Update method summary

README.md

@@ -56,7 +56,7 @@ The `callback` algorithm consists of three simple steps:
 
 1. First, we generate synthetic null variables, formally called knockoff features, where we augment the single-cell data being analyzed with "fake" genes that are known not to contribute to any unique cell type. 
 2. Second, we perform both preprocessing and clustering on this augmented dataset. 
-3. Third, we calibrate the number of inferred clusters by using a hypothesis testing strategy with a data-dependent threshold to determine if there is a statistically significant difference between groups and if re-clustering should occur.
+3. Third, we calibrate the number of inferred clusters by using a hypothesis testing strategy with a data-dependent threshold to determine if there is a statistically significant difference between groups. If any pair of groups does not have statistically significant differences then re-clustering occurs.
 
 The synthetic knockoff genes act as negative control variables; they go through the same analytic steps as the real data and are presented with the same opportunity to be identified as marker genes. The `callback` algorithm uses the guiding principle that well-calibrated clusters (i.e., those representing real groups) should have significantly differentially expressed genes after correcting for multiple hypothesis tests, while over-clustered groups will not. We use this rule to iteratively re-cluster cells until the inferred clusters are well-calibrated and the observed differences in expression between groups are not due to the effects of double-dipping.