Skip to main content

Glossary

Quick reference for terminology used in CellType-Refinery documentation.

A​

Anti-Marker​

A marker that should NOT be expressed in a cell type. Used to penalize incorrect assignments. For example, CD45 is an anti-marker for epithelial cells.

Anti-Penalty​

The score deduction applied when a cluster expresses anti-markers for a cell type. Higher anti-marker expression results in larger penalties.

Ambiguous Root​

A stop reason indicating the top two root categories scored too close together (gap < root_gap_threshold). The cluster cannot be confidently assigned to a single root.

Ambiguous Siblings​

A stop reason indicating children at the current hierarchy level scored too close together (gap < min_gap[level]). Assignment stops at the parent level.

Assignment Path​

The sequence of nodes traversed during hierarchical descent, from root to the final assigned cell type.

B​

Base Score​

The fundamental score component derived from marker expression, before applying DE bonuses or penalties.

Batch Effect​

Technical variation between experimental batches that can affect marker expression measurements and clustering results.

C​

Cell State​

A transient or activation-dependent phenotype of a cell, such as "activated," "exhausted," or "proliferating." Distinguished from cell type identity.

Cell Type​

A distinct biological category of cells defined by characteristic marker expression patterns and functional properties.

Cluster​

A group of cells identified by unsupervised clustering algorithms based on expression similarity.

Commonness Penalty​

A score reduction applied to markers that are frequently expressed across many cell types, reducing their discriminative power.

Confidence​

The minimum margin (score gap) along the entire hierarchical assignment path. Higher confidence indicates more decisive assignments. Range: 0 to ~5+.

Coverage​

The fraction of a cell type's expected markers that are present in the marker panel. Coverage of 0.8 means 80% of markers are available for scoring.

D​

DE Bonus​

Extra score contribution for markers that appear in the cluster's top differentially expressed genes. Rank-weighted to give higher weight to top-ranked DE genes.

DE Component​

The differential expression portion of the marker score. Calculated from rank-weighted DE hits with commonness penalty.

DE Hits​

The count of a cell type's markers that appear in the cluster's differentially expressed gene list.

Differential Expression (DE)​

Statistical comparison identifying genes with significantly different expression between a cluster and other cells.

E​

Enrichment​

Z-score of a cluster's marker expression compared to the global mean. Positive enrichment indicates the cluster expresses the marker more than average.

Enrichment Threshold​

The minimum Z-score required for a marker to be considered enriched in a cluster.

Expression Matrix​

The data structure containing gene expression values for each cell, with genes as rows and cells as columns.

F​

Final Score​

The composite score combining base score, DE component, and anti-penalty used for ranking cell type candidates.

Frac Markers On​

The fraction of a cell type's markers that are expressed above threshold in the cluster.

G​

Gap Threshold​

The minimum score difference required between candidates for confident assignment. Different thresholds may apply at different hierarchy levels.

Gating​

The process of checking whether a cluster meets threshold requirements to be considered for a cell type. Gates include coverage, positive fraction, enrichment, and anti-penalty checks.

Gap​

The score difference between the best candidate and the runner-up at any hierarchy level. Larger gaps indicate more confident assignments.

Global Mean​

The average expression of a marker across all cells in the dataset, used as reference for enrichment calculations.

H​

Hard Requirements​

Mandatory marker expression criteria that must be met for a cell type to be considered. Failure to meet hard requirements results in immediate exclusion.

Hierarchical Descent​

The top-down traversal through the marker hierarchy, starting from root categories and descending to specific cell types.

Hierarchy Level​

The depth in the marker map tree structure. Level 0 is root, with increasing levels representing more specific cell type categories.

I​

Intermediate Node​

A non-leaf node in the marker hierarchy that has children. Assignment may stop at intermediate nodes when children are ambiguous.

L​

Leaf Node​

A terminal node in the marker hierarchy with no children. Represents the most specific cell type category available.

Leaf Reached​

A stop reason indicating the algorithm successfully descended to the deepest level of the hierarchy. This is the ideal outcome.

