Scratch Pad

This page is meant to be used as a scratch pad for discussion around different aspeccts of the project's development.

---

Initial Ideas for Mapping Qualitative Mouse Data to Human Phenotypes

September 16th, 2018

Since connecting quantitative mouse data to dichotomous human phenotype data is not something that I have done before, I wanted to start with a relatively straightforward phenotype -- Chronic Mild Stress. This is a great initial dataset to use from GeneNetwork because it contains molecular data (hippocampus mRNA), but also has clinical measures which easily mapped to human phenotypes.

With that in mind, here is one strategy we could consider using with this data (assuming the knowledge graph has been built):

1. Use the distribution of one of the serum corticosterone measures (at time of death or after 15 minutes of restraint) to rank the mice within the strain. Once ranked, obtain those strains at the tails to create two groups: low and high cortisol.
2. Perform differential expression analysis using the hippocampal mRNA from the low/high cortisol mouse groups to obtain significantly differentially expressed genes (DEGs).
3. Identify human homologs for mouse DEGs from Step 2.
4. Use the human knowledge graph to identify which of the homologs have been linked to human phenotypes and diseases related to stress → this gives us what is “known” for humans and stress-related disease.
5. Use Gemma or GEO to obtain human:
   - Hippocampal (preferably on “normal” patients/samples) gene expression data.
   - Gene expression data from diseases related to stress and adrenal function (hold-out human PBMC data, if found).
6. Perform differential expression analysis (meta-analysis) on data obtained from Step 5 (if data is from GEO).
   - Overlap DEGs from Steps 4 and 6 to obtain lists of:
   • Genes important to normal hippocampal function.
   • Genes important to stress-related diseases.
     - Convert all overlapping genes back to mouse
7. Using the lists of overlapping genes from Step 6 and the individual mouse arrays (with the initial list of DEGs) from Step 2:
   - Identify which mice have genes from Step 6 with gene expression sig above/below average expression level.
   - Examine differences in the remaining clinical data from GeneNetwork for mice identified from Step 7.
   - For each mouse, obtain and examine mechanism pathways from the mouse knowledge graph.
8. Additional evaluation:
   - If available, could compare overlapping DEGs obtained from Step 6 with significant DEGs identified from Human stress PBMC studies. The added benefit of studies like these is that PBMCs can be used as biomarkers which can be followed up experimentally -- getting us one step closer to precision medicine.
   - If above works, expand to include a greater number of datasets and consider implementing a stack autoencoder framework.