Error in pathway_errorbar() ( Some features have less than 3 nonzero values! )

[Valentin-Bio]

Hello developer! thanks so much for this tool. I'm trying to run the full workflow with ggpicrust2() function. The problem is I got the following error message when running this code:

results_file_input <- ggpicrust2(file = "./functional_analysis/KO_paths_mod.tsv",
                                 metadata = metadata_f,
                                 group = "treatment", # For example dataset, group = "Environment"
                                 pathway = "KO",
                                 daa_method = "LinDA",
                                 ko_to_kegg = TRUE,
                                 order = "pathway_class",
                                 p_values_bar = TRUE,
                                 x_lab = "pathway_name")

Error in pathway_errorbar(abundance = abundance, daa_results_df = daa_sub_method_results_df, :
The 'group1' or 'group2' column in the 'daa_results_df' data frame contains more than one group. Please filter each to contain only one group.
In addition: Warning message:
In MicrobiomeStat::linda(feature.dat = feature.dat, meta.dat = meta.dat, :
Some features have less than 3 nonzero values!
They have virtually no statistical power. You may consider filtering them in the analysis!

Im thinking this error may be due my experimental design of my samples. I only have 8 samples, with 3 groups (control with 3 samples, another control of 2 samples and 2 treatment samples and another group of 1 sample). I thought the error may be due to the group of one sample so I decided to remove that group from my data (In the picrust2 output table and in the metadata table) But I still have the same problem. Any ideas?

Thanks!

Valentín.

[cafferychen777]

Hello @Valentin-Bio,

Thank you for using ggpicrust2! I understand the issue you're encountering. The error message "The 'group1' or 'group2' column in the 'daa_results_df' data frame contains more than one group" occurs because when you have more than 2 groups in your experimental design, the LinDA method performs multiple pairwise comparisons, which creates multiple group1/group2 combinations in the results dataframe.

Root Cause

When LinDA analyzes data with 3+ groups, it performs pairwise comparisons between each group and a reference group. This results in a dataframe where:

• group1 contains the reference group name
• group2 contains different comparison group names

The pathway_errorbar() function expects only one unique comparison (i.e., only one pair of groups), but receives multiple comparisons, hence the error.

Solutions

Option 1: Specify a Reference Group (Recommended)

When you have more than 2 groups, you should specify a reference parameter in your ggpicrust2() call:

results_file_input <- ggpicrust2(file = "./functional_analysis/KO_paths_mod.tsv",
                                 metadata = metadata_f,
                                 group = "treatment",
                                 pathway = "KO",
                                 daa_method = "LinDA",
                                 ko_to_kegg = TRUE,
                                 order = "pathway_class",
                                 p_values_bar = TRUE,
                                 x_lab = "pathway_name",
                                 reference = "control")  # Add this line

Replace "control" with the exact name of your control group as it appears in your metadata.

Option 2: Filter to Two Groups Only

If you want to compare only two specific groups, filter your metadata and abundance data to include only those two groups before running the analysis:

# Filter metadata to include only two groups
metadata_filtered <- metadata_f[metadata_f$treatment %in% c("group1", "group2"), ]

# Filter abundance data to match
# (assuming your abundance file has sample names as column headers)
samples_to_keep <- metadata_filtered$sample_id  # adjust column name as needed
# Then filter your abundance file accordingly

results_file_input <- ggpicrust2(file = "./filtered_abundance.tsv",
                                 metadata = metadata_filtered,
                                 group = "treatment",
                                 pathway = "KO",
                                 daa_method = "LinDA",
                                 ko_to_kegg = TRUE,
                                 order = "pathway_class",
                                 p_values_bar = TRUE,
                                 x_lab = "pathway_name")

Option 3: Use a Different DAA Method

Some methods like ALDEx2 handle multiple groups differently. You could try:

results_file_input <- ggpicrust2(file = "./functional_analysis/KO_paths_mod.tsv",
                                 metadata = metadata_f,
                                 group = "treatment",
                                 pathway = "KO",
                                 daa_method = "ALDEx2",  # Changed from LinDA
                                 ko_to_kegg = TRUE,
                                 order = "pathway_class",
                                 p_values_bar = TRUE,
                                 x_lab = "pathway_name")

About the Warning Message

The warning "Some features have less than 3 nonzero values!" is from the LinDA method itself and indicates that some pathways have very low abundance across samples. This is common with small sample sizes and doesn't necessarily indicate an error, but suggests those features have limited statistical power.

Recommendation

Given your experimental design (8 samples across multiple groups), I recommend Option 1 - specifying a reference group. This will allow LinDA to perform proper pairwise comparisons against your control group.

Please let me know if you need any clarification or if you encounter any other issues!

Best regards,
Chen Yang