Experimental methods for drug target identification and
drug–target interaction are costly and time consuming. A compu-
tational predictor at the human proteome level will be of great
importance for the pharmaceutical industry. Computational meth-
ods can also expedite the process of drug target identification by
reducing the time consumed in the experimental methods.
The three important challenges that are being tackled in the
drug-discovery pipeline are the drug target identification/predic-
tion, drug–target interaction prediction, and prediction of drug
binding residues. Accurate and most significant prediction methods
can facilitate in reducing the target search space. The prediction of
drug–target interaction is a challenging task as it presents a situa-
tion of many to many mapping (i.e., a single drug can interact with
many targets and vice versa). Furthermore, the prediction of drug
binding residues is also an inevitable job in target identification
study. Availability of very limited number of target structures in
repositories directs most of the studies to adopt sequence feature
extraction based prediction [9–11], however a structure based
analysis of druggable pockets using physicochemical properties
have also been reported [12]. A similar druggable microenviron-
ment analysis resulted in the development of DrugFEATURE
[13]. There are only few validated and well-proven drug target
prediction methods available. Further reliable improvements and
innovative methods can be used to increase the prediction rate in
drug target discovery. Discovery of a large number of putative drug
targets and their interactions with known drugs are still a basic need
of medicinal sciences for curing life-threatening diseases. Compu-
tational methods can reduce the cost and time which is involved in
drug target identification. The promiscuity of drug–target interac-
tions makes it difficult to develop machine learning based classifica-
tion models. Prediction of new drug targets and drug target
interactions can open new doors for drug discovery and drug
repurposing for a given disease. The drug targets and their interac-
tions may facilitate the understanding of drug action at molecular
level and increase the success rate of drug discovery. The major
steps for prediction of drug targets and their interactions are shown
in Figs.1 and 2, respectively.2 Materials
2.1 Machine
Learning Platforms
For the development of machine learning models, there are many
open source platforms like WEKA [14], KNIME [15], RapidMiner
[16], H 2 O[17], and Scikit-learn [18]. WEKA, KNIME, RapidMi-
ner, and Scikit-learn provide a plethora of data preprocessing,
classification, and clustering algorithms, while H 2 O is mostly dedi-
cated to deep learning neural networks. H 2 O can be implemented
in both R and python. Graphical User Interface (GUI) facility is22 Abhigyan Nath et al.