Similarity measure for sparse time course data with Gaussian processes

Zijing Liu 2021-05-22

Introduction

This repository contains MATLAB functions for modelling time course data with Gaussian processes (GP) and computing a pair-wise similarity measure in the form of a Bayes factor. It uses the GPML toolbox (http://www.gaussianprocess.org/gpml/code/matlab/doc/).

• BF_onehyp.m - a function for computing the pair-wise similarity matrix, where the hyperparameters are optimised for the whole dataset.

• BF_twohyp.m - a function for computing the pair-wise similarity matrix, where the hyperparameters are optimised for each pair of time courses.

• BF_async.m - a function for computing the pair-wise similarity matrix, where the hyperparameters are optimised for the whole dataset and the time courses are asynchronous.

• script_synthetic_data.m - script to test on the synthetic data

• script_gene_data.m - script to cluster the gene expression data

• lib/ - the required Matlab packages including:

  • GPML toolbox for Gaussian process
  • Ncut and ZPclustering for spectral clustering
  • InfoTheory toolbox for NMI

• R/ - contains the R code to compute the BHI z-score.

• gene_data.mat - it is the Matlab data file containing the gene expression data.

References

Liu, Zijing, and Mauricio Barahona. "Similarity measure for sparse time course data based on Gaussian processes." arXiv preprint https://arxiv.org/abs/2102.12342 (2021). Accepted at 37th Conference on Uncertainty in Artificial Intelligence (UAI 2021), July 27-29, 2021. Link: https://proceedings.mlr.press/v161/liu21a.html