Skip to main content
Fig. 4 | BMC Bioinformatics

Fig. 4

From: Denoising large-scale biological data using network filters

Fig. 4

Schematic of K-nearest neighbors regression framework. We designed a weighted K-nearest neighbors regression framework to predict the protein expression changes a healthy tissue would undergo when becoming cancerous, given a vector of protein expression profile of a healthy tissue. First, we extract features from the training set of 19 healthy tissue protein expression vectors by PCA. Second, we project the left out healthy vector down to the same PCA space, and third, determine K-nearest neighbors to use for the prediction task. Fourth, we extract the features from the 19 delta vectors by PCA, and fifth, predict the delta vector for the left-out healthy sample by taking the weighted average of the K-nearest neighbors’ delta vectors. Finally, sixth, we project the predicted delta vector from PCA space back to a vector of protein expression values to calculate the error

Back to article page
\