puma logo
PUMA: Propagating Uncertainty in Microarray Analysis

PUMA is  a research project based in the School of Computer Science at the University of Manchester in the UK.


9 July, 2009 - The puma software paper is published and is a featured article on the BMC Bioinformatics website.
13 June, 2007 - The TIGRA project, which builds on work done in PUMA, has been funded by the EPSRC.
27 April, 2007 - The puma package has been released through Bioconductor.


We are developing probabilistic models for the analysis of microarray data. These models allow us to include various sources of noise and uncertainty, be they biological or technical, within a unifying analysis framework. We are currently working on three main themes:
  1. We are developing methods to extract accurate measurements from the probe-level analysis of Affymetrix arrays. By taking a probabilistic perspective, we can associate these measurements with a level of technical measurement error while integrating out probe-specific effects.

  2. We are developing a suite of methods which use the technical measurement error from 1 and propagate it through higher level analyses. We have done this for a number of important applications, e.g. we have published methods for identifying differentially expressed genes from replicated and unreplicated experiments, for dimensionality reduction using principal component analysis, and for model-based clustering of genes.

  3. We are developing latent variable models for inferring the activity and effect of transcription factors from time-series gene expression data. We have developed discrete time state-space models for genome-wide inference, e.g. integrating gene expression data with ChIP-chip data, and we have developed continuous time Gaussian process models for small ODE systems of transcriptional regulation. This work is now being continued in the TIGRA project.
In all these cases we have developed practical methods for parameter estimation and probabilistic inference. Our methods are implemented in a number of freely available software packages.






Please note that the R packages mmgmos, pplr and pumaclust are no longer supported. All the functionality of these packages is now included in the puma package, which is still being actively supported and developed. The links for mmgmos, pplr and pumaclust can still be used to access historic versions of these packages.


This work is supported by a BBSRC award to N. Lawrence and M. Rattray "Improved Processing of microarray data using probabilistic models".

email: magnus.rattray@manchester.ac.uk, neill@cs.man.ac.uk