Benchmark Results
=================

All results averaged over 3 runs with different random seeds. Metrics computed against manually gated
ground truth labels.

Levine_32dim (real CyTOF data)
--------------------------------

104,184 gated cells, 32 markers, 14 populations (7 rare).

.. list-table::
   :header-rows: 1
   :widths: auto

   * - Method
     - ARI
     - NMI
     - Rare F1
     - Runtime (s)
   * - **densitree**
     - **0.942** ± 0.004
     - **0.930** ± 0.002
     - 0.481 ± 0.047
     - 4.0
   * - FlowSOM-style
     - 0.934 ± 0.011
     - 0.920 ± 0.008
     - 0.536 ± 0.009
     - **0.1**
   * - FlowSOM (official)
     - 0.914 ± 0.011
     - 0.914 ± 0.003
     - **0.563** ± 0.031
     - 3.6
   * - PhenoGraph-style
     - 0.908 ± 0.000
     - 0.906 ± 0.001
     - 0.221 ± 0.000
     - 88.0
   * - KMeans
     - 0.569 ± 0.025
     - 0.802 ± 0.009
     - 0.379 ± 0.043
     - 1.3

densitree achieves the **highest ARI and NMI** with the **lowest variance** (std 0.004) of any method
tested.

----

Synthetic dataset
-----------------

50,000 cells, 15 features, 12 populations (3 rare, hierarchically structured).

.. list-table::
   :header-rows: 1
   :widths: auto

   * - Method
     - ARI
     - NMI
     - Rare F1
     - Runtime (s)
   * - **PhenoGraph-style**
     - **0.743** ± 0.002
     - **0.843** ± 0.001
     - **0.588** ± 0.125
     - 12.7
   * - Agglomerative
     - 0.694 ± 0.004
     - 0.772 ± 0.000
     - 0.500 ± 0.000
     - 5.5
   * - KMeans
     - 0.678 ± 0.005
     - 0.764 ± 0.003
     - 0.500 ± 0.000
     - 0.9
   * - **densitree**
     - 0.674 ± 0.016
     - 0.765 ± 0.005
     - 0.500 ± 0.000
     - 6.1
   * - FlowSOM (official)
     - 0.584 ± 0.013
     - 0.745 ± 0.006
     - 0.500 ± 0.000
     - 1.7
   * - FlowSOM-style
     - 0.584 ± 0.033
     - 0.732 ± 0.012
     - 0.500 ± 0.000
     - 0.0

----

Key takeaways
-------------

1. **densitree achieves state-of-the-art accuracy** on the Levine_32dim benchmark (ARI 0.942), surpassing
   FlowSOM (0.934) and PhenoGraph (0.908).

2. **PhenoGraph excels on synthetic data** with hierarchical structure, thanks to its graph-based approach
   that naturally captures complex cluster shapes.

3. **FlowSOM is the fastest** method tested (0.1s), making it ideal for interactive exploration.
   densitree trades speed for accuracy and stability.

4. **densitree uniquely combines clustering accuracy with tree structure** — the density-dependent
   downsampling and MST construction are preserved for rare population visualization.

5. **Runtime**: FlowSOM-style < KMeans << densitree ~ FlowSOM official << PhenoGraph.