-
Notifications
You must be signed in to change notification settings - Fork 3
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Merge pull request #35 from spriyansh/main
Object Updates with Link, Diagnostics and Offset
- Loading branch information
Showing
16 changed files
with
330 additions
and
13 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,235 @@ | ||
| #' @title Plot Model Diagnostics | ||
| #' | ||
| #' @description | ||
| #' Plots model residuals for the requested gene. Can be used for model | ||
| #' diagnostics of optimized model or full model. | ||
| #' | ||
| #' @importFrom stats residuals fitted rstandard predict | ||
| #' @importFrom stringr str_remove | ||
| #' | ||
| #' @param scmpObj An object of class \code{\link{ScMaSigPro}}. | ||
| #' @param feature_id Name of the gene to be plotted. | ||
| #' @param model Type of model to be used for plotting. Can be either 'optimized', | ||
| #' 'intercept'and 'full'. Default is 'optimized'. | ||
| #' | ||
| #' @return ggplot2 plot object. | ||
| #' | ||
| #' @author Priyansh Srivastava \email{spriyansh29@@gmail.com} | ||
| #' | ||
| #' @export | ||
| #' | ||
| plotDiagnostics <- function(scmpObj, | ||
| feature_id, | ||
| model = "optimized") { | ||
| # Check if the results exist | ||
| assert_that(!isEmpty(scmpObj@Significant@genes), | ||
| msg = paste("No Significant genes found, please run the workflow first") | ||
| ) | ||
|
|
||
| # Check model type | ||
| assert_that(all(model %in% c("optimized", "full", "intercept")), | ||
| msg = paste("The requested gene is not available in the significant genes list") | ||
| ) | ||
|
|
||
| # Get all the available genes | ||
| avail_genes <- unique(unlist(scmpObj@Significant@genes)) | ||
|
|
||
| # Check if the requested gene is availble as singificant gene | ||
| assert_that(all(feature_id %in% avail_genes), | ||
| msg = paste("The requested gene is not available in the significant genes list") | ||
| ) | ||
|
|
||
| # Get the counts for the genes | ||
| y_mat <- as.matrix(scmpObj@Dense@assays@data$bulk.counts) | ||
|
|
||
| # Extract gene counts | ||
| y_df <- y_mat[feature_id, , drop = TRUE] %>% | ||
| as.data.frame() | ||
| colnames(y_df) <- "y" | ||
|
|
||
| # Set theme for the plot | ||
| theme_set <- theme_minimal(base_size = 12) + | ||
| theme( | ||
| legend.position = "bottom", | ||
| panel.grid.major = element_line( | ||
| color = "grey90", linewidth = 0.3, linetype = "dashed" | ||
| ), | ||
| panel.grid.minor = element_blank() | ||
| ) | ||
|
|
||
| # Based on model | ||
| if (model == "full" || model == "intercept") { | ||
| # Create Model input | ||
| regression_matrix <- scmpObj@Design@predictor_matrix | ||
| } else { | ||
| # Get coefficients | ||
| coeff_matrix <- showCoeff(scmpObj) %>% | ||
| data.frame() | ||
|
|
||
| # Subset | ||
| coeff_matrix_sub <- coeff_matrix[feature_id, , drop = FALSE] | ||
| all_terms <- colnames(coeff_matrix_sub)[-1] | ||
|
|
||
| # Remove any occurance of beta | ||
| all_terms <- str_remove(string = all_terms, pattern = "beta") | ||
|
|
||
| # Remove intercept | ||
| all_terms <- all_terms[-1] | ||
|
|
||
| # Get the terms | ||
| terms <- coeff_matrix_sub[apply(coeff_matrix_sub, 2, FUN = function(i) { | ||
| i != 0 | ||
| })] | ||
|
|
||
| # Extract the prediction | ||
| prediction_matrix <- scmpObj@Design@predictor_matrix | ||
|
|
||
| # Create Model input | ||
| regression_matrix <- prediction_matrix[, all_terms, drop = FALSE] | ||
| } | ||
|
|
||
| # Combine regression matrix with counts | ||
| model_df <- cbind(y_df, regression_matrix) | ||
|
|
||
| # Extract offsets | ||
| if (scmpObj@Parameters@offset) { | ||
| offset_data <- scmpObj@Design@offset | ||
| } else { | ||
| offset_data <- NULL | ||
| } | ||
|
|
||
| if (model == "intercept") { | ||
| glm_s3 <- glm(y ~ 1, | ||
| data = as.data.frame(model_df), | ||
| family = scmpObj@Parameters@distribution, | ||
| epsilon = scmpObj@Parameters@epsilon, | ||
| offset = offset_data, | ||
| weights = NULL, | ||
| maxit = scmpObj@Parameters@max_it | ||
| ) | ||
| } else { | ||
| # Fit Model | ||
| glm_s3 <- glm(y ~ ., | ||
| data = as.data.frame(model_df), | ||
| family = scmpObj@Parameters@distribution, | ||
| epsilon = scmpObj@Parameters@epsilon, | ||
| offset = offset_data, | ||
| weights = NULL, | ||
| maxit = scmpObj@Parameters@max_it | ||
| ) | ||
| } | ||
| # Extract residuals | ||
| raw_residuals <- residuals(glm_s3, type = "response") | ||
| pearson_residuals <- residuals(glm_s3, type = "pearson") | ||
| deviance_residuals <- residuals(glm_s3, type = "deviance") | ||
