applied to identify and analyze authentic CME sites by support
vector machine classifiers [57]. Furthermore, on the organism
level, compared to the morphological characteristics and molecular
data, using image analysis and machine learning approaches (i.e.,
artificial neural network) is another invasive approach to investigate
the house shrew, where an automated identification system is devel-
oped to reveal the shape characteristic features differentiating the
specimens [58].3.1.3 Application in
Interaction-Focused
Analysis
Protein-protein interactions (PPIs) may represent one of the next
major classes of therapeutic targets [59], and such intricate
biological systems cannot be cost-efficiently tackled using conven-
tional high-throughput screening methods. To overcome the
inherent problem of rigid approach on predicting the binding
affinities when the modeling assumptions are not confirmed, a
new RF-Score that circumvents the need for problematic modeling
assumptions via nonparametric machine learning was used to
implicitly capture binding effects that are hard to model explicitly
[60]. And the protein-ligand interaction also requires predictive
model for high-throughput screen: the machine learning-based
models, PPI-HitProfiler, mainly decision trees, have been devel-
oped to determine a global physicochemical profile for putative PPI
inhibitors, so that it can screen drug-like compound collection
from any chemical library enriched in putative PPI inhibitors
[61]; and to screen potential drug (or target) candidates for bio-
chemical verification on drug-target interactions, the similarity-
based machine learning-based approaches have been proposed to
combine drug and target similarities to generate models for pre-
dicting new drug-target interactions [62]. Besides, to screen
genome-wide targets of transcription factors (TFs) on regulatory
level, the regulatory interaction predictor (RIP) with condition
independent employs SVMs trained on a set of experimentally
proven RIs from TRANSFAC, where the features of such RIs are
extracted from the common TF (TF-module) of co-regulated
genes by integrating the meta-analysis of gene expression correla-
tion and in silico predictions of TF binding sites [63]. And on the
epigenetic regulation, many computational methodologies for
miRNA-mRNA target gene prediction have been developed based
on cross-species sequence conservation of the seed segment of the
miRNA and the region of the mRNA target [64]. Meanwhile, the
methods that do not rely on conservation are increasing due to
analyzing non-conserved genomic sequences. For example, the
NBmiRTar adopts machine learning by a naive Bayes classifier and
has shown higher sensitivity and specificity than algorithms that rely
on conserved genomic regions [65]; and the TargetSpy can predict
target sites regardless of the presence of a seed match, which is also
based on machine learning and automatic feature selection using a
wide spectrum of compositional, structural, and base-pairing fea-
tures covering current biological knowledge [66].Revisit of Machine Learning Supported Biological and Biomedical Studies 191