Not yet reviewed by Pith; the record is open.

There is a growing need for unbiased clustering methods, ideally automated. We have developed a topology-based analysis tool called Two-Tier Mapper (TTMap) to detect subgroups in global gene expression datasets and identify their distinguishing features. First, TTMap discerns and adjusts for highly variable features in the control group and identifies outliers. Second, the deviation of each test sample from the control group in a high-dimensional space is computed and the test samples are clustered in a global and local network using a new topological algorithm based on Mapper. Validation of TTMap on both synthetic and biological datasets shows that it outperforms current clustering methods in sensitivity and stability; clustering is not affected by removal of samples from the control group, choice of normalization nor subselection of data. There is no user induced bias because all parameters are data-driven. Datasets can readily be combined into one analysis. TTMap reveals hitherto undetected gene expression changes in mouse mammary glands related to hormonal changes during the estrous cycle. This illustrates the ability to extract information from highly variable biological samples and its potential for personalized medicine.