Citation

If you use funscoR please cite:

Ochoa et al. The functional landscape of the human phosphoproteome bioRxiv (2019). https://doi.org/10.1101/541656

Installation

First, install funscoR from github. It requires to have devtools installed.

devtools::install_github("evocellnet/funscoR")

Getting started

To get started, fire up the packages and load some sample data.

## Load the required packages
library(funscoR)
library(knitr)
library(dplyr)
library(stringr)

Datasets

phosphoproteome %>%
  head() %>%
  kable()

psp %>%
  head() %>%
  kable()

Phosphoproteome annotation

An extensive functional annotation for the human phosphoproteome is provided as part of this package. Next you can find a list of all the annotations available.

data(package = "funscoR")$results %>% 
  as.data.frame() %>% 
  filter(str_detect(Item, "feature")) %>% 
  select(Item) %>%
  kable()

Each of the annotation objects can be used indepedently. For example, it’s described next the object that contains the ancestral reconstruction of all available phosphosites. The column w0_mya contains the inferred age of the last common ancestor for the phosphosite. w3_mya contains the equivalent information using a window of +/-3 residues to asses the conservation. More information about the dataset can be found using using ?feature_ptmdb_age.

feature_ptmdb_age %>%
  head() %>%
  kable()

In order to train a model, you might be interested in annotating the phosphoproteome with all the available features. You can use the annotate_sites function for this.

## annotate phosphoproteome with features
annotated_phos <- annotate_sites(phosphoproteome)

Model training

A preprocessing step must be run to ensure the features are properly provided to the model. The function preprocess_features defaults to a series of methods but additional tunning can be applied using the methods= and features_to_exclude= arguments. Different preprocessing steps are necessary for “ST” and “Y” residues, as some of the features are exclusive to each of the sets.

## preprocess features for training
ST_features <- preprocess_features(annotated_phos, "ST")
Y_features <- preprocess_features(annotated_phos, "Y")

Once the features are ready, a model can be trained using a provided gold standard. The default algorithm is a Gradient Boosting Machine with a series of hyperparameters optimized to the default set. Different algorithms and settings can be provided using the parameters= argument. The training process can be parallelized using the doParallel package if the ncores parameter exceeds 1.

## train new model
ST_model <- train_funscore(ST_features, "ST", psp, ncores = 4)
Y_model <- train_funscore(Y_features, "Y", psp, ncores = 4)

Predicting functional scores

Given an annotated phosphoproteome with preprocessed features and a trained model, new functional scores can be predicted for “ST” and “Y” separately.

## predict funcscoR for all sites
ST_scores <- predict_funscore(ST_features, ST_model, ncores = 4)
Y_scores <- predict_funscore(Y_features, Y_model, ncores = 4)

## gather all predictions
all_scores <- bind_rows(ST_scores, Y_scores) %>%
  mutate(probabilities = log_scaling(probabilities))