Fill missing responses with predictions from per-subject linear models

Completes the data to every subject x every sample and fills each missing response cell with predict() of that subject's fitted lm, on the response scale. Observed values are left unchanged. The imputed values carry no noise.

Usage

impute_from_model(model, lfqdata)

Arguments

model: a Model with lm fits, as returned by build_model or build_model_impute
lfqdata: the LFQData the model was fitted on (same response and subject_id)

Value

list with lfqdata, a new LFQData holding every subject x sample row with missing responses filled, and summary, a tibble with the subject_id columns and n_observed, n_imputed and route per subject

Details

Each subject gets one route, checked in this order:

complete: no missing cell.
lod_refit: the subject was refitted at the LOD by build_model_impute (model_df$imputed); that fit's predictions fill the missing cells.
fitted: the observed-data fit has no NA coefficient and knows every factor level of the missing cells; its predictions fill them.
none: no usable fit; the missing cells stay NA.

Other modelling: AnovaExtractor, Contrasts, ContrastsDEqMSFacade, ContrastsDEqMSVoomFacade, ContrastsFacadeBase, ContrastsFirth, ContrastsFirthFacade, ContrastsFirthNestedFacade, ContrastsLMFacade, ContrastsLMImputeFacade, ContrastsLMMissingFacade, ContrastsLimma, ContrastsLimmaFacade, ContrastsLimmaImputeFacade, ContrastsLimmaVoomFacade, ContrastsLimmaVoomImputeFacade, ContrastsLimpaFacade, ContrastsLimpaNestedFacade, ContrastsLmerNestedFacade, ContrastsMissing, ContrastsModerated, ContrastsModeratedDEqMS, ContrastsPlotter, ContrastsRLMFacade, ContrastsROPECA, ContrastsROPECANestedFacade, ContrastsRfitFacade, ContrastsRfitImputeFacade, ContrastsTable, INTERNAL_FUNCTIONS_BY_FAMILY, LR_test(), Model, ModelFirth, ModelLimma, StrategyLM, StrategyLimma, StrategyLimpa, StrategyLmer, StrategyLogistf, StrategyRLM, StrategyRfit, build_contrast_analysis(), build_model(), build_model_glm_peptide(), build_model_glm_protein(), build_model_impute(), build_model_limma(), build_model_limma_impute(), build_model_limma_voom(), build_model_limma_voom_impute(), build_model_limpa(), build_model_logistf(), compute_borrowed_variance(), compute_borrowed_variance_limma(), compute_contrast(), compute_lmer_contrast(), df.residual.rfit_prolfqua(), get_anova_df(), get_complete_model_fit(), get_p_values_pbeta(), group_label(), impute_refit_singular(), linfct_from_model(), linfct_matrix_contrasts(), list_facades(), lookup_facade(), merge_contrasts_results(), model_analyse(), model_summary(), moderated_p_deqms(), moderated_p_deqms_long(), moderated_p_limma(), new_imputed_model(), pivot_model_contrasts_to_wide(), register_facade(), sigma.rfit_prolfqua(), sim_build_models_lm(), sim_build_models_lmer(), sim_build_models_logistf(), sim_make_model_lm(), sim_make_model_lmer(), strategy_limma(), strategy_limpa(), strategy_logistf(), summary_ROPECA_median_p.scaled(), vcov.rfit_prolfqua()

Examples

istar <- sim_lfq_data_protein_config(Nprot = 30, weight_missing = 0.5)
#> creating sampleName from file_name column
#> completing cases
#> completing cases done
#> setup done
lfqdata <- LFQData$new(istar$data, istar$config)
strat <- strategy_lm(paste(lfqdata$response(), "~ group_"))
mod <- build_model_impute(lfqdata, strat)
res <- impute_from_model(mod, lfqdata)
table(res$summary$route)
#> 
#>  complete    fitted lod_refit 
#>         8        16         6 
n_missing <- function(x) sum(is.na(x$data_long()$abundance))
stopifnot(n_missing(res$lfqdata) <= n_missing(lfqdata))