FIGURE 4.

Approaches for missing value imputation. Global approaches such as (A) Singular value decomposition (SVD), (B) Multi-layered perceptron (MLP), and (C) Recurrent Neural Network use global correlation information extracted from gene expression array to estimate missing values. Local approaches such as (D) K-nearest neighbor (KNN) and (E) Support vector regression (SVR) utilize local similarity in the gene expression data. The solid green circles in (B,C) represent known gene expression values and solid red circles represent missing ones. Hollow circles in (D,E) represent samples.