Quantitative-Percentage
Description
The “quantitative-percentage” module takes phenotype/bioactivity data where samples have a measured percentage of bioactivity/phenotype at a specified concentration.
Multiple measurements can be specified for each sample, and multiple assays (each at a determined concentration) can be provided. The algorithm works as follows:
- Values below zero are set to zero.
- Duplicate measurements per sample are averaged using either the mean or median.
- The areas of molecular features detected in more than three samples are z-transformed along with the phenotype measurements.
- Transformed feature areas and percentage values are correlated using Pearson correlation.
- The resulting p-values are corrected for multiple hypothesis testing using a user-specified correction method (e.g. Bonferroni).
- Features that exceed user-defined thresholds for both correlation coefficient and adjusted p-value are considered phenotype-associated.
Limitations
- This method assumes that the prerequisites with regard to sample reproducibility are met (see Input/Output).
- This method assumes a positive linear correlation between phenotype (percentage) and concentration (area of feature).
- This method does not take into account any synergistic or quenching effects.
Parameters
Key | Possible Values | Default |
activate_module | true, false | false |
sample_avg | mean, median | mean |
value | area | area |
algorithm | pearson | pearson |
fdr_corr | bonferroni, sidak, holm-sidak, holm, simes-hochberg, hommel, fdr_bh, fdr_by, fdr_tsbh, fdr_tsbky | bonferroni |
p_val_cutoff | 0.0-1.0 | 0.05 |
coeff_cutoff | 0.0-1.0 | 0.7 |
Explanation
sample_avg
: specifies the algorithm to summarize multiple measurements per sample for same assay. Possible algorithms aremean
andmedian
.value
: specifies value per feature to be correlated with percentage. Onlyarea
is currently allowed.algorithm
: specifies the statistical algorithm to use. Onlypearson
is currently allowed.fdr_corr
: the method used for false-discovery-rate correction. FERMO uses the statsmodels library for this purpose - please see their documentation for information on the different algorithms.p_val_cutoff
: Maximum FDR-corrected p-value to consider, with zero disabling cutoff filtering for both p-value and coefficient.coeff_cutoff
: Minimum correlation coefficient to consider, with zero disabling cutoff filtering for both p-value and coefficient.