Skip to main content
. 2022 May 10;29(5):465–482. doi: 10.1089/cmb.2021.0403

Table 2.

Correlation Completion Accuracy

  MSE
chu_time darmanis CMD
chu_time darmanis
chu chu
Blind estimate 0.01008 0.00796 0.00804 0.21217 0.15855 0.22897
SAVER 0.01949 0.01850 0.01476 0.44604 0.40991 0.52235
drImpute 0.01181 0.00796 0.01371 0.23595 0.16050 0.33769
scRMD 0.01125 0.00868 0.00879 0.23598 0.17621 0.26170
PRIME 0.03707 0.03508 0.02661 0.47846 0.42137 0.33218
SCENA_average 0.01285 0.01119 0.01313 0.27283 0.21596 0.42946
SCENA_ridge 0.01212 0.00708 0.00513 0.20497 0.12942 0.12474

MSE and CMD between the reference correlation and the estimated correlation derived from various methods. SCENAridge has the lowest MSE and CMD among imputation methods in chu_time and darmanis data, and is lower than the baseline (“Downsample”), which is the sample correlation matrix of the downsampled data. All other imputation approaches perform worse than the baseline.

CMD, correlation matrix distance; MSE, mean squared error.