I am attempting to use the Seurat FindAllMarkers function to validate markers for rice taken from the plantsSCRNA-db. I want to use the ROC test in order to get a good idea of how effective any of the markers are. While doing a bit of research, different stats forums say: "If we must label certain scores as good or bad, we can reference the following rule of thumb from Hosmer and Lemeshow in Applied Logistic Regression (p. 177):
0.5 = No discrimination 0.5-0.7 = Poor discrimination 0.7-0.8 = Acceptable discrimination 0.8-0.9= Excellent discrimination0.9 = Outstanding discrimination "
https://www.statology.org/what-is-a-good-auc-score/#:~:text=0.5%2D0.7%20%3D%20Poor%20discrimination,%3E0.9%20%3D%20Outstanding%20discrimination
For more background, the output of the function returns a dataframe with a row for each gene, showing myAUC: area under the Receiver Operating Characteristic, and Power: the absolute value of myAUC - 0.5 multiplied by 2. Some other statistics are included as well such as average log2FC and the percent of cells expressing the gene in one cluster vs all other clusters.
With this being said, I would assume a myAUC score of 0.7 or above would imply the marker is effective. However given the formula used to calculate power, a myAUC score of 0.7 would correlate to a power of 0.4. So with this being said, would it be fair to assume that myAUC should be ignored for the purposes of validating markers? Or should both values be taken into account somehow?
In the Seurat R package for analyzing single-cell RNA-seq data, "power" and "myAUC" are both functions used for selecting the most informative features or genes in the dataset. However, they employ different approaches and criteria to achieve this.
In summary, the "power" function identifies highly variable genes based on their expression dispersion relative to mean expression, while the "myAUC" function ranks genes based on their ability to discriminate between two predefined groups or conditions using the area under the ROC curve. Both functions aim to identify genes that are potentially important for distinguishing between different cell types, states, or experimental conditions, but they use different statistical and computational approaches to achieve this goal.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology