poeli/disside repository overview

⁠DISSIDE: Dynamic In Silico Sample Identification for Discrete Evaluation

DISSIDE uses novel unsupervised learning to select samples that best represent the strongest a priori discrete pattern in a given data set. It further removes major outliers and "noisy" samples that do not fit discrete patterns well or represent outliers in groups below a defined n value. It then estimates the fit and strength of the a priori discrete pattern using both unconstrained and constrained methods for raw and cleaned data.

⁠Software Overview

DISSIDE accepts raw sample x feature data, distance matrices, or adjacency matrices as input. DISSIDE is best run with default parameters and requires no user input beyond the input data but can be heavily parameterized by the user if desired. DISSIDE first identifies samples with no reasonably related samples and removes those samples from the dataset. DISSIDE then confirms that all data is coherent for relational analysis and comparison. If samples exist without coherence for comparison, DISSIDE will automatically partition and output the data into coherent data sets for the user. On coherent data, DISSIDE uses hierarchical clustering to build a tree of sample relationships. DISSIDE forces a range of k groupings onto the tree. DISSIDE calculates a unique DISSIDE score for each k using PERMANOVA and ANOSIM tests without allowing for overfitting by using a unique algorithm including weighting that allows DISSIDE to know when to stop evaluating k's. The highest DISSIDE score is selected as optimal k. Constrained, unconstrained, and MANOVA-type visualization and analysis is performed on the data at optimal k. Groups at optimal k with fewer than n samples are cleaned from the data as pattern outliers and constrained, unconstrained, and MANOVA-type visualization and analysis is performed on the cleaned data.

⁠USAGE

The following command will display the help page.

docker run poeli/disside -h

Here is an example for running DISSIDE using the test dataset 9010_otu_no_noise.csv. The user needs to mount the host path(s) to the container to provide input and output directories.

docker run -v $PWD:/data poeli/disside -i /data/TEST_DATA/Strong_pattern_PA/9010_otu_no_noise.csv -o /data/9010_otu_no_noise

⁠Repository

Please visit our Gitlab repository⁠ for more information.

⁠Contact

Erick LeBrun: [email protected]⁠