Integrating genomic features improves the performance of the KNN algorithm.

<p>(A) Shown is a comparison of the training R² values for genes in either the basic KNN algorithm (x-axis) or the extension of the KNN algorithm that integrates genomic features (<i>KNN-IGF</i>, y-axis). Only genes that have training R²≥0.05 in the basic KNN were included in the analysis. Results are shown for each cross validation scheme, along with the fraction of genes with improved training R² of 0.1 or more in KNN-IGF (pink). (B) Shown is a comparison of the test R² values for same genes in either the basic KNN algorithm (x-axis) or the KNN-IGF (y-axis). Results are shown for each cross validation scheme, along with the fraction of genes that better predicted in the basic KNN (under the blue line) or in KNN-IGF (above blue line). (C) Same as (b), but filtered for genes whose training R² when integrating genomic features was at least 0.1 higher than the training R² of the basic KNN algorithm. As seen, most genes with improved training R² also have better test R².