WeightedMeanCompletion

Quick Start

# installation
library(devtools)
install_github('bkellman/WeightedMeanCompletion')
library(WeightedMeanCompletion)

# interpolation in mtcars
X = data.matrix(mtcars)
X[sample(prod(dim(X)),20)] = NA
# default correlation based interpolation for rows (i) and columns (j)
X_out1 = interp_weightedMean(X,t=5)
# default correlation-based interpolation for columns (j), integer distance informed interpolation with exponential spreading function in rows (i)
X_out2 = interp_weightedMean(X,t=10,D_i=dist_integer(X),sim_func_i=sim_exp_func)
# integer distance informed interpolation with exponential spreading function in rows (i) and columns (j)
X_ou32 = interp_weightedMean(X,t=10,
  D_i=dist_integer(X),sim_func_i=sim_exp_func,
  D_j=dist_integer(X),sim_func_j=sim_exp_func)

For more information on parameter optimization, see the notes at the end of each function in R/benchmarking.R

Curated Phenotype and Phylogenic Database Summary

• Both databases match in species represented with a total of 742 bacteria representing the type strain in the species (50% match between databases).
• The Phenotype database has a total of 156 metadata factors flattened from the Bacdive DSMZ api.
• The Phylogenic database is based on cophenetic distance between the 742 species.

Phenotype data

• Curated phenotype information was collected from the Bacdive DSMZ api for type strain species.

Phylogeny data

• A cophenetic distance matrix was curated based on the available species from Phylophlan
• The benefit of matching species to Phylophlan is two fold. First, the Phylophlan included taxonomy represent the best classified species in each genus. While mainting a good phylogenic spread of all bacterial groups. Second, this method allows for "new" bacterial species to be accurately inserted into the distance matrix based on the Phylophlan genome insertion method.

Benchmarking

Benchmarking code

• base_impute: simple imputation methods such as mean,median,common fill, and zeros
• error_impute: can calculate RMSE,forbenious error and a function for measuring the explained varance of dataset through PCA
• benchmarking_methods: method that takes a dataset and runs 8 different methods for imputation for benchmarking ** TODO: (must add our method to the methods benchmarked)

Benchmarking Dataset

• TODO: find Common imputation benchmarking dataset
• Benchmarking performed on our dataset

TODO

Coding Tasks

• perform standard matrix-completion/interpolation: EM, regulariztion, nuclear norm...
• machine learning approach to matrix-completion/interpolation
• Weighted Mean Completion approach to matrix-completion/interpolation

Paper

https://www.overleaf.com/6466536zjchwf#/21826208/