README.md

In shihuang047/crossRanger: Cross-Application of Ranger Models using microbiome data

crossRanger

This package provide functionalities to perform random forests (RF) meta-analysis of microbiome data

Introduction

crossRanger: this package enables random forests classification or regression on a single target variable for the multiple datasets and allows users to apply the prebuilt random forests model from one dataset to predict the target variable in other datasets.

ranger is a fast implementation of random forests (Breiman 2001) or recursive partitioning, particularly suited for high dimensional data. Classification, regression, and survival forests are supported. To meet the meta-analysis requirements we further use ranger as a core RF implementation function to develop this package for microbiome-centric analyses.

This R package basically provides a variety of functions for the RF analyses within a single or across multiple microbiome datasets. Specifically, this package allows RF classification or regression on a single target variable in a single microbiome dataset with CV approaches RF classification or regression on a single target variable (such as diseased status or age) stratified by another covariate (such as sex or body sites) for a single microbiome dataset RF classification or regression on a single covariate for multiple microbiome datasets (such as microbiome studies focused on the same phenotype) The prediction performance comparisons across multiple RF models (including AUROC, AUPRC, and accuracy etc. for classification models; MAE, MSE, RMSE, MAPE, R squared etc. for regression models) the application of multiple RF models to each other and output the prediction performance (such as application of an RF regression model on aging on the female to male cohort and vice versa) to output the univariate associations in the microbiome with a single covariate (e.g Wilcoxon rank-sum test on the clr-transformed relative abundance of each microbial feature in microbiome data) * to visualize the results above and produce publication-quality figures

Authors

Shi Huang, UC San Diego

Installation

The development version is maintained on GitHub and can be downloaded as follows:

## install.packages('devtools') # if devtools not installed
devtools::install_github('shihuang047/crossRanger')

Examples

• Compositional microbiome data simulation using multinomial distribution

set.seed(123)
df <- data.frame(rbind(t(rmultinom(7, 75, c(.201,.5,.02,.18,.099))),
            t(rmultinom(8, 75, c(.201,.4,.12,.18,.099))),
            t(rmultinom(15, 75, c(.011,.3,.22,.18,.289))),
            t(rmultinom(15, 75, c(.091,.2,.32,.18,.209))),
            t(rmultinom(15, 75, c(.001,.1,.42,.18,.299)))))
metadata<-data.frame(f_s=factor(c(rep("A", 15), rep("B", 15), rep("C", 15), rep("D", 15))),
                     f_c=factor(c(rep("C", 7), rep("H", 8), rep("C", 7), rep("H", 8),
                                  rep("C", 7), rep("H", 8), rep("C", 7), rep("H", 8))),
                     f_d=factor(rep(c(rep("a", 5), rep("b", 5), rep("c", 5)), 4)))

• RF modeling on one responsive variable

rf.out.of.bag(x=df, y=metadata$f_s, imp_pvalues=FALSE)

• RF modeling on one responsive variable stratified by another covariate

res_list<-rf_clf.by_datasets(df, metadata, s_category='f_s', c_category='f_c', positive_class="C")
summary(res_list)

