Figure 2 (IMAGE)
Caption
Overview of the scICE method.
(a) scICE begins by selecting a range of “resolution” values that produce a specific target number of clusters. These resolutions are then tested to find the most stable one.
(b) To measure stability, scICE calculates the Inconsistency Coefficient (IC) for each resolution setting — with an IC of 1.0 representing perfect stability, and higher values indicating greater inconsistency. The resolution with the lowest IC is selected as the most stable for that cluster number.
(c) The selected result is further validated by repeating the clustering multiple times to confirm that its IC remains low and stable across tests.
(d) scICE identifies a “consistent clustering set”: only cluster numbers that consistently show low IC values are accepted as reliable, while unstable results are excluded. This ensures that only the most trustworthy cluster groupings are reported.
Credit
Institute for Basic Science
Usage Restrictions
Attribution Required
License
Original content