Skip to main content
[Preprint]. 2023 Mar 19:2023.02.22.529573. [Version 2] doi: 10.1101/2023.02.22.529573

Figure 2:

Figure 2:

Exploration of PCA, RPCA and AE normalization across hyperparameters. A, (Left) Precision-recall (PR) performance of AE-normalized data generated with latent space sizes 1, 3 and 5 evaluated against CORUM protein complexes. (Right) Corresponding contribution diversity plots depicting CORUM complex contributions from AE-reconstructed and AE-normalized data. B, (Left) PR performance of RPCA-normalized data generated with λ set to 0.0049, 0.007 and 0.0091 evaluated against CORUM protein complexes. (Right) Corresponding contribution diversity plots illustrating complex contributions in RPCA-reconstructed and RPCA-normalized data. C, Scatter plot of Pearson correlation coefficients between un-normalized data and reconstructed data as well as between un-normalized data and normalized data generated by PCA, AE and RPCA normalization. Y-axis contains Pearson correlation coefficient values, and the X-axis contains the number of removed principal components (first 1, 3, 5, 7, 9, 11, 13, 15, 17, 19) for PCA-normalization, latent space sizes (1, 2, 3, 4, 5, 10) for AE-normalization and λ (approximately 0.0049, 0.0056, 0.0063, 0.007, 0.0077, 0.0084, 0.0091) for RPCA-normalization.