Amanida meta-analysis: a tutorial
Here we show how to conduct a meta-analysis using Amanida approach. We are using a dataset obtained in systematic review and meta-analysis: Comprehensive Volatilome and Metabolome Signatures of Colorectal Cancer in Urine: A Systematic Review and Meta-Analysis Mallafré et al. Cancers 2021, 13(11), 2534; https://doi.org/10.3390/cancers13112534.
Before using Amanida is important the curation of the identifiers. The objective of a meta-analysis is to combine the different results of a same feature, therefore, it is important that all identifiers are in the same format.
Import data
Dataset to analyse must include the following columns: identifier, p-value, fold-change, study size (N) and reference.
Once we have the dataset in a supported format (xls/xlsx, csv or txt), we import as follows:
coln = c("Compound Name", "P-value", "Fold-change", "N total", "References")
input_file <- getsampleDB()
datafile <- amanida_read(input_file, mode = "quan", coln, separator=";")
#> Loaded dataset with 143 rows which contains 120 different identifiers. There are 18 rows skipped from original dataset because contained NA values.It creates a tibble which contain the data needed to perform the amanida meta-analysis. Note: missing data is ignored and negative values of fold-change are transformed to positive (1/value).
Analysis
In this step will be performed the combination of overall results weightening by study size. Methods for combine results are:
P-value: weighted p-values combination, which is a variant of Fisher’s method. A gamma distribution is used to assign non-integral weights proportional to study size to each p-value.
Fold-change: logarithmic transformation (base 2) and then averaged with weighting by study size.
Note: Only with the object created by amanida_read function in mode = "quan" are able to compute the amanida meta-analysis.
It is also included a qualitative analysis, the vote-counting, that is computed by the sum of votes assigned as follows: votes are +1 for up-regulation, -1 for down-regulation and 0 if no trend.
amanida_result <- compute_amanida(datafile)Note: The output include compound identifiers from public repositories such as PubChem or the Human Metabolite DataBase, if compounds are not used this could be avoided with comp.inf = F.
Results and visualization
Combination results are disclosed in two tables
- meta-analysis results acces by
amanida_result@stat - vote-counting results acces by
amanida_results@vote
Quantitative results
Amanida meta-analysis gives the combination for p-value and fold-change for each feature.
Trend variable correspond to the direrction of the combination, 1 means that the feature is increased in case group and -1 the opposite. N_total is the sum of the study sizes for that feature, which are disclosed in reference variable.
head(amanida_result@stat)
#> # A tibble: 6 × 6
#> id trend pval fc N_total reference
#> <chr> <dbl> <dbl> <dbl> <int> <chr>
#> 1 1,1,6-Trimethyl-1,2-dihydronaphthalene -1 0.000391 0.224 60 [29] Por…
#> 2 1,3-Dithiane 1 0.0189 3.43 60 [29] Por…
#> 3 1,5,5-Trimethyl-6-methylene-cyclohexene -1 0.0342 0.596 60 [29] Por…
#> 4 1,6,7-Trimethylnaphthalene -1 0.000194 0.320 60 [29] Por…
#> 5 2,2,6-Trimethylcyclohexanone -1 0.0298 0.348 60 [29] Por…
#> 6 2,3-Butanediol -1 0.000091 0.271 135 [43] Lie…To observe the results of meta-analysis graphically is done with a volcano plot, where the log10(p-values) are plotted against the log2(fold-change). The cut-offs can be selected by the user, in case of fold-change we recommend values higher than 2, where it is considered to have biological meaningfulness.
Note: cut-off have to be indicated without log scale.
volcano_plot(amanida_result, cutoff = c(0.05,3.5))
#> The cut-off used are 0.05 for p-value (1.3 in log10 scale) and 3.5 for fold-change (1.81 in log2 scale).
We will consider that a feature have statistical significance if it is over the cut-offs threshold (dashed lines).
Qualitative results
Optionally we van observe the qualitative results. A bar plot shows the result of vote-counting.
Note: output is restricted to 30 compounds to facilitate the readability. To visualize less compounds increase counts parameter.
vote_plot(amanida_result, counts = 1)
#> Cut-off for votes is 1.
#> Too much values, only showing 30 highest values. Please check counts parameter.
Note: The output include compound identifiers from public repositories such as PubChem or the Human Metabolite DataBase, if compounds are not used this could be avoided with comp.inf = F.
With vote plot discrepancies in compounds behaviour are not detected at first glance, and we suggest to combine the results with the explore plot. Discrepancies in results are shown when type = "mix" is used:
- counts parameter: this value indicate the minimum votes to show the results.
Note: output is restricted to 25 compounds to facilitate the readability. To visualize less compounds increase counts parameter.
explore_plot(sample_data, type = "mix", counts = 1)
#> Cut-off for votes is 1.
In sample data we observe some compounds over the cut-off threshold, then we check the consistency of result. In this case, compounds with more than one report and statistical significance are Hippuric acid and Phenol. Both of them have consistency as all reports results are in the same trend. We can conclude that for Hippuric acid and Phenol have evidences to reject the null hypothesis.
Note: compounds with both up-regulated and down-regulated tendencies points to low consistency.