Intelligence Algorithms for Protein Classification by Mass Spectrometry

Table 1

Typical classification algorithms and their characters and samples.

Method	Feature	Samples
Logistic Regression	can predicate the peak intensity patterns exactly and simplify a SVD decomposition [13].	Tandem mass spectrometry
KNN algorithm	by Euclidean distance or by one minus correlation. [11]	ovarian cancer MALDI-MS database
	a modification of Euclidean distance formula [16].	patients with mild cognitive impairment and patients with clinical symptoms of Alzheimer’s disease [16].
Support vector machines	using 4 genes	colon cancer database
	suitable for noisy high-throughput proteomics and microarray data and outperforming in the robustness to noise	SELDI-TOF-MS
	an unsupervised feature selection phase, restriction of the coefficient of variation and wavelet analysis for classification [17].	ovarian cancer database [17].
Decision tree algorithm	a new high-throughput proteomic classification system, and developed by a nine-protein mass pattern [18]	blood samples from prostate cancers and healthy man cohort [18]
Classification tree	partitioning the learning sample into smaller and smaller subsamples to ensure the disease status within each subsample is relatively homogeneous [19].	clinical specimens [19].
	combining MALDI-TOF MS with WCX magnetic beads, and with high sensitivity (98.3%) and high specificity (84.4%) [20].	patients with pulmonary tuberculosis [20].
	boosted feature extraction coupled with the nearest centroid classifier with high accuracy [21].	OCWCX2a [21].
Random Forest	used as both feature extractors and classifier and suit for the small sample [4].	serum samples from patients with ovarian cancer [4].
	a complex proteome with a wide range of protein concentrations [22].	signature peptides [22]
	nonlinear random and combined with a discrete mapping approach [23].	phosphorylation data set [23].
Neural Networks algorithm	a multilayer perceptron ANN with a backpropagation algorithm [24].	SELDI-MS data [24].
	using Naive Bayes with a multilayer perceptron [25].	mass data set with InfoGain and Relief-F [25].
	basing on SRNG and FLSOM [26].	breast cancer listeria and tissue data set [26].
	convolutional neural networks [27].	Q-TOF and IT [27].