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 57

The receiver operating characteristic (or ROC curve for short) is a graphical
representation of the relation between sensitivity (expressed as TPR, i.e. true positive
rate) and specificity (determined by FPR - false positive rate) for a binary classi fi cation
which bases upon some measurable quantity.
ROC constitutes the graphical equivalent of the so-called contingency table (or
confusion matrix):

Actual value

Prediction True Positive (TP) False Positive (FP)

False Negative (FN) True Negative (TN)

TPR (equivalent to sensitivity) is expressed as:

TP

TPR =

Y-axis

TP + FN

FPR (equivalent to 1-specificity) is expressed as:

FP

FPR =

FP

X-axis

TN +

As the ROC curve is meant to visualize the dependence of TPR on FPR, FPR
values are typically plotted along the X axis, while the Y axis represents TPR
values.
The “fuzzy oil drop” model focuses on

Δ H - the difference between the
expected and observed potency of hydrophobic interactions at specific points in
the protein molecule. We assume that significant discrepancies between these two
values point to the presence of a ligand which distorts the protein's own structural
form. In this sense, the binary classification mentioned above determines which
amino acids, representing local maxima (or minima) on the

Δ H scale are actually
involved in binding ligands. This comparative method of determining the accu-
racy of theoretical predictions is only applicable to the “fuzzy oil drop” model due
to variations in the

Δ H cutoff values for TPR and FPR parameters respectively.
Since classification can focus either on local maxima (hydrophobicity deficiency)
or minima (hydrophobicity excess), ROC curves are plotted for each criterion
separately. The goal is to determine whether the ligand binds to a cavity repre-
senting a local hydrophobicity deficiency, or is attracted to residues characterized
by excess hydrophobicity where its presence can shield such areas from direct
contact with water.
The ROC analysis is applicable solely for “fuzzy oil drop” model since this
model only bases on the quantitative measurements of the identification criterion
which is the value (cutoff level) for

Δ H values. The other models deliver only binary
solutions expressed as YES - residue engaged in complexation and NO - residue
not engaged in complexation. This is why the ROC curve analysis is presented only
to interpret the results based on “fuzzy oil drop” model.