A project from Summer 2018 that compared the utility of newer ChIP methods ChIP-exo and ChIP-nexus on its effects on Segway's ability to generate annotations from this data of higher signal to noise ratio and near single base pair resolution.
- Data Cleaning (data_preprocessing/)
- Download raw data from ENCODE project and NCBI SRA (getSRR.sh)
- Convert to bedgraph and sort data based on threshold cut off (fq_to_bam.sh)
- QC with PhantomPeakQualTools + peak calling with MACS2 to generate bedgraphs (MACS2/ & PhantomPeakQualTools/)
- Store data in genomedata archive (bedgraph_to_genomedata.sh)
- Run Segway (segway/)
- Training then identification rounds (trainsegway.sh & annotate.sh)
- Set minibatch training of 10 round on 1% of the genome
- Try with 5 different resolutions: 100bp, 1bp, 2bp, 30bp, 50bp
- Analyze Results
- Recolour segway annotations with 10 different colours for visualization in genome browser (segway/)
- Run stable marriage Hungarian algorithm on annotations (stablemarriage/)
- Graph heatmaps and bipartite graphs in R. Account for negative/NaNs by adding pseudocount (LOD/2) to all data points prior to normalizing (pseudocount/)
- All graph generating functions in datavisualization.R file, generated some in R notebook (can find in lab notebook)
- Miscellaneous
- Script to clean up on the cluster (segway/seg_cleanup.sh)
- Alternate attempt at finding the LOD via the genomedata archives that were already generated (segway/runthroughcoords.py)
- Get average counts from bedgraph file (data_preprocessing/getavg.sh)
- Optional conversion from bigwig to wiggle format (data_preprocessing/bigwig_to_wiggle.sh)
Lab Notebook Final Presentation
###Acknowledgements The student researcher would like to thank Dr. Hoffman for the opportunity + resources and Francis Nguyen for mentorship. Special thanks to Coby, Sam, and Davide of the Hoffman Lab as well.