Package for common analyses in FinnGen.

Overview

The fganalysis is an R package designed for common analyses performed in FinnGen. First functionality provides functions for data processing, summarization, and visualization of lab measurements and drug purchases.

Installation

You can install the package from the local directory using the following command after installing devtools package:

Need to make precompiled packages of everything for sandbox.

Some packages might get installed from source and to speedup that, can add multithreaded compilation.... add environment variable to enable 4 threads. 
Sys.setenv(MAKEFLAGS = "-j 4")

## if installing from source, install devtools and
devtools::install("path/to/fganalysis")

## in sandbox, you can just


library(devtools)
load_all("/finngen/shared_nfs/finngen/code/fganalysis/")
conn <- connect_fgdata("/finngen/shared_nfs/finngen/code/fganalysis/config/db_config_sb.json")

Usage

Functions

The package includes several key functions:

• create_drug_response(): Generates a drug response dataset based on lab measurements and drug purchases.
• summarize_drug_response(): Creates a summary PDF and text tables of drug response data.
• get_lab_measurements and get_drug_purchases to query for lab values and purchases.

Examples

Here is a simple example of how to use the package:

# Load the package

load_all("/finngen/shared_nfs/finngen/code/fganalysis/")
## get connection to data sources. in sanbox you can find data source configuration in /finngen/shared_nfs/finngen/code/drugResponsePackage/config/db_config_sb.json
conn <- connect_fgdata("config/db_config.json")
### SANDBOX
conn <- connect_fgdata("/finngen/shared_nfs/finngen/code/drugResponsePackage/config/db_config_sb.json")



##returned object has attributes that are lazy loaded data frames of different phenotype data.
## you can start writing dplyr queries and e.g. joining to other tables. Nothing will happen before you actually request the data to be localized.
## behind the scenes, a query engine optimizes the query and returns only the data matching your query....

## query for individuals with ICD-10 code K51 (IBD)
ibd <- conn$pheno %>% filter( (SOURCE=="INPAT"|SOURCE=="OUTPAT") & CODE1=="K51" & ICDVER=="10") %>% group_by(FINNGENID) %>% summarize(n_diagnoses=n())
## look at the number of rows..

>nrow(ibd)
NA
### you get NA because nothing has been queried before you ask for the data. use function collect to execute the query and return results
ibd <- ibd %>% collect()
nrow(ibd)
258

## get all labs with omopid 3007461
labs <- get_lab_measurements(conn$labs, c("3007461"))

## get all drug purchases with ATC codes starting with L01B
dr <- get_drug_purchases(conn, c("L01B"))


# Create drug response data of lab changes after initiating a drug.
## first define time intervals from drug purchase to summarise lab values
## here defining pre-measurements drug measurements to be 1 year before drug and 
## after period to be 1month to 1 year.
before_period <- c(-1, 0)
after_period <- c(1/12, 1)

## create a dataframe containing LDL (omopid 3001308) response to first statin purchase (ATC codes starting with C10AA) for each finngen ID  
resp <- create_drug_response(conn,c("3001308"), 
                             druglist=c("C10AA"),before_period,after_period)
## create plots and tables of the respons
summarize_drug_response(resp, out_file_prefix="3001308_A10_resp")

Development & Data storage

Install devtools package. When in root folder of package you can load everything with devtools::load_all(). Read more about package dev with devtools here https://cran.r-project.org/web/packages/devtools/readme/README.html and https://r-pkgs.org/

Database connection is defined in config/db_config.json. Currently data is stored in parquet files and queried via duckdb. This way there are no external dependencies on databases.
If new ways to access data are introduced, add handling of such datatypes in R/connections.R connect_fgdata. Returned objects should be lazy loaded dplyr::tbl objects so further processing can be done via dbplyr (https://dbplyr.tidyverse.org/)

Testing

The package includes unit tests to ensure the functionality of its core functions. You can run the tests using:

devtools::test()

When adding new functionality add unit tests. See tests/testthat/test-drug_response_functions.R for examples.

Author

[Mitja Kurki]

License

This package is licensed under the MIT License.