We propose a novel method for imputing missing values in microarray data.
The proposed method works in a global learning with local preservation scheme.
We design a strategy for automatically selecting similar genes.
Experimental results demonstrate its effectiveness on real-world datasets.