optimize.constant.budget.restrictedDoublets: Optimizing cost parameters to maximize detection power for a...

In heiniglab/scPower: Experimental design framework for scRNAseq population studies (eQTL and DE)

Optimizing cost parameters to maximize detection power for a given budget and 10X design

Description

This function determines the optimal parameter combination for a given budget. The optimal combination is thereby the one with the highest detection power. Of the three parameters sample size, cells per sample and read depth, two need to be set and the third one is uniquely defined given the other two parameters and the overall budget.

Usage

optimize.constant.budget.restrictedDoublets(
  totalBudget,
  type,
  ct,
  ct.freq,
  costKit,
  costFlowCell,
  readsPerFlowcell,
  ref.study,
  ref.study.name,
  cellsPerLane,
  read.umi.fit,
  gamma.mixed.fits,
  disp.fun.param,
  nSamplesRange = NULL,
  nCellsRange = NULL,
  readDepthRange = NULL,
  mappingEfficiency = 0.8,
  multipletRate = 7.67e-06,
  multipletFactor = 1.82,
  min.UMI.counts = 3,
  perc.indiv.expr = 0.5,
  cutoffVersion = "absolute",
  nGenes = 21000,
  samplingMethod = "quantiles",
  multipletRateGrowth = "linear",
  sign.threshold = 0.05,
  MTmethod = "Bonferroni",
  useSimulatedPower = FALSE,
  simThreshold = 4,
  speedPowerCalc = FALSE,
  indepSNPs = 10,
  ssize.ratio.de = 1,
  reactionsPerKit = 6
)

Arguments

totalBudget	Overall experimental budget
type	(eqtl/de) study
ct	Cell type of interest (name from the gamma mixed models)
ct.freq	Frequency of the cell type of interest
costKit	Cost for one 10X kit
costFlowCell	Cost of one flow cells for sequencing
readsPerFlowcell	Number reads that can be sequenced with one flow cell
ref.study	Data frame with reference studies to be used for expression ranks and effect sizes (required columns: name (study name), rank (expression rank), FoldChange (DE study) /Rsq (eQTL study))
ref.study.name	Name of the reference study. Will be checked in the ref.study data frame for it (as column name).
cellsPerLane	Maximal number of cells per 10X lane
read.umi.fit	Data frame for fitting the mean UMI counts per cell depending on the mean readds per cell (required columns: intercept, reads (slope))
gamma.mixed.fits	Data frame with gamma mixed fit parameters for each cell type (required columns: parameter, ct (cell type), intercept, meanUMI (slope))
disp.fun.param	Function to fit the dispersion parameter dependent on the mean (required columns: ct (cell type), asymptDisp, extraPois (both from taken from DEseq))
nSamplesRange	Range of sample sizes that should be tested (vector)
nCellsRange	Range of cells per individual that should be tested (vector)
readDepthRange	Range of read depth values that should be tested (vector)
mappingEfficiency	Fraction of reads successfully mapped to the transcriptome in the end (need to be between 1-0)
multipletRate	Expected increase in multiplets for additional cell in the lane
multipletFactor	Expected read proportion of multiplet cells vs singlet cells
min.UMI.counts	Expression cutoff in one individual: if cutoffVersion=absolute, more than this number of UMI counts for each gene per individual and cell type is required; if cutoffVersion=percentage, more than this percentage of cells need to have a count value large than 0
perc.indiv.expr	Expression cutoff on the population level: if number < 1, percentage of individuals that need to have this gene expressed to define it as globally expressed; if number >=1 absolute number of individuals that need to have this gene expressed
cutoffVersion	Either "absolute" or "percentage" leading to different interpretations of min.counts (see description above)
nGenes	Number of genes to simulate (should match the number of genes used for the fitting)
samplingMethod	Approach to sample the gene mean values (either taking quantiles or random sampling)
multipletRateGrowth	Development of multiplet rate with increasing number of cells per lane, "linear" if overloading should be modeled explicitly, otherwise "constant". The default value for the parameter multipletRate is matching the option "linear".
sign.threshold	Significance threshold
MTmethod	Multiple testing correction method (possible options: "Bonferroni","FDR","none")
useSimulatedPower	Option to simulate eQTL power for small mean values to increase accuracy (only possible for eQTL analysis)
simThreshold	Threshold until which the simulated power is taken instead of the analytic (only for the eQTL analysis)
speedPowerCalc	Option to speed power calculation by skipping all genes with an expression probability less than 0.01 (as overall power is anyway close to 0)
indepSNPs	Number of independent SNPs assumed for each loci (for eQTL Bonferroni multiple testing correction the number of tests are estimated as number expressed genes * indepSNPs)
ssize.ratio.de	In the DE case, ratio between sample size of group 0 (control group) and group 1 (1=balanced design)
reactionsPerKit	Reactions (=lanes) per kit, defines the total number of tested individuals per kit

Value

Data frame with overall detection power, power and expression power for each possible parameter combination given the budget and the parameter ranges

heiniglab/scPower documentation built on Jan. 9, 2025, 12:13 p.m.