Table 4.

Supervised machine learning methods and examples of their application to bioinformatics.

Supervised ML method	Description	Strength	Weaknesses	Application examples
Linear Regression	Data Regression to its mean value Best Fit Line (Mean Pattern of the dataset) Gradient Descent Least Square Function Continuous Output Normality Assumption Linearity Assumption	Computationally inexpensive Weighed Sum Prediction Reduces complex dataset to a singular function Less prone to overfitting	Reduces larger complex dataset to a singular function Assumption of Linearity Relationship is seldom applicable Does not distinguish Outliers which might bias regression	Prediction and Evolutionary info analysis of protein structure [137] Prediction and evolutionary information analysis of protein solvent accessibility [138] Genetic Expression inference [139] Genotype Prediction based on Single Nucleotide Polymorphism [140] Prediction of protein secondary structure [141]
Logistic Regression	Extension of Linear Regression Logistic line fitting Probability modelling Non-linearity acceptance	Probability-based classification (rather than final classifications) Fast Training Extension to Multiclass Classifications Less prone to overfitting	Complex Multiplicative weighted function Complete Separation of classes Unrepresentative for classes that highly overlap	Cellular Phenotype classification based on gene expression profile [142] Gene Selection [143] Disease Classification from microarray data [144] Molecular Classification of Cancer [145]
Support Vector Machine	Hyperplane Data classification Classes separation on higher dimensionality Kernel Transformation	Easy implementation to well defined classified categories Effective in high dimensional spaces Non-linear input acceptance	Not suitable for overlapping classes Can be prone to overfitting when number of features exceeds the number of samples No probabilistic explanation for classification	Classification on Gene functional annotations from a combination of protein sequence and structure data [146] Cancer Classification from genetic expression [147] Protein subcellular classification prediction [148] Structural Classification of proteins [149]
Naïve- Bayers	Probabilistic Classification Probabilistic Bayer’s theorem Conditional Independence between variables Most used for classification	Reduced risk of overfitting on small datasets Probabilistic classification Fast training Computationally inexpensive Scales linearly	Does not incorporate feature interactions Performance sensitive to skewed data Requires assumption that variables are conditionally independent	MicroRNA target prediction [150] Prediction of Protein Interaction Sites [151] Prediction of Protein coupling specificity [152]
Decision Tree Classifier & Forest Tree	Classification or Regression modelling Parameter based Data splitting of variable with highest information gain Data entropy Information Gain Theory Gini Coefficient	Easy interpretation and analysis Valued on smaller datasets Multiclassification applicability White box model highlighting classification pattern Easily assembled	Tendency to overfit Lack of linear smoothness	Prediction of microorganism growth temperatures and enzyme catalytic optima [153] Protein Structure Prediction from enzymatic turnovers [129] Microbial Genome prediction [154] MS cancer data classification [155] Gene Selection for Cancer identification [156] Human protein function prediction [157] Prediction of Protein interaction [158]
k-NN	Classification and Regression modelling Clustering classification Instance-based learning, i.e. lazy learning Parameter selection on Kernel basis Higher dimensions for clustering	Simple to implement Learn non-linear boundary Robust to noise in input data	Inefficiency on training larger datasets Expensive computational cost K value evaluation based on mixed heuristics Unclear	Gene selection for sample classification based on gene expression data [159] Classification for Cancer diagnosis [160] Prediction of Metabolic pathways dynamics [161]