A number of rules were used to filter the dataset.
We filter the dataset to distinguish certain potential artifacts in the assignment of atoms during refinement.
Thus, it is not possible to filter the dataset based on OMP class alone.
Here we see that these approaches significantly outperform ReliefF and SURF when the task is to filter the dataset to the 99 th or 95 th percentiles of SNPs.
To demonstrate how to filter the dataset by the length of peaks, we used "Peak length" filter criteria to obtain a subset of peaks that have length range from 100 bp to 150 bp.
To filter the dataset to codon pairs where (K a / K s )Y|Xa could be estimated with some accuracy, we restricted the analysis to cases where N YaXa + N YsXa > 100.
FDR cut-off value 0.007 (i.e., a conservative cut-off value for correcting over 28 samples separately) was used to filter the dataset obtained before proceeding with further analysis.
After carefully pruning and filtering the datasets, all the datasets were clustered (see Methods section) to avoid biased results due to overrepresentation of similar molecules.
We first filtered the datasets down to approximately 10% of total genes (800 genes for dataset I, 2000 for dataset II, and 4000 for dataset III) based on Median Absolute Deviation MADD).
Thus, we filtered the datasets with high N/A rates before performing other survival analyses.
