Marker Gene

The gene must be exclusively expressed or highly enriched in the target cell type compared to all other populations in the dataset.

• Binary pattern: Ideally, expression is detected in >80% of target cells and <5% of background cells.
• Avoids shared lineages: Genes expressed across closely related subtypes (e.g., CD4+ and CD8+ T cells) fail to uniquely resolve boundaries.
• Example: INS (insulin) is highly specific to pancreatic beta cells and absent from alpha, delta, and exocrine cells in a healthy pancreas.

The gene must exhibit a strong signal-to-noise ratio to withstand technical dropout events inherent to single-cell assays.

• Dropout resistance: High expression reduces the probability of a transcript being missed during reverse transcription or amplification.
• Clear fold change: A log2 fold change >2 between the target cluster and the next highest cluster is a standard threshold.
• Example: HBB (hemoglobin subunit beta) shows extremely high counts in erythrocytes, making it detectable even in low-quality cells.

Expression must be stable across biological covariates such as age, sex, and disease state, unless the marker is specifically designed to track a pathological transition.

• Activation state independence: Avoid genes that fluctuate with cell cycle phase or metabolic stress unless annotating a specific activation state.
• Batch consistency: The gene should rank consistently as a top differentially expressed feature across independent experiments and sequencing platforms.
• Example: FOXP3 is a stable lineage-defining transcription factor for regulatory T cells, unlike IL2RA (CD25) which varies with activation.

The marker gene should not be highly expressed by a fragile or abundant cell type that contributes significant ambient RNA to the suspension, as this creates false-positive background signal.

• Ambient RNA source check: Genes from lysed erythrocytes or stressed hepatocytes often contaminate droplets globally.
• Validation: Use negative control cells (e.g., empty droplets or orthogonal assays) to confirm signal is intracellular.
• Example: MALAT1, while highly expressed, is a poor marker because its ubiquitous presence in ambient RNA masks true cell-type specificity.

The gene must validate across orthogonal technologies and independent cohorts to rule out protocol-specific artifacts.

• Technology concordance: Expression should be confirmed in 10x Genomics, Smart-seq2, and spatial transcriptomics platforms.
• Protein validation: When possible, the transcript should correlate with surface protein detection via CITE-seq or flow cytometry.
• Example: CD3E is a robust pan-T cell marker detectable at both mRNA and protein levels across all common single-cell platforms.

The ideal marker is not just a correlative feature but a lineage-defining transcription factor or a cell-type-specific structural protein with a known biological role.

• Transcription factors: Master regulators (e.g., PAX6 for retinal progenitors) provide mechanistic grounding.
• Structural proteins: Cytoskeletal or extracellular matrix components (e.g., COL1A1 for fibroblasts) define tissue architecture.
• Avoid metabolic genes: Housekeeping and ribosomal genes are poor markers due to ubiquitous expression.