Skip to content

26-190

Latest

Choose a tag to compare

@cakirb cakirb released this 09 Jul 08:52

This release adds new QC and integration controls, plus resource tuning for large scAutoQC runs.

QC changes

  • CONS_FLOOR is no longer hardcoded in qc.py; it is now configurable via --consensus_floor / params.consensus_floor.
  • --consensus_floor controls the minimum consensus fraction used for consensus rescue/filtering.
  • The default value is 0.90, matching the previous hardcoded behaviour.
  • The selected consensus floor is stored in adata.uns["consensus_floor"].
  • --consensus_floor is validated to be between 0 and 1.
  • Multires cell-mode default percent_mito upper threshold is now 80.

Integration changes

  • Added --hvg_batch_key / params.hvg_batch_key for batch-aware HVG selection during integration.
  • When set, the value is passed to scanpy.pp.highly_variable_genes(batch_key=...).
  • The default is empty, preserving the previous non-batch-aware HVG behaviour.
  • The HVG batch key is validated before HVG selection starts, so missing columns fail early with a clear error.
  • Both cell_nuclei_union and global HVG strategies support the new batch-aware selection.

Resource tuning

Memory requests were increased based on a large run report:

  • pool_all: now requests 1.65 GB * numSamples, with a 10 GB floor.
  • finalize_qc and finalize_qc_basic: now request 16 GB + 6.5x input H5AD size.
  • integrate: now requests 18x input H5AD size, rounded up to the next 8 GB.

These changes should reduce memory-limit retries for large pooled datasets and integration runs.