Reverse Engineering (Aracne)

1
Reverse Engineering (Aracne)
2
Aracne

• Aracne stands for Algorithm for the
  Reconstruction of Accurate Cellular NEtworks.
• Aracne uses information theory to compute the
  mutual information between pairs of markers (or
  genes) in a set of microarray experiments.
• From these mutual information computations, an
  interaction network is inferred.

3
Aracne C Implementation

• Andreas group actively maintains a C
  implementation of Aracne.
• Supports creating networks over all markers, a
  subset of markers, or a single hub marker.
• This version is can also be used to do large,
  parallel runs for large data sets.
• Manjunath is developing a Workbench component
  that calls this C implementation natively.

4
Workbench Implementation

• The Workbench implementation is called Reverse
  Engineering. It only supports running Aracne for
  a single hub gene.
• It is run in two stages compute interactions
  with the hub gene (Analyze 2D) and build the
  network (Create Network).
• Cytoscape is used to display the networks created
  by Reverse Engineering.

5
Issues - Correctness

• The results do not match those of the C
  implementation. It has been confirmed with
  Andreas group that the Workbench version is
  incorrect.
• Two of the interaction computation modes (Mutual
  Information Fast, and Linear) are not
  functional.
• Adjacency matrices cannot be loaded from the
  current version of the C Aracne implementation.

6
Issues - Design

• The words gene and marker are often confused
  in the interface.
• The MYC gene is hard-coded in to the interface.
• The Histograms do not lay out properly.
• All sub-panels of Reverse Engineering except the
  first are probably not useful to anybody but
  Andreas team, but even they do not use Workbench
  for Aracne.

7
Issues - Code

• The codebase is enormous for the relatively
  limited amount of functionality.
• There are many areas of redundancy. There seems
  to have been an attempt to rewrite parts of the
  algorithm, but both the old and new parts still
  exist and are now entangled. (For example, the
  CSInformationTheory class is approximately the
  same as GeneGeneRelationship).
• Code documentation is fairly reasonable in some
  areas, but absent in others.

8
Proposed Fixes

• The existing interaction computation methods
  appear to be correct. This can be retained, but
  the network calculation code must be rewritten.
• The interface is very poorly implemented, and it
  is not clear what most of it is for. A simpler
  and more convenient interface should be
  implemented from scratch.
• Support for more than one hub gene should be
  included in the new implementation.