assess_immune_response: Assess easier score as predictor of patients' immune response

In olapuentesantana/easier: Estimate Systems Immune Response from RNA-seq data

Assess easier score as predictor of patients' immune response

Description

Evaluates the predictive performance of easier score as predictor of patients' immune response. This is done for each quantitative descriptor, an ensemble descriptor based on the average of the individual ones, and the gold standard scores. If provided, tumor mutational burden (TMB) is also used as predictor for comparison. Since both immune response and TMB are essential for effective immunotherapy response, an integrated score is calculated given two different approaches based on a applying either a weighted average or penalty to patients' easier score depending on their TMB category.

Usage

assess_immune_response(
  predictions_immune_response = NULL,
  patient_response = NULL,
  RNA_tpm = NULL,
  select_gold_standard = NULL,
  TMB_values = NULL,
  easier_with_TMB = "none",
  weight_penalty = NULL,
  verbose = TRUE
)

Arguments

predictions_immune_response	list containing the predictions for each quantitative descriptor and for each task. This is the output from predict_immune_response.
patient_response	character vector with two factors (Non-responders = NR, Responders = R).
RNA_tpm	numeric matrix of patients' gene expression data as tpm values.
select_gold_standard	character string with names of scores of immune response to be computed. Default scores are computed for: "CYT", "Roh_IS", "chemokines", "Davoli_IS", "IFNy", "Ayers_expIS", "Tcell_inflamed", "RIR", "TLS".
TMB_values	numeric vector containing patients' tumor mutational burden (TMB) values.
easier_with_TMB	character string indicating which approach should be used to integrate easier with TMB. If TMB_values provided, one of "weighted_average" (default) or "penalized_score". If TMB_values not provided, default is "none".
weight_penalty	integer value from 0 to 1, which is used to define the weight or penalty for combining easier and TMB scores based on a weighted average or penalized score, in order to derive a score of patient's likelihood of immune response. The default value is 0.5.
verbose	logical flag indicating whether to display messages about the process.

Value

When patient_response is provided, a roc curve plot and a bar plot that displays the average (across tasks) area under the ROC curve (AUC) values is returned. If patient_response is not provided, the easier score is represented as box plots (10 tasks) for each patient.

When patient_response is provided and easier_with_TMB = weighted_average or easier_with_TMB = penalized_score, an scatter plot that shows the AUC values of the integrated approach, easier score and TMB is returned. If in this case, patient_response is not provided, the integrated score is represented as a dot plot for each patient.

Examples

# using a SummarizedExperiment object
library(SummarizedExperiment)
# Using example exemplary dataset (Mariathasan et al., Nature, 2018)
# from easierData. Original processed data is available from
# IMvigor210CoreBiologies package.
library("easierData")

dataset_mariathasan <- easierData::get_Mariathasan2018_PDL1_treatment()
RNA_tpm <- assays(dataset_mariathasan)[["tpm"]]
cancer_type <- metadata(dataset_mariathasan)[["cancertype"]]

# Select a subset of patients to reduce vignette building time.
pat_subset <- c(
  "SAM76a431ba6ce1", "SAMd3bd67996035", "SAMd3601288319e",
  "SAMba1a34b5a060", "SAM18a4dabbc557"
)
RNA_tpm <- RNA_tpm[, colnames(RNA_tpm) %in% pat_subset]

# Computation of TF activity (Garcia-Alonso et al., Genome Res, 2019)
tf_activities <- compute_TF_activity(
  RNA_tpm = RNA_tpm
)

# Predict patients' immune response
predictions <- predict_immune_response(
  tfs = tf_activities,
  cancer_type = cancer_type,
  verbose = TRUE
)

# retrieve clinical response
patient_ICBresponse <- colData(dataset_mariathasan)[["BOR"]]
names(patient_ICBresponse) <- colData(dataset_mariathasan)[["pat_id"]]

# retrieve TMB
TMB <- colData(dataset_mariathasan)[["TMB"]]
names(TMB) <- colData(dataset_mariathasan)[["pat_id"]]

patient_ICBresponse <- patient_ICBresponse[names(patient_ICBresponse) %in% pat_subset]
TMB <- TMB[names(TMB) %in% pat_subset]

# Assess patient-specific likelihood of response to ICB therapy
output_eval_with_resp <- assess_immune_response(
  predictions_immune_response = predictions,
  patient_response = patient_ICBresponse,
  RNA_tpm = RNA_tpm,
  select_gold_standard = "IFNy",
  TMB_values = TMB,
  easier_with_TMB = "weighted_average",
)


RNA_counts <- assays(dataset_mariathasan)[["counts"]]
RNA_counts <- RNA_counts[, colnames(RNA_counts) %in% pat_subset]

# Computation of cell fractions  (Finotello et al., Genome Med, 2019)
cell_fractions <- compute_cell_fractions(RNA_tpm = RNA_tpm)

# Computation of pathway scores (Holland et al., BBAGRM, 2019;
# Schubert et al., Nat Commun, 2018)
pathway_activities <- compute_pathway_activity(
  RNA_counts = RNA_counts,
  remove_sig_genes_immune_response = TRUE
)

# Computation of ligand-receptor pair weights
lrpair_weights <- compute_LR_pairs(
  RNA_tpm = RNA_tpm,
  cancer_type = "pancan"
)

# Computation of cell-cell interaction scores
ccpair_scores <- compute_CC_pairs(
  lrpairs = lrpair_weights,
  cancer_type = "pancan"
)

# Predict patients' immune response
predictions <- predict_immune_response(
  pathways = pathway_activities,
  immunecells = cell_fractions,
  tfs = tf_activities,
  lrpairs = lrpair_weights,
  ccpairs = ccpair_scores,
  cancer_type = cancer_type,
  verbose = TRUE
)

# Assess patient-specific likelihood of response to ICB therapy
output_eval_with_resp <- assess_immune_response(
  predictions_immune_response = predictions,
  patient_response = patient_ICBresponse,
  RNA_tpm = RNA_tpm,
  TMB_values = TMB,
  easier_with_TMB = "weighted_average",
)

olapuentesantana/easier documentation built on Feb. 25, 2024, 3:39 p.m.