M​

Marker​

A gene whose expression is characteristic of a particular cell type or category.

Marker Map​

The hierarchical data structure defining cell types, their relationships, and associated markers.

Marker Panel​

The set of genes measured in the experiment that can be used for cell type annotation.

Mean Enrichment​

Average Z-score of a cluster's expression for a cell type's markers compared to global expression.

Mean Positive Fraction​

Average fraction of cells in a cluster expressing each of a cell type's markers above the Q75 threshold.

Min Gap​

The minimum score difference required between the best and second-best candidates at each hierarchy level. Below this threshold, assignment stops as "ambiguous."

N​

No Child Passed​

A stop reason indicating none of the children at the current hierarchy level passed the gating requirements. Assignment stops at the current (parent) level.

No Root Passed​

A stop reason indicating no root category passed the gating requirements. The cluster is labeled "Unassigned."

Node​

An entry in the marker hierarchy representing a cell type category at any level.

O​

Overclustering​

Excessive subdivision of cell populations resulting in biologically redundant clusters.

P​

Parent Node​

The immediate ancestor of a node in the marker hierarchy. All siblings share the same parent.

Per-Cell Voting​

A process that calculates cell-level scores when siblings are ambiguous. Provides evidence for refinement but does NOT override cluster-level assignments.

Positive Fraction​

The fraction of cells in a cluster expressing a marker above the Q75 (75th percentile) threshold.

Positive Fraction Threshold​

The minimum fraction of cells required to express a marker for the cluster to be considered positive for that marker.

Q​

Q75 Threshold​

The 75th percentile expression value across all cells, used as the cutoff for determining positive marker expression.

R​

Rank Weight​

The weight applied to DE genes based on their rank position, giving higher influence to top-ranked genes.

Refinement​

The process of improving cell type assignments through subclustering, merging, or manual adjustment.

Root Category​

A top-level category in the marker hierarchy (e.g., Immune, Epithelial, Stromal). All hierarchical descent begins at root level.

Root Gap Threshold​

The minimum score difference required between the top two root categories for confident root assignment.

Root Hard Requirements​

Mandatory marker expression requirements for root categories. For example, Immune Cells may require CD45 expression above a threshold.

Root Veto Markers​

Markers that, if expressed above threshold, block assignment to a root category. For example, Pan-Cytokeratin expression may veto Immune Cells assignment.

Runner-Up​

The second-highest scoring candidate at any hierarchy level. Used to calculate the gap for confidence assessment.

S​

Score​

The numerical value representing how well a cluster matches a cell type based on marker expression patterns.

Sibling Comparison​

The principle that only cell types sharing the same parent are compared at each hierarchy level. This ensures fair comparison within biological categories.

Sibling Node​

A node sharing the same parent as another node in the marker hierarchy.

Stop Reason​

The explanation for why hierarchical descent stopped at a particular level. Values: leaf_reached, ambiguous_siblings, no_child_passed, no_root_passed, ambiguous_root.

Subcluster​

A refinement operation that re-clusters a cluster at finer resolution to separate mixed populations.

T​

Technical Marker​

A marker used for quality control or batch correction rather than biological cell type identification.

Top-Down Traversal​

The annotation strategy of starting at broad root categories and descending through the hierarchy to specific cell types, rather than comparing all cell types directly.

Total Score​

Synonym for final score; the aggregate score used for ranking candidates.

U​

Unassigned​

The label given to clusters that fail to pass gating requirements for any root category.

Underclustering​

Insufficient subdivision of cell populations resulting in mixed clusters containing multiple cell types.

V​

Veto​

The blocking of a cell type assignment due to expression of exclusionary markers.

Voting Fraction​

In per-cell voting, the proportion of cells within a cluster that vote for a particular cell type.

W​

Weight​

A multiplier applied to marker scores to adjust their influence on final scoring. Some markers may be weighted higher due to specificity.

Z​

Z-Score​

A statistical measure indicating how many standard deviations a value is from the mean. Used for enrichment calculations.

See Also​