FALSE)



dput(df1)
structure(list(Genes_names = c("microRNA-1", "microRNA-2", "microRNA-3", 
"microRNA-4", "microRNA-5", "microRNA-6", "microRNA-7", "microRNA-8", 
"microRNA-9", "microRNA-10"), Total_mRNAs = c(61064L, 61064L, 
61064L, 61064L, 61064L, 61064L, 61064L, 61064L, 61064L, 61064L
), Total_targets_targets = c(1918L, 7807L, 3969L, 771L, 2850L, 
1355L, 1560L, 2478L, 1560L, 2478L), subset_mRNAs = c(17571L, 
17571L, 17571L, 17571L, 17571L, 17571L, 17571L, 17571L, 17571L, 
17571L), subset_targets = c(544L, 2109L, 1137L, 213L, 793L, 394L, 
430L, 686L, 430L, 686L)), class = "data.frame", row.names = c(NA, 
-10L))


df1$pvalue <- phyper(df1$subset_targets, df1$Total_targets_targets, df1$Total_mRNAs-df1$Total_targets_targets, df1$subset_mRNAs, lower.tail= FALSE)

Now the question is how can I Bonferroni correct this values? Is this calculation theoretically right?

You can use the apply function for this. Please give more details on what you want. From your code listenings I do not know how you want to calculate what exactly. — MacOS
Sorry I corrected the question lots. But ill try and explain it a bit better. — Amaranta_Remedios
I don't understand where this df1$Total-df1$targets comes from. — MacOS
df1$Total = all the mRNAs present in the genome of that specie 61064 df1$targets = all the targets each microRNA can theoretically have if all the mRNAs are present 1918 for the first one. — Amaranta_Remedios

StupidWolf StupidWolf · Accepted Answer · 2020-11-23T15:23:25

If you have not a lot of samples, avoid this pain and simply use a fisher test and do bonferroni using p.adjust:

library(broom)
result = lapply(1:nrow(df1),function(i){
       not_target_subset = df1$Total_targets_targets[i] - df1$subset_targets[i]
       not_subset = df1$Total_mRNAs[i] - df1$subset_mRNAs[i] - not_target_subset
       
       
       M = cbind(c(df1$subset_targets[i],df1$subset_mRNAs[i]-df1$subset_targets[i]),
             c(not_target_subset,not_subset))
      
       res = data.frame(Genes_names=df1$Genes_names[i],
                tidy(fisher.test(M,alternative="greater")))

       return(res)
})

result= do.call(rbind,result)
result$padj = p.adjust(result$p.value,"bonferroni")

Your hyper-geometric code is slightly off. And note that you are doing one sided hyper-geometric test.

You can check this post for how to put the tables into phyper and this for why you need the -1 . So we calculate the hypergeometric p-value:

result$hyper_p = with(df1, 
phyper(subset_targets-1,subset_mRNAs,Total_mRNAs-subset_mRNAs, Total_targets_targets, lower.tail= FALSE)
)

And you can see it tallies:

   Genes_names  estimate   p.value  conf.low conf.high
1   microRNA-1 0.9793710 0.6655527 0.8984025       Inf
2   microRNA-2 0.9047305 0.9998968 0.8647701       Inf
3   microRNA-3 0.9933480 0.5791759 0.9350214       Inf
4   microRNA-4 0.9441864 0.7722712 0.8229140       Inf
5   microRNA-5 0.9520878 0.8789562 0.8863785       Inf
6   microRNA-6 1.0151760 0.4119600 0.9168998       Inf
7   microRNA-7 0.9404619 0.8641420 0.8539585       Inf
8   microRNA-8 0.9454359 0.8942082 0.8756678       Inf
9   microRNA-9 0.9404619 0.8641420 0.8539585       Inf
10 microRNA-10 0.9454359 0.8942082 0.8756678       Inf
                               method alternative   hyper_p
1  Fisher's Exact Test for Count Data     greater 0.6655527
2  Fisher's Exact Test for Count Data     greater 0.9998968
3  Fisher's Exact Test for Count Data     greater 0.5791759
4  Fisher's Exact Test for Count Data     greater 0.7722712
5  Fisher's Exact Test for Count Data     greater 0.8789562
6  Fisher's Exact Test for Count Data     greater 0.4119600
7  Fisher's Exact Test for Count Data     greater 0.8641420
8  Fisher's Exact Test for Count Data     greater 0.8942082
9  Fisher's Exact Test for Count Data     greater 0.8641420
10 Fisher's Exact Test for Count Data     greater 0.8942082

How to calculate hypergeometric test for each row in data frame with Bonferroni correction

2 Answers

EDIT