scran_norm
Scaling normalization of single-cell data
|
Options for sanitize_size_factors()
.
More...
#include <sanitize_size_factors.hpp>
Public Attributes | |
SanitizeAction | handle_zero = SanitizeAction::ERROR |
SanitizeAction | handle_negative = SanitizeAction::ERROR |
SanitizeAction | handle_nan = SanitizeAction::ERROR |
SanitizeAction | handle_infinite = SanitizeAction::ERROR |
Options for sanitize_size_factors()
.
SanitizeAction scran_norm::SanitizeSizeFactorsOptions::handle_zero = SanitizeAction::ERROR |
How should we handle size factors of zero?
If set to SANITIZE
, they will be automatically set to the smallest valid size factor (or 1, if all size factors are invalid). This ensures that the normalized values will be large to reflect the extremity of the scaling, but still finite for sensible downstream processing.
In most cases, size factors of zero are only obtained for cells with all-zero counts. By replacing the size factor with a finite value, we ensure that any all-zero cells are represented by all-zero columns in the normalized matrix, which is a reasonable outcome if those cells cannot be filtered out during upstream quality control.
SanitizeAction scran_norm::SanitizeSizeFactorsOptions::handle_negative = SanitizeAction::ERROR |
How should we handle negative size factors?
If set to SANITIZE
, they will be automatically set to the smallest valid size factor (or 1, if all size factors are invalid), following the same logic as SanitizeSizeFactorsOptions::handle_zero
.
SanitizeAction scran_norm::SanitizeSizeFactorsOptions::handle_nan = SanitizeAction::ERROR |
How should we handle NaN size factors?
If set to `SANITIZE, NaN size factors will be automatically set to 1, meaning that scaling is a no-op.
SanitizeAction scran_norm::SanitizeSizeFactorsOptions::handle_infinite = SanitizeAction::ERROR |
How should we handle infinite size factors?
If set to SANITIZE
, infinite size factors will be automatically set to the largest valid size factor (or 1, if all size factors are invalid). This ensures that any normalized values will be, at least, finite; the choice of a relatively large replacement value reflects the extremity of the scaling.