| standardized_residuals <- rstandard(glm_s3) | ||
|
|
||
| # Extract fitted values | ||
| fitted_values <- fitted(glm_s3) | ||
|
|
||
| # Calculate predicted responses | ||
| predicted_responses <- predict(glm_s3, type = "response") | ||
|
|
||
| # Q-Q plot | ||
| # This plot of the standardized residuals allows you to check if the | ||
| # residuals are approximately normally distributed, which is an | ||
| # Calculate standardized residuals | ||
|
|
||
| # Create a dataframe for plotting | ||
| qq_df <- data.frame(std_residuals = standardized_residuals) | ||
|
|
||
| # Create the Normal Q-Q plot | ||
| qq_plot <- ggplot(qq_df, aes(sample = .data$std_residuals)) + | ||
| stat_qq(color = "#EE446F", size = 2) + | ||
| stat_qq_line( | ||
| color = "#15918A", alpha = 1, | ||
| linewidth = 1, linetype = "solid" | ||
| ) + | ||
| labs( | ||
| subtitle = feature_id, | ||
| title = "A. Normal Q-Q Plot of Standardized Residuals", | ||
| x = "Theoretical Quantiles", | ||
| y = "Standardized Residuals" | ||
| ) + | ||
| theme_set | ||
|
|
||
| # Residuals vs Fitted Plot: | ||
| # This plot helps to detect non-linearity, unequal error variances, | ||
| # and outliers. It is used to check the assumption of homoscedasticity. | ||
| # Assuming your glm model is stored in 'glm_s3' | ||
|
|
||
| # Create a dataframe for plotting | ||
| res_vs_fit_df <- data.frame( | ||
| Fitted = fitted_values, | ||
| StdResiduals = standardized_residuals | ||
| ) | ||
|
|
||
| # Create the Residuals vs Fitted plot | ||
| stdRes_vs_fitted_plot <- ggplot(res_vs_fit_df, aes(x = .data$Fitted, y = .data$StdResiduals)) + | ||
| geom_point(color = "#EE446F", size = 2) + # Plot the points | ||
| geom_hline(yintercept = 0, linetype = "dashed", color = "#9F7BB8") + # Add a horizontal line at 0 | ||
| labs( | ||
| subtitle = feature_id, | ||
| title = "B. Standardized Residuals vs Fitted Plot", x = "Fitted Values", y = "Standardized Residuals" | ||
| ) + | ||
| theme_set + | ||
| geom_smooth( | ||
| method = "loess", color = "#15918A", alpha = 0.4, formula = y ~ x, | ||
| linewidth = 1, linetype = "solid", se = TRUE, fill = "#FDC659" | ||
| ) # Add a loess smoothed line | ||
|
|
||
| # Scale-Location Plot (or Spread-Location Plot): This shows if residuals | ||
| # are spread equally along the ranges of predictors. This is a way to check | ||
| # the assumption of equal variance (homoscedasticity). | ||
| # Calculate square root of absolute standardized residuals | ||
| sqrt_abs_std_residuals <- sqrt(abs(standardized_residuals)) | ||
|
|
||
| # Create a dataframe for plotting | ||
| scale_loc_df <- data.frame( | ||
| Fitted = fitted_values, | ||
| SqrtAbsStdResiduals = sqrt_abs_std_residuals | ||
| ) | ||
|
|
||
| # Create the Scale-Location plot | ||
| scale_loc_plot <- ggplot(scale_loc_df, aes(x = .data$Fitted, y = .data$SqrtAbsStdResiduals)) + | ||
| geom_point(color = "#EE446F", size = 2) + # Plot the points | ||
| geom_smooth( | ||
| method = "loess", color = "#15918A", formula = y ~ x, | ||
| linewidth = 1, linetype = "solid", se = TRUE, | ||
| fill = "#FDC659", alpha = 0.4 | ||
| ) + | ||
| labs( | ||
| subtitle = feature_id, | ||
| title = "C. Scale-Location Plot", | ||
| x = "Fitted Values", | ||
| y = "Sqrt of Absolute Standardized Residuals" | ||
| ) + | ||
| theme_set | ||
|
|
||
| # A Dfun plot is less standard and typically refers to a plot that can help | ||
| # assess the appropriateness of the link function in a GLM. This might involve | ||
| # plotting observed versus expected responses or derivatives of the link function. | ||
| # The exact implementation can depend on the specific family and link function used | ||
| # in your model. For a basic implementation, you might plot observed vs. predicted responses. | ||
|
|
||
| # Create a dataframe for plotting | ||
| dfun_df <- data.frame(Observed = model_df$y, Predicted = predicted_responses) | ||
|
|
||
| # Create the Dfun plot | ||
| defunc_plot <- ggplot(dfun_df, aes(x = .data$Observed, y = .data$Predicted)) + | ||
| geom_point(color = "#EE446F", size = 2) + | ||
| geom_abline( | ||
| intercept = 0, slope = 1, linetype = "solid", color = "#15918A", | ||
| linewidth = 1, alpha = 1 | ||
| ) + | ||
| labs( | ||
| subtitle = feature_id, | ||
| title = "D. Observed vs Predicted", | ||
| x = "Observed", | ||
| y = "Predicted" | ||
| ) + | ||
| theme_set | ||
|
|
||
| # Return Plots | ||
| return((qq_plot + stdRes_vs_fitted_plot) / (scale_loc_plot + defunc_plot)) | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.