Almost every cells in human's body contain the same number of genes so what makes them different is which genes are expressed at any time. Measuring gene expression can be done by measuring the amount of mRNA molecules. However, it is a very expensive and time consuming task. Using computational methods can help biologists to perform gene expression measurements more efficiently by providing prediction techniques based on partial measurements. In this paper we describe how we can recover a gene expression dataset by employing Euclidean distance, Pearson correlation coefficient, Cosine similarity and Robust PCA. To do this, we can assume that the gene expression data is a matrix that has missing values. In that case the rows of the matrix are different genes and columns are different subjects. In order to find missing values, we assume that the data matrix is low rank. We then used different correlation metrics to find similar genes. In another approach, we employed RPCA method to differentiate the underlying low rank matrix from the sparse noise. We used existing implementations of state-of-the-art algorithms to compare their accuracy. We describe that RPCA approach outperforms the other approaches with reaching improvement factors beyond 4.8 in mean squared error.