R/model_cluster_id_by_feature.R

In momeara/MPStats: Tools for analyzing Morphological Profiling data

model_cluster_label_by_feature_lm <- function(cell_features, cluster_labels){
    data <- cell_features %>%
        dplyr::mutate(cluster_label = as.factor(cluster_labels$cluster_label))
    model <- lm(cluster_label ~ ., data=data)
}


# bstDense <- xgboost(
#     data = as.matrix(cell_features),
#     label=cluster_labels$cluster_label == 39,
#     max.depth = 2,
#     eta = 1,
#     nthread = 40,
#     nrounds = 2,
#     objective = "binary:logistic")


# bstDense <- xgboost(
#     data = as.matrix(cell_features),
#     label=cluster_labels$cluster_label == 2,
#     max.depth = 2,
#     eta = 1,
#     nthread = 40,
#     nrounds = 2,
#     objective = "binary:logistic")

#                                     Feature        Gain      Cover Frequency
# 1:     Intensity_IntegratedIntensity_Hoe_ER 0.837153462 0.34957503 0.1666667
# 2:             Texture_SumAverage_CMO_20_00 0.119862828 0.15042497 0.1666667
# 3:          Texture_Correlation_Lipids_8_00 0.022304826 0.04777903 0.1666667
# 4:          Texture_SumAverage_Hoe_ER_20_00 0.011423058 0.30179601 0.1666667
# 5:          Texture_Correlation_Hoe_ER_5_00 0.006264655 0.02315743 0.1666667
# 6: Intensity_IntegratedIntensityEdge_Hoe_ER 0.002991171 0.12726753 0.1666667

# importance_matrix <- xgb.importance(model = bstDense)
# print(importance_matrix)