envest / ballad Goto Github PK

Documentation:

• Preamble
• Functions

Error handling:

• Check inputs

Include a value when missing ALT in extract_variant_sites_liftover.py

VEP returns more than one prediction per variant. Pick the best one (eg the one on canonical transcript) and select most important information. Full annotation may be stored separately.

Question: run within filter_loom.py or after the fact

Given a set of SNPs, determine genotype of a sample at each SNP. Take coverage variability and allele dropout into consideration. Use this as input to unpooling loom files.

Currently output file has column names in alphabetical order (from the list of loom file row attribute keys). The order could be more meaningful, but would have to be pre-specified / harder to adapt to future file formats.

Try using loompy.combine along a shared axis (GRCh38 variant) to get everything in the same order, rather than loop through each variant

Add a command line flag for a file with list of variants or cell barcodes to force to be included in the filtered output

--force_include_variants file.tsv --force_include_cell_barcodes file.tsv

Flat variants + barcodes file with so much repeated info takes lots of time to write. Maybe just output the filtered loom file? Possibly even just the indexes of kept variants and barcodes

Set sensible defaults, but let user specify non-default values for each parameter one by one or as a set (via parameter file) which overrides all other given parameters. Then check that each parameter makes sense. Force explicit overwrite

• Need to know consensus genotypes for each patient at a set of SNPs
• For each cell, measure how similar the available genotypes are to known consensus profiles
• If one is clearly more similar, assign barcode to that patient
• Output unpooled loom files

Use ensemble-vep to predict effect of each variant in a loom file. Given a loom file, extract the variant sites to use as input to VEP.

Document with additional directions for downloading VEP resources in README

Run timing tests to see what steps are slowest, and improve. Try to make everything run in < 1 minute.