• Apply RF models to other datasets

cross_rf<-rf_clf.cross_appl(rf_model_list=res_list$rf_model_list, 
                            x_list=res_list$x_list, 
                            y_list=res_list$y_list, positive_class="C")
cross_rf$perf_summ
# the output
Train_data Test_data   Validation_type  Accuracy       AUC       Kappa Sensitivity Specificity Pos_Pred_Value Neg_Pred_Value
1           A         A   self_validation 0.9333333 1.0000000  0.86725664   1.0000000       0.875      0.8750000      1.0000000
2           A         B cross_application 0.4666667 0.3303571  0.00000000   1.0000000       0.000      0.4666667            NaN
3           A         C cross_application 0.4666667 0.6964286  0.00000000   1.0000000       0.000      0.4666667            NaN
4           A         D cross_application 0.4666667 0.4285714  0.00000000   1.0000000       0.000      0.4666667            NaN
5           B         B   self_validation 0.5333333 0.4464286  0.05405405   0.4285714       0.625      0.5000000      0.5555556
6           B         A cross_application 0.4666667 0.4017857  0.00000000   1.0000000       0.000      0.4666667            NaN
7           B         C cross_application 0.4666667 0.3482143  0.00000000   1.0000000       0.000      0.4666667            NaN
8           B         D cross_application 0.4666667 0.4553571 -0.03448276   0.7142857       0.250      0.4545455      0.5000000
9           C         C   self_validation 0.5333333 0.3571429  0.05405405   0.4285714       0.625      0.5000000      0.5555556
10          C         A cross_application 0.4666667 0.4642857 -0.11111111   0.1428571       0.750      0.3333333      0.5000000
11          C         B cross_application 0.4666667 0.6071429  0.00000000   1.0000000       0.000      0.4666667            NaN
12          C         D cross_application 0.4000000 0.5267857 -0.13445378   0.8571429       0.000      0.4285714      0.0000000
13          D         D   self_validation 0.4000000 0.3214286 -0.21621622   0.2857143       0.500      0.3333333      0.4444444
14          D         A cross_application 0.5333333 0.4464286  0.10256410   0.8571429       0.250      0.5000000      0.6666667
15          D         B cross_application 0.6000000 0.5267857  0.19642857   0.5714286       0.625      0.5714286      0.6250000
16          D         C cross_application 0.4000000 0.3482143 -0.26168224   0.0000000       0.750      0.0000000      0.4615385
   Precision    Recall        F1 Prevalence Detection_Rate Detection_Prevalence Balanced_Accuracy
1  0.8750000 1.0000000 0.9333333  0.4666667     0.46666667            0.5333333         0.9375000
2  0.4666667 1.0000000 0.6363636  0.4666667     0.46666667            1.0000000         0.5000000
3  0.4666667 1.0000000 0.6363636  0.4666667     0.46666667            1.0000000         0.5000000
4  0.4666667 1.0000000 0.6363636  0.4666667     0.46666667            1.0000000         0.5000000
5  0.5000000 0.4285714 0.4615385  0.4666667     0.20000000            0.4000000         0.5267857
6  0.4666667 1.0000000 0.6363636  0.4666667     0.46666667            1.0000000         0.5000000
7  0.4666667 1.0000000 0.6363636  0.4666667     0.46666667            1.0000000         0.5000000
8  0.4545455 0.7142857 0.5555556  0.4666667     0.33333333            0.7333333         0.4821429
9  0.5000000 0.4285714 0.4615385  0.4666667     0.20000000            0.4000000         0.5267857
10 0.3333333 0.1428571 0.2000000  0.4666667     0.06666667            0.2000000         0.4464286
11 0.4666667 1.0000000 0.6363636  0.4666667     0.46666667            1.0000000         0.5000000
12 0.4285714 0.8571429 0.5714286  0.4666667     0.40000000            0.9333333         0.4285714
13 0.3333333 0.2857143 0.3076923  0.4666667     0.13333333            0.4000000         0.3928571
14 0.5000000 0.8571429 0.6315789  0.4666667     0.40000000            0.8000000         0.5535714
15 0.5714286 0.5714286 0.5714286  0.4666667     0.26666667            0.4666667         0.5982143
16 0.0000000 0.0000000       NaN  0.4666667     0.00000000            0.1333333         0.3750000

References

• Breiman, L. (2001). Random forests. Mach Learn, 45:5-32. https://doi.org/10.1023/A:1010933404324.
• Wright, M. N. & Ziegler, A. (2017). ranger: A fast implementation of random forests for high dimensional data in C++ and R. J Stat Softw 77:1-17. https://doi.org/10.18637/jss.v077.i01.

License

All source code freely availale under GPL-3 License.

Documentation

Each exported function in the package has been documented and we have also written an introductory vignette that is accessible by calling

vignette('crossRanger--intro', package='crossRanger')

Bugs/Feature requests

I appreciate bug reports and feature requests. Please post to the github issue tracker here.

Acknowledgements

This work is supported by IBM Research AI through the AI Horizons Network. For more information visit the IBM AI Horizons Network website.

shihuang047/crossRanger documentation built on Feb. 7, 2023, 10:03 p.m.