What it does

The re-sampling method is a comparative analysis the allows that can be used to determine conditional essentiality of genes. It is based on a permutation test, and is capable of determining read-counts that are significantly different across conditions.

This technique has yet to be formally published in the context of differential essentiality analysis. Briefly, the read-counts at each genes are determined for each replicate of each condition. The total read-counts in condition A is subtracted from the total read counts at condition B, to obtain an observed difference in read counts. The TA sites are then permuted for a given number of “samples”. For each one of these permutations, the difference is read-counts is determined. This forms a null distribution, from which a p-value is calculated for the original, observed difference in read-counts.

Note : Can be used for both Himar1 and Tn5 datasets

Inputs

Input files for Resampling need to be:

• .wig files : Tabulated files containing one column with the TA site coordinate and one column with the read count at this site.
• annotation .prot_table : Annotation file generated by the Convert Gff3 to prot_table for TRANSIT tool.

Parameters

Optional Arguments:

The resampling method is non-parametric, and therefore does not require any parameters governing the distributions or the model. The following parameters are available for the method:

• Samples: The number of samples (permutations) to perform. The larger the number of samples, the more resolution the p-values calculated will have, at the expense of longer computation time. The re-sampling method runs on 10,000 samples by default.
• Output Histograms:Determines whether to output .png images of the histograms obtained from resampling the difference in read-counts.
• Adaptive Resampling: An optional “adaptive” version of resampling which accelerates the calculation by terminating early for genes which are likely not significant. This dramatically speeds up the computation at the cost of less accurate estimates for those genes that terminate early (i.e. deemed not significant). This option is OFF by default.
• Include Zeros: Select to include sites that are zero. This is the preferred behavior, however, unselecting this (thus ignoring sites that) are zero accross all dataset (i.e. completely empty), is useful for decreasing running time (specially for large datasets like Tn5).
• Normalization Method: Determines which normalization method to use when comparing datasets. Proper normalization is important as it ensures that other sources of variability are not mistakenly treated as real differences. See the Normalization section for a description of normalization method available in TRANSIT. - TTR (Default) : Trimmed Total Reads (TTR), normalized by the total read-counts (like totreads), but trims top and bottom 5% of read-counts. This is the recommended normalization method for most cases as it has the beneffit of normalizing for difference in saturation in the context of resampling. - nzmean : Normalizes datasets to have the same mean over the non-zero sites. - totreads : Normalizes datasets by total read-counts, and scales them to have the same mean over all counts. - zinfnb : Fits a zero-inflated negative binomial model, and then divides read-counts by the mean. The zero-inflated negative binomial model will treat some empty sites as belonging to the “true” negative binomial distribution responsible for read-counts while treating the others as “essential” (and thus not influencing its parameters). - quantile : Normalizes datasets using the quantile normalization method described by Bolstad et al. (2003). In this normalization procedure, datasets are sorted, an empirical distribution is estimated as the mean across the sorted datasets at each site, and then the original (unsorted) datasets are assigned values from the empirical distribution based on their quantiles. - betageom : Normalizes the datasets to fit an “ideal” Geometric distribution with a variable probability parameter p. Specially useful for datasets that contain a large skew. See Beta-Geometric Correction . - nonorm : No normalization is performed.

Outputs

The re-sampling method outputs a tab-delimited file with results for each gene in the genome. P-values are adjusted for multiple comparisons using the Benjamini-Hochberg procedure (called “q-values” or “p-adj.”). A typical threshold for conditional essentiality on is q-value < 0.05.

More Information

See TRANSIT documentation

• TRANSIT: https://transit.readthedocs.io/en/latest/index.html
• TRANSIT Gumbel: https://transit.readthedocs.io/en/latest/transit_methods.html#re-sampling