Marker Scores in Nygen Analytics

Marker scores help evaluate the strength of marker genes that define specific cell populations and cluster quality. Marker scores range from 0 to 1, measuring the specificity and consistency of gene expression within a cluster. Higher scores indicate genes more uniquely expressed in a particular cluster, with 1 being the theoretical maximum for a perfect marker. This score balances expression magnitude and consistency, providing a robust measure of marker gene quality comparable across different genes, clusters, and datasets.

Source: Dhapola et. al, Nature Comm 2022

‍

Methodology

Marker scores are calculated using a ranking-based method that assesses gene expression specificity across cell clusters. The process begins by applying a dense ranking to library-size normalized gene expression values across all cells. These ranks are then grouped by cluster and averaged, producing mean ranks for each gene in each cluster. The final score for a gene in a particular cluster is computed by dividing its mean rank by the sum of its mean ranks across all clusters. This results in a normalized score between 0 and 1, where higher values indicate genes that are both highly and specifically expressed in a given cluster relative to others.

‍

FAQ

Q: How do marker scores compare to traditional p-values?

Marker scores offer several advantages over p-values:

• Consistency across cluster sizes: Unlike p-values, which can be artificially lowered for larger clusters, marker scores provide a more balanced measure of gene specificity regardless of cluster size.
• Comparability: Marker scores can be easily compared across different datasets and clusters, offering a standardized metric for assessing gene importance.
• Intuitive interpretation: The 0-1 scale of marker scores is more intuitive than the exponential scale of p-values, making it easier for researchers to gauge a marker's significance quickly.

Q: Can marker scores be used to refine clustering?

Yes, marker scores can be valuable for iterative clustering refinement. If you notice clusters with consistently low marker scores, it might indicate over-clustering (too many clusters) or under-clustering (not enough clusters). This information can guide adjustments to your clustering parameters for more biologically meaningful results.

‍