From Experimental Approaches to Computational Techniques: A Review on the Prediction of Protein-Protein Interactions

Table 5

A summary of related work in inferring PPI networks. The first column presents the study, this is followed by advantages and limitations of the study.

Related Work	Advantages	Limitations	Ref
A Bayesian networks approach for predicting protein-protein interactions from genomic data	Pioneering study which applied a Bayesian approach to infer PPI in yeast by integrating diverse genomic data	Only four features were integrated. By integrating more features an improvement in interactome coverage and classification predictive performance may be achieved	[11]
A Bayesian networks approach for predicting protein-protein interactions from genomic data	Sixteen diverse features were integrated using a NB classifier to predict PPI in yeast.	The NB classifier approach was applied—this classifier assumes feature independence. Subtle dependencies between features may have an adverse affect on the NB performance. ROC curves were the only assessment method applied to measure the predictive performance of the classifier.	[10]
Information assessment on predicting protein-protein interactions	Application of RF, NB and logistic regression for the prediction of PPI in yeast. Discovered MIPS and GO annotation data were dominant features	Only used subset of data. Missing data was removed.	[9]
Probabilistic model of the human protein-protein interaction network	One of the first studies to integrate diverse “omic” data for the prediction of PPI in human.	A NB approach was employed to infer PPI. Three gene coexpression datasets will employed however only the maximum likelihood ratio per gene coexpression data source per protein pair was considered	[60]