With the help of single-cell RNA sequencing (scRNA-seq), scientists try to understand what makes individual cells unique. This is important in everything from tracking immune cell behavior to identifying early disease markers or monitoring how cells respond to treatment. A key step in this process is identifying cell markers, genes that are highly active in certain cell types but not in others. These markers help researchers label and classify different cell types.

Traditionally, researchers use clustering algorithms to group cells with similar gene expression profiles and then look for markers specific to those clusters. However, clustering has its limitations. It forces cells into groups that may not fully reflect their subtle differences or complex relationships, especially in tissues where the boundaries between cell types aren’t so clear-cut.

To address this challenge, a research team at Yale University developed a new tool called Localized Marker Detector (LMD). Rather than depending on predefined clusters, LMD identifies what the authors call “localized genes”, genes that are expressed only in tight-knit groups of very similar cells.

LMD Workflow

How does it work? Imagine all your cells as dots on a map, where closer dots represent cells that are more alike. LMD builds a “cell-cell affinity graph” from this map, showing how connected each cell is to its neighbors based on gene expression. It then simulates how the expression of a gene would “spread” across this map. If a gene’s activity stays concentrated in one small neighborhood, it’s flagged as a good candidate marker for a localized cell type or state.

LMD doesn’t just spit out a list of genes, it also organizes them into functional gene modules, helping researchers understand what roles those genes might be playing, such as involvement in the cell cycle or specific signaling pathways. The method is multi-scale, meaning it works whether you’re looking for broad cell types or fine subtypes.

The researchers applied LMD to datasets from mouse bone marrow and hair follicle dermal condensates, two systems with complex cell diversity. LMD helped them detect shared and sample-specific gene signatures, as well as identify novel cell populations, all without the need for batch effect correction or stitching together data from different samples. This is a major advantage because batch correction can introduce bias or mask true biological variation.

To benchmark its performance, the team tested LMD on ten different scRNA-seq datasets and compared it against eight other widely used tools. LMD consistently performed better, identifying more precise and biologically meaningful markers. This makes it a valuable tool for researchers seeking to understand cellular diversity, map developmental pathways, or find disease-related biomarkers.

For scientists working in drug discovery, this kind of precision matters. Drugs often target specific cell types or states, and being able to clearly distinguish these with reliable markers can improve the development of therapies and diagnostics. Tools like LMD open new doors for analyzing complex tissues with greater accuracy, revealing rare or transitional cells that might otherwise go unnoticed.

Availability – The R package of LMD and the code used for data analysis are available on GitHub (https://github.com/KlugerLab/LocalizedMarkerDetector).