NOTE: lgpr2 has been released! It allows scaling to larger data as it uses approximate Gaussian processes.

Using the lgpr package

This repository contains code for experiments with the R package lgpr (https://github.com/jtimonen/lgpr).

Getting started

Tutorials in html format, package documentation and installation instructions can be found here.

Reproducibility

The reproducibility directory contains codes for reproducing the experiments in the manuscript using lgpr version 0.33, which can be installed via

devtools::install_github('jtimonen/lgpr', ref = "v0")

Experiments involve fitting models for different data sets in parallel, and shell scripts for doing this on a computing cluster are provided. There are also scripts for collecting the final results and plotting ROC curves after the models have been fit for all data sets.

There are seven different experiments:

• 01_lme4 - comparison with linear mixed effect modeling and scalability testing
• 02_longp - comparison with LonGP (https://www.nature.com/articles/s41467-019-09785-8)
• 03_heter - heterogeneous disease effect modeling
• 04_effect_time - modeling uncertainty of disease effect time
• 05_nb - experiment with negative binomial distributed count data
• 06_proteomics - longitudinal proteomics data set [1] modeling with a homogeneous disease effect
• 07_proteomics_heter - longitudinal proteomics data set [1] modeling with a heterogeneous disease effect
• 08_rna-seq - analysis of longitudinal RNA-seq data from CD4+ cells [2]

Data

The data directory contains the preprocessed longitudinal proteomics data [1].

References

[1] Liu et al. Temporal expression profiling of plasma proteins reveals oxidative stress in early stages of Type 1 Diabetes progression (2018). Journal of proteomics 172: 100-110. doi:10.1016/j.jprot.2017.10.004

[2] Kallionpää et al. Early Detection of Peripheral Blood Cell Signature in Children Developing β-Cell Autoimmunity at a Young Age (2019). Diabetes 68(10):2024-2034. doi:10.2337/db19-0287