Computational biology · BEGINNER LESSON
ICE
ICE, short for Imputation-based Cell Enrichment, is a computational method for single-cell RNA sequencing data that looks for cells enriched for a chosen gene set. It combines expression imputation with gene-set refinement to improve detection when marker genes are weak or sparse.
What ICE is and what it measures
ICE is a single-cell gene set enrichment analysis method. Single-cell RNA sequencing (RNA-seq) measures which messenger RNA molecules are present in individual cells, giving a gene expression profile for each cell. ICE takes that profile plus a predefined gene set, then estimates an enrichment score (ES), which summarizes whether the genes in the set are concentrated where they matter in the ranked data. The goal is to identify cells significantly enriched for the chosen gene set, especially when the marker genes are weak or show many zeros in the data.
[1] ICE: robust detection of cellular senescence from weak single ...
How ICE works
ICE first uses MAGIC, a diffusion-based imputation method. Imputation means estimating missing or weak values from patterns in similar cells; MAGIC builds a k-nearest neighbor graph, where each cell is linked to similar cells by Euclidean distance, then spreads expression values across that graph to smooth noise and recover dropouts. ICE then refines the marker gene set and computes an enrichment score with a running-sum approach similar to Gene Set Enrichment Analysis (GSEA), which is a way to test whether members of a gene set appear near the top or bottom of a ranked list. Finally, ICE examines the derivative, or rate of change, of the ES trend and uses peaks in that derivative to set a cutoff that separates enriched from non-enriched cells.
[1] ICE: robust detection of cellular senescence from weak single ...
Why ICE was developed
The source study says that detecting senescent cells from single-cell RNA-seq data is hard because canonical markers can be weak and non-specific. It also reports that simply adding more low-signal marker genes does not improve accuracy by itself. ICE was designed to address this by improving weak signals rather than only expanding the marker list.
[1] ICE: robust detection of cellular senescence from weak single ...
Limitations
ICE depends on an initial imputation step, and the source notes that this step is the main bottleneck. The MAGIC component can become computationally expensive on larger datasets, with cost increasing quadratically with the number of cells. ICE is also built for predefined gene sets, so its output depends on the quality and specificity of that set.
[1] ICE: robust detection of cellular senescence from weak single ...
SEE THE IDEA
ICE workflow for single-cell gene set enrichment
Single-cell gene expression profiles plus a predefined gene set.
MAGIC smooths values across similar cells to reduce zeros and noise.
A running-sum method ranks whether the gene set is concentrated in the data.
Peaks in the score’s derivative define which cells are called enriched.
ICE moves from noisy cell measurements to a binary enriched/not-enriched call through smoothing, scoring, and thresholding.
A simplified view of how ICE turns sparse single-cell data into an enrichment decision.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.