Intelligence Algorithms for Protein Classification by Mass Spectrometry
Table 2
Biomarker analysis algorithms and their advantages as well as disadvantages.
Method
Advantages
Disadvantages
Samples
Support Vector Machine
High robustness to noise and good ability to recover informative features, could work well on nonlinear problems. Stable classification rate of candidate biomarkers, high classification accuracy
Inferior in terms of the number of recovered informative genes, must according to the collaborative information of multiple genes, hard to train and hard to find kernel function
Noisy high-throughput proteomics and microarray data set Sphingosine and progesterone Metabolomics datasets
Decision Tree
easy to interpret, nonparametric method
May be stuck in local minima, overfitting data, could not be learned online
Volatile oils and S. mutans
Neural Networks algorithm
Identify masses that accurately predict tumour grade, high cross-validation on test data sensitivity rate and specificity rate
Need huge volume of samples, computational expansive to train, black box model, overfitting, hard to select meta-parameter