In generable/rgeco: R package for accessing the Generable API

NOT_CRAN <- !is.null(Sys.getenv("GECO_API_PASSWORD")) && Sys.getenv("GECO_API_PASSWORD") != ''
knitr::opts_chunk$set(
  purl = NOT_CRAN,
  collapse = TRUE,
  comment = "#>",
  fig.height = 6,
  fig.width = 6
)

library(ggplot2)
library(tidyverse)
library(rgeco)

First we will login to the Generable API with our username and password (here, stored in environment variables).

login(user = Sys.getenv('GECO_API_USER'), password = Sys.getenv('GECO_API_PASSWORD'))
demo_project <- Sys.getenv('GECO_API_TEST_PROJECT')

Loading data into memory

For this analysis we will work with survival data. The information we need is stored in the subjects table.

subjects <- fetch_subjects(project = demo_project, event_type = 'overall_survival')

This command returns a data.frame of records, one per subject

glimpse(subjects)

Notice here that we have loaded subjects + events data in one step. For endpoints that are non-terminal or that have more than one event per subject, we would want to load those data separately by calling fetch_subjects(..., event_type = NULL) and fetch_events(...). For this project, it is reasonable to load data in one step.

This project contains synthetic data for 5 trials.

subjects %>%
  dplyr::filter(is.na(parent_trial_id)) %>%
  distinct(trial_name)

Note that, in this project, we have a number of so-called "validation" trials used for testing in the context of cross-validation. For now we are going to ignore these.

subjects <- subjects %>%
  dplyr::filter(is.na(parent_trial_id))

Compute Kaplan-Meier survival estimates

Next we will compute the Kaplan-Meier survival estimates for our 5 studies.

km_data <- prep_km_data(data = subjects,
                        formula = survival::Surv(time = overall_survival_event_trial_day,
                                                 event = overall_survival_event_flag == 'occurred') ~
                          trial_name + trial_arm_name)

This function returns data in a structure that is designed to make it easy to prepare survival plots using standard ggplot2 or other libraries, and eventually to overlay those with posterior inferences from our Inference API.

glimpse(km_data)

For example, we can easily plot K-M estimates using ggplot2:

ggplot(km_data, aes(x = time, y = estimate, colour = trial_arm_name)) + 
  facet_wrap(~ trial_name, ncol = 2) + 
  geom_step() +
  theme_minimal() + 
  ggtitle('Kaplan-Meier estimates for overall survival') +
  scale_y_continuous(labels = scales::percent) +
  guides(colour = guide_legend('Trial arm'))