Construct TF-CRE-Gene Regulon by combine all genomic features

[lhqing]

TF-CRE-Gene Regulon

1. TF - CRE link See motif enrichment analysis below
   • [x] Scan motifs ecker-hanqing-analysis/220924-dmr-motif-scan cemba.get_mc_dmr_ds(add_motif=True) # to get DMR RegioDS with motif scan matrix
   • [x] perform motif enrichment for each cluster and meaningful cluster groups
   • [x] Create graph adjacency matrix for TF-CRE
2. Gene - CRE link
   • [ ] correlation
   • [ ] GBT model predictability
   • [x] 3C physical approximation #16
   • [ ] Create graph adjacency matrix for Gene-CRE
3. Gene - TF link
   • [x] correlation
   • [ ] GBT model predictability
   • [ ] Create adjacency matrix for Gene-TF
4. Construct Regulon
   • [ ] link eRegulon by raw adjacency matrix
   • [ ] filter by GSEA Leading edge analysis
   • [ ] quantification in each cluster
   • [ ] QC regulon

[lhqing]

Motif Enrichment Analysis

1. Use pycistarget motif collection (4096 motif / motif clusters with a mouse TF annotation) to scan DMR regions (slop -b 150), resulting a motif-by-dmr dataset, can be loaded with DMR RegionDS
2. Determine region sets to run motif enrichment analysis min_cov = 5; quantile = 0.1 a. for each sample, with cov > min_cov filter, choose top and bottom quantile DMR as the sample-based hypo- and hyper-DMR, hypo-DMR status -1, hyper-DMR status 1, no status 0 b. for each dmr, with cov > min_cov filter, choose top and bottom quantile DMR as the sample-based hypo- and hyper-DMR, hypo-DMR status -1, hyper-DMR status 1, no status 0 c. combine hypo- and hyper- status from a and b, only save consistent status when a == b, otherwise save 0 d. sample can be all the L4Regions, and the SubCluster labels
3. For each set identified from 2, run motif enrichment with DME and CisTarget methods, take the consistent hypo- or hyper- enriched motifs
4. Combine all results to get cistrome of all TFs by all regions

After complete comments

CisTarget Hypo enrichment show good cell type specificity, hyper has little enrichment; DEM does not work well, motifs uniformly enriched, no clear cell type specificity. The final motif hits will be cistarget-hypo-DMR motif hits.

TF Example

tf_nes = dmr_ds["tf_nes"].to_pandas() # use NES per TF, agg from max(motifs) cemba.get_gene_fracs() # TF mCH frac

How to get results

dmr_ds = cemba.get_mc_dmr_ds(add_motif_hits=True)

# final TF-by-DMR hits (or adj matrix for the TF DMR graph)
dmr_ds['tf_gene_hits']

[rachelzeng98]

TF_GENE correlation

1. get l4region TF and gene expression matrix:

• a. get tf name from mm10.get_tf_gene_table()
• b. get mc fraction matrix by add_mc_frac to wmb.cemba.CEMBA_SNMC_CLUSTER_L4Region_SUM_ZARR_PATH
• c. output: tf-by_l4region matrix and gene_by_l4region matrix mc fraction matrix

2. correlation: 1838 TF, 4673 l4region and 30370 gene

• a. pearson correlation between each gene and each tf
• b. null distribution: shuffle sample order in tf-by_l4region matrix and do pearson correlation
• c. use a filter cutoff = 0.3 to judge if tf and gene have a connection
• c. output: tf_by_gene correlation matrix and adjacency matrix

How is the result

1. histplot showing pearsoncorrelation value distribution

   

2. scatterplot showing TF and Gene expression

   

3. TF and Gene expression plot on l4region clusters

   

How to get results combine all results to zarr: ecker-rachel-analysis/tf_gene/TF_Gene_Corelation.zarr