How to Run Analysis

Close

How to prepare your dataset and parameters

scGHSOM is an unsupervised hierarchical clustering algorithm designed for single-cell expression data analysis. The input dataset is expected to contain the following types of columns:

• Index – sample or cell identifiers (optional)
• Feature – numerical expression values used for clustering
• Label – known cell-type annotations for evaluation (optional)

---

1. Index column (optional)

If your dataset contains an index column (e.g., cell IDs or barcodes), please specify its column name in the Index field.

This prevents the index from being mistakenly treated as a feature. Leave this field blank if no index column is present.

---

2. Label column (optional)

If true cell-type annotations are available, specify the corresponding column name in the Label field.

• Enables correct computation of external evaluation metrics
• Allows result interpretation based on known cell types

The label column does not affect the clustering process itself. Leave this field blank if labels are unavailable.

---

3. Feature columns

All remaining columns are treated as features, typically representing protein or gene expression levels.

• All feature columns must be numeric
• If categorical or string-valued features are present, scGHSOM may not be suitable for the dataset

---

4. Parameter settings (Tau₁ / Tau₂)

Tau₁ controls horizontal expansion of the hierarchy.
Tau₂ controls vertical expansion.

Both parameters range from 0 to 1. If unsure, the default value 0.1 is recommended.

For improved clustering accuracy, explore values around 0.1 using a step size of 0.05.