Comparative Analysis of Techniques Oriented on the Recognition of Ligand Binding Area in Proteins - Identification of Ligand Binding Site and Protein-Protein Interaction Area - page 55

transfer (triggered by later stages of the respiratory chain), an electrochemical
gradient emerges. ATP synthase can then exploit this gradient to synthesize universal
energy carriers (ATP molecules) (Pollak et al. 2007 ) .

4.2.2 Protein Data Set

The list of proteins taken for analysis was determined from PDB based on a keyword
search (with FMN and NAD + as the keywords). Proteins complexed to more than
the selected ligands were excluded from analysis (Rose et al. 2011 ) .
In order to limit the redundancy of the protein test set we selected targets with pair-
wise sequence identity below 30 %, as reported by ClustalW (Chenna et al. 2003 ) .

4.3 Comparison Methodology

Comparative analysis focused on the following criteria of correctness:

1. F-measure
2. MCC
3. ROC curves

These parameters were calculated to validate the end results from each of the
presented programs. All of them are based on the following metrics: TP ( true
positive ) - the number of residues correctly identified as involved in binding ligands;
FP ( false positive ) - the number of residues falsely suspected of involvement
in binding ligands (contrary to experimental data); TN ( true negative ) - the number
of residues correctly identified as not involved in binding ligands, and FN ( false
negative ) - the number of residues which are experimentally known to be involved
in binding ligands but were not identified as such by the tested software.
TP and TN calculations were based upon the “gold standard” provided by the
PDBSum database (Laskowski 2009 ) .

4.3.1 F-Measure

F-measure specifies the so-called precision and recall properties, which are often
referred to when determining the correctness of pattern recognition algorithms.
They give a formal meaning to the notion of accuracy, expressing the number of
cases (instances) in which a correct solution has been reached (Olson and Delen
2008 ; van Rijsbergen 1979 ) .
The measure of exactness (or fidelity) is assumed to express precision, while
completeness is measured by recall. Recall is computed as the fraction of correct