Missing value estimation methods for DNA microarrays
Troyanskaya, O., Cantor, M., Sherlock, G., Brown, P., Hastie, T., Tibshirani, R., Botstein, D., and Altman, R. B. (2001). Missing value estimation methods for dna microarrays. Bioinformatics, 17(6):520–525. 10.1093/bioinformatics/17.6.520
SVD, KNN and row average imputation are evaluated with different parameter settings on real data sets with regard to robustness, sensitivity and accuracy.
List the paper results concerning method comparison and benchmarking:
Rank of performance: KNN, SVD, row average, zero filling.
"KNN is relatively insensitive to .. K within the range of k=10-20" (Figure 1)
SVD "is sensitive to the type of data" and "is ideally suited .. in terms of .. constituent patterns"
Study design and evidence level
Just 4 imputation algorithms (SVD,KNN) are evaluated from which 2 are singular value substitutions (average,zero).
Analysis is performed over a broad range of hyperparameters (KNN: k=[1,1000], SVD: Eigengenes=[5,30]).
The imputation methods are only analyzed on data with less than 20%.