Establishing initial data analysis strategy

[swiri021]

dataset summary (what we have, what we can do), data analysis strategy

[swiri021]

Early diagnosis biomarker:

1. Getting DEG (Early vs Late vs Normal)
2. Getting Differential pathways (Early vs Late vs Normal)
3. Feature Extraction(RFE) (Early vs Normal)
4. PCA (Early vs Late vs Normal) with clustering methods
5. Intersected features among all methods and curation with scientific insight(Kicheol)
6. Performance test (AUC curve) with validation dataset
7. Performance test (AUC curve) with external dataset

[swiri021]

@kicheolkim We need a normalization dataset, I am able to do it by myself but I think we have to make consistency for the normalization method. A possible option could be TMM or quantile norm, let me know your thoughts.

[kicheolkim]

• We don't have actual late-stage patients. The majority of patients are early stage.

• For DEG, it should compare within cell types. I used sex and age as covariates (model: ~ Last_Known_Treat_Stat + Sex + AgeAtExamGrp). The age was binned as 10-30, 30-40, 40-50, 50-60, and 60-90.

• Github R code for previous DEG (used in the publication)

[swiri021]

@kicheolkim

Initial feature extraction

• Experiment group: early stage (DiseaseDuration <= median)
• Control group: Late stage (DiseaseDuration > median)

Step1- Ranksum and get pvalues<0.05 from Activation Score
Step2- Recursive Feature Elimination(w/ Cross Validation) and get ranking=1 from Step1 result
Step3- Ranksum and get pvalues<0.05 from Step2 result
Step4- Recursive Feature Elimination(w/ Cross Validation) and get ranking=1 from Step3 result

• Final number of genes = 62
  Notebook Link Step4

[swiri021]

New comments for actual analysis work #24