Skip to content

Qualitative-Concentration

Description

The “quantitative-concentration” module takes phenotype/bioactivity data, with the minimum inhibitory concentration per sample specified (concentration/dilution at which a phenotypic signal was still observed).

Samples inactive at any concentration are specified with 0. Multiple measurements can be specified for each sample, and multiple assays can be provided. The algorithm works as follows:

  • Duplicate measurements per sample are averaged using either the mean or median.
  • Only molecular features detected in more than three samples are retained for correlation testing.
  • MIC values are converted to their reciprocals (1 / measurement) or left as zero if the concentration was zero.
  • Feature areas and MIC values are correlated using one of the correlation methods (e.g. Pearson).
  • 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 classified as phenotype-associated.

Limitations

  • This method assumes that the prerequisites with regard to sample reproducibility are met (see Input/Output).
  • This method assumes a negative relationship between phenotype (MIC) and concentration (area of feature) - the lower the minimal inhibitory concentration, the higher the concentration.
  • 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, spearman, spearman_permutation 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 are mean and median.
  • value: specifies value per feature to be correlated with concentration. Only area is currently allowed.
  • algorithm: specifies the statistical algorithm to use. spearman_permutation includes a two-sided permutation test (scipy.stats.permutation_test) for accurate p-values for small sample sizes.
  • 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.