CADET

RAIN: automated image alignment for 2-D gel electrophoresis

The RAIN 2-D gel alignment algorithm improves on the existing image registration approach of MIR with:

  1. A third order B-spline transformation that provides a realistic smooth deformation model of the 2-DE process.
  2. Volume-invariant warping through explicit modelling of expression density, which ensures protein abundance levels remain invariant under spot deformation.
  3. Relative bias-field correction of expression inhomogeneities between gels using a multi-resolution third order B-spline approach interleaved inside the image registration process.
  4. Shifted-log image model for variance stabilisation in the registration process.
  5. A registration strategy and validation for DIGE gels, for which previous image-based gel matching algorithms were unable to register reliably due to the increased bias and dynamic range.
  6. A GPGPU (General Purpose Computation on Graphics Processing Units) implementation to harness the scientific parallel computing power available in modern graphics cards.

 

Figure: Two overlaid gel images (magenta and green) before (left) and after (right) RAIN alignment.
 

RAIN is implemented with GPGPU techniques together with features found only on Nvidia graphics card hardware. Other vendors’ hardware (e.g. ATI, Matrox) will not work. The minimum requirement is an Nvidia Geforce 6 series graphics card with 128Mb of memory. More recent Nvidia graphics card should also be 100% compatible, but please note that the algorithm has only been tested on the following models: 6800Ultra, 6800Ultra Go, 7800GT, 7800GTX and 8800GTX. Further resources on GPGPU can be found at:

  1. Introduction to GPGPU techniques
  2. A comprehensive portal for GPGPU research over the past five years
  3. Nvidia’s page on their Cg programming language for GPUs
  4. J.D. Owens et al. (2007) A Survey of General-Purpose Computation on Graphics Hardware. Comp. Graph. Forum, 26, 2007, 80-113.

The command line program for Windows can be downloaded here. The gel images used for validation can also be downloaded:

  1. Veeser et al. (4Mb)
  2. Imperial (48Mb)
  3. UCD Silver Stain (160Mb)
  4. UCD DIGE (260Mb)

Key References: