Fix tcga heatmap

[jyaacoub]

TCGA heatmap has a couple issues that need to be addressed:

• [ ] Make sure we are using the right TCGA MAF file (i.e: not CRCh37)
• [ ] Filtering out mutations that do not impact MANE transcripts
• [ ] Merge TCGA with biomart file on Transcript_ID for MANE to get the gene length for normalizing counts by
  • [ ] instead of normalizing by average gene length.

Scope

Everything from how to load up TCGA, adding cancer type info, and plotting figures should be available in https://github.com/jyaacoub/MutDTA/blob/main/tcga_analysis.py.

Relevant functions are:

• load_TCGA for loading TCGA and adding cancer type info
• normalized_heatmap for plotting the normalized heatmap

Feel free to modify whatever you need to, everything is already tracked with git so even if you delete the whole file it wont matter.

Files needed:

• [ ] TCGA MAF - I used this but is outdated and uses CRCh37.
• [ ] tcga_code_tables: bcr and tss are needed for getting cancer types from barcodes. Download all code tables from: https://gdc.cancer.gov/system/files/public/file/tcga_code_tables.zip
• Dataset csvs (davis, kiba, pdbbind) are not needed to fix this issue but you can download the test set from https://github.com/jyaacoub/MutDTA/tree/main/splits
• [ ] Biomart file for merging with TCGA to get gene length.
• [ ] MANE csv for filtering out entries in TCGA that do not relate to a MANE transcript.

[jyaacoub]

Relevant links

• DeepDTA article (see section 2.1 for info on how sequences were grabbed): https://academic.oup.com/bioinformatics/article/34/17/i821/5093245?login=true
  • GitHub page - https://github.com/hkmztrk/DeepDTA/blob/master/data/README.md
• PDBbind dataset -> http://www.pdbbind.org.cn/
• KIBA dataset (see supplementary info) -> https://pubs.acs.org/doi/10.1021/ci400709d
• Our test splits: https://raw.githubusercontent.com/jyaacoub/MutDTA/main/splits/all_tests.csv
  • code column are the protein IDs that they provided, prot_id are uniprot ids (if available), and gene_name is the mapped gene name using biomart (for davis its just a duplicate of code)