REFLECT (REcurrent Features LEveraged for Combination Therapy) is a bioinformatics tool to accelerate drug discovery and clinical translation that will result in real patient benefit. REFLECT identifies co-actionable, co-occurring oncogenic alterations that are recurrent within patient cohorts. REFLECT aims to enable researchers and clinicians to develop more effective targeted therapy combinations and better select benefiting patient cohorts in preclinical and clinical settings.

The REFLECT website (https://bioinformatics.mdanderson.org/reflect/) provides co-occurrence analyses of ~200 patient cohorts each identified by a potentially actionable aberration.

The reflect package provides a set of functions that runs the reflect pipeline:

• get_best_wbound() gets the optimal tunning parameter that maximizes gap statistic.
• sparse_hclust() performs a sparse hierarchical clustering.
• get_recur_pval() gets recurrence p values based on clustered matrix.
• get_recur_actionable_features() gets recurrent and actionable features.
• get_coaltered_targets() gets combination targets based on stratification biomarker and REFLECT-selected features.
• reflect_pipeline() runs an end-to-end REFLECT pipeline.
• reflect_pipeline2() runs a REFLECT pipeline given a precomputed tunning parameter.

The reflect package also provides a set of functions that plots the results:

• plot_bar_tumortypes() plots a bar chart of tumor type distribution.
• plot_bar_tumortypes_stack_tcga_ccl() plots a stacked bar chart of tumor types for TCGA and cell line samples.
• plot_gapstat() plots a gap statistic profile.
• plot_bar_weights() plots a bar chart of feature weights.
• plot_heatmap() plot a heatmap with covariate bars of feature weights and sample types.

You can learn more about them in vignette("reflect").

To use the latest development version, you can install reflect from GitHub:

# install.packages("devtools")
devtools::install_github("korkutlab/reflect")

library(reflect)

# selecting optimal tunning parameter
# this is computationally costly and may take take a long time 
# gapstat_bestwbound <- get_best_wbound(egfr_data$mat_value)
# wbound <- gapstat_bestwbound$best_wbound

# run a REFLECT pipeline given a precomputed tunning parameter
wbound <- 2.0
res <- reflect_pipeline2(wbound, 
                         egfr_data$mat_value, 
                         egfr_data$df_sample, 
                         egfr_data$df_feature)
#> 123456

Li X., et al. (2020) Precision combination therapies from recurrent oncogenic co-alterations. doi: https://doi.org/10.1101/2020.06.03.132514