Computational Drug Discovery and Design

available in WKEA, KNIME, and Rapidminer which makes it easy

to experiment with a large number data processing and machine

learning algorithms.

One of the common formats consisting of feature values for a

group of instances, which is almost supported by all the above-

mentioned machine-learning platforms, is the Attribute-Relation

File Format (ARFF). The ARFF file consists of header and data

part. The header part consists of title, name of the relation, a list of

attributes (features) with their types and the data part consists of

the values of the calculated features with the class information for

each instance. In Fig.3 we have presented dummy ARFF file

consisting of features for two classes—drug targets and nontargets

with four features (Molecular weight, mean hydrophobicity, aro-

matic amino acid composition, and charged amino acid

composition).

Dataset consisting of

known drug targets

and non drug targets

Fixed length representation of the protein sequences by

calculating measurable properties such as amino acid

composition, dipeptide composition, physicochemical

properties, pseudo amino acid composition etc.

Training using machine

learning algorithms

Selection of best

prediction model

Performance evaluation using K-fold cross

validation or Leave one out cross validation

or by training/testing set split

Fig. 1Schematic representation of the steps for drug target prediction

Human Drug Targets and Their Interactions 23