Microarray Missing Value Imputation: Supplementary Materials

References

1. Z. Cai, M. Heydari, and G. Lin.
   Iterated Local Least Squares Imputation for Microarray Missing Values.
   Journal of Bioinformatics and Computational Biology.
   4(5)(2006), 935-957.
2. Z. Cai, M. Heydari, and G. Lin.
   Microarray Missing Value Imputation by Iterated Local Least Squares.
   In Proceedings of the Fourth Asia-Pacific Bioinformatics Conference (APBC 2006).
   Taipei, Taiwan. February 13-16, 2006.
   Pages 159-168, 2006.

Release Notes

1. On May 12, 2006, the following supplementary materials were released:
   • Figure 4 in black-and-white (tar gzipped, 10KB);
   • Figure 5 in black-and-white (tar gzipped, 9KB);
   • Full ANOVA test results (tar gzipped, 564KB).
2. On March 11, 2006, the following supplementary materials were released:
   • Six microarray datasets (Alpha, Cyc.a, Cyc.b, Elu, Env, Ta.crc) in MATLAB format: each with a complete dataset and 10 versions of randomly created missing values with 1%, 2%, 3%, 4%, 5%, 10%, 15%, and 20% missing rates (zipped, 62.1MB).
   • 10 NRMSE values of each of the six methods KNNimpute, SKNN, BPCA, LLSimpute, LinCmb, and ILLSimpute on the six microarray datasets with eight different missing rates (zipped, 13KB).
   • Average runtimes per run (in minutes) of each of the six methods KNNimpute, SKNN, BPCA, LLSimpute, LinCmb, and ILLSimpute on the six microarray datasets with eight different missing rates (zipped, 2KB).
   • 10 NRMSE values of each of the four methods KNNimpute, INNimpute, LLSimpute, and ILLSimpute on the six microarray datasets with eight different missing rates (zipped, 9KB).
   • Average runtimes per run (in minutes) of each of the four methods KNNimpute, INNimpute, LLSimpute, and ILLSimpute on the six microarray datasets with eight different missing rates (zipped, 2KB).
   • Average NRMSE values (over 10 runs) of ILLSimpute method on the Elu- and Cyc.b- datasets with 10% missing rate and 5 iterations, using 0.5-1.5 ratios (468B);
     Average NRMSE values (over 10 runs) of ILLSimpute method on the Elu- and Cyc.b- datasets with 10% missing rate and ratio 0.6, after 1-10 iterations (470B).
   • The (average) numbers of coherent genes selected by LLSimpute and ILLSimpute (w/5 iterations) methods on the Cyc.b-datasets with eight different missing rates.

     	1%	2%	3%	4%	5%	10%	15%	20%
     k in LLSimpute NRMSE	140 0.2856	140 0.3346	140 0.4234	140 0.4496	210 0.4595	210 0.6459	210 0.6736	210 0.7076
     Ratio δ in ILLSimpute Average k NRMSE	0.9 139 0.2043	0.9 139 0.1782	1.2 181 0.2315	0.9 139 0.2550	1.3 190 0.2586	0.8 119 0.5141	1.4 197 0.5742	1.3 190 0.6409

   • MATLAB code for ILLSimpute method (8K).

Acknowledgments

This research is partially supported by CFI and NSERC.

Feedback

Please email ghlin[at]cs[dot]ualberta[dot]ca for any additional questions you might have.