cMonkey Principles

In order to use cMonkey for a computation, a user needs to setup a cMonkey run. A run is a set of specific input data and configuration parameters and a set of scoring algorithms that are activated at certain iterations and combined using user-specified weights.

cMonkey Python supports the user by providing the CMonkeyRun class which contains a framework for executing a run. The user can override certain configuration parameters or its methods to customize the run.

Core algorithm

At the heart of the cMonkey algorithm is a scheduler that executes specific scoring modules at user-defined iterations and with a user-defined scoring weight. There are two distinct classes of scoring modules:

• Row scoring modules are based on the input matrix's rows (or genes). The user may define an arbitrary number of row scoring modules in a run.
• Column scoring modules are based on the input matrix's columns (or conditions). Currently, only a single column scoring function per run is supported

Anatomy of a scoring module

A cMonkey scoring module is a class that implements the method

def compute(self, iteration_result, reference_matrix=None)

which returns a DataMatrix object with

• num_genes rows and num_clusters columns for a row scoring module where each cell contains the score for each gene-cluster combination
• num_conditions rows and num_clusters columns for a column scoring module where each cell contains the score for each condition-cluster combination

System defined scoring modules

Row scoring modules

• microarray.RowScoringFunction: This is the standard row scoring function
• network.ScoringFunction: Uses a list of weighted networks to calculate scores
• motif.MemeScoringFunction: Calculates scores based on an external tool pipeline of meme and mast
• motif.WeederScoringFunction: Calculates scores based on an external tool pipeline of weeder and mast
• scoring.ScoringFunctionCombiner: A function module that represents an aggregate of one or more functions that are weighted and combined.

It is recommended to extend the base class scoring.ScoringFunctionBase, to automatically inherit access to some useful and commonly used functionality.

Column scoring modules

• scoring.ColumnScoringFunction: This is the default condition scoring function

How to implement your own scoring module

The simplest way to implement your own scoring module is to derive from scoring.ScoringFunctionBase. This base class provides all necessary infrastructure for scheduling and basic configuration.