In berkgurdamar/predatoR: Mutation Impact Prediction Based on Network Properties

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "inst/extdata",
  out.width = "100%"
)

predatoR: Mutation Impact Prediction Based on Network Properties

Overview

predatoR is a tool for mutation impact prediction based on network properties.

predatoR() function is the wrapper function of predatoR package.

predatoR() works on each PDB respectively. For each PDB;

Downloads/Reads PDB file
Calculates distance between each atom in the structure
Creates a network from PDB
Calculates Degree Centrality Z-Score of each atom
Calculates Eigen Centrality Z-Score of each atom
Calculates Shortest Path Z-Score of each atom
Calculates Betweenness Z-Score of each atom
Calculates Clique Z-Score of each atom
Calculates PageRank Z-Score of each atom
Gets gnomAD Synonymous Z-Score, Missense Z-Score, and pLI Score
Gets BLOSUM62 score of the mutation
Finds the number of KEGG Pathways which contains the input gene
Gets Genic Intolerance Score
Finds the number of GO terms associated with the input gene
Finds the number of diseases associated with the input gene from DisGeNET
Gets the Gene Essentiality Score from Online Gene Essentiality (OGEE) Database
Calculates the median gene expression value of the input gene from GTEx
Calculates 6 different features from Accessible Surface Area and Hydrophobicity Scale of reference and mutant amino acids
Makes prediction

Installation

You can install the predatoR via devtools:

``` {r installation_guide, eval=FALSE} library(devtools) install_github("berkgurdamar/predatoR")

# Usage

Mutation impact prediction can be done via `predatoR()` function:

`predatoR()` uses data.frame structures as an input. data.frame should consist of __'PDB_ID'__, __'Chain'__, __'Position'__, __'Orig_AA'__, __'Mut_AA'__ and __'Gene_Name'__ (optional). Predictions can be made by using 2 different models, 5 Angstrom (Å)-all atoms model and 7Å-carbon alpha (Cα) atoms only model.

```r
test_data <- as.data.frame(rbind(c("3SQJ", "A", 196, "GLN", "LEU", "ALB"),
                               c("3SQJ", "A", 396, "GLU", "LYS", "ALB")))
colnames(test_data) <- c("PDB_ID", "Chain", "Position", "Orig_AA", "Mut_AA", "Gene_Name")
knitr::kable(test_data, align = c("c", "c", "c", "c", "c", "c"))

predatoR() can work with input which has partially included gene names.

library(predatoR)
# Gene name included
input_df <- as.data.frame(rbind(c("3SQJ", "A", 196, "GLN", "LEU", "ALB"),
                               c("3SQJ", "A", 396, "GLU", "LYS", "ALB")))
pred_res <- predatoR(info_df =  input_df, n_threads = 8, gene_name_info = TRUE)
# Gene name not included
input_df <- as.data.frame(rbind(c("3SQJ", "A", 196, "GLN", "LEU"),
                               c("3SQJ", "A", 396, "GLU", "LYS")))
pred_res <- predatoR(info_df =  input_df, n_threads = 8, gene_name_info = FALSE)
# Partially included gene names
input_df <- as.data.frame(rbind(c("3SQJ", "A", 196, "GLN", "LEU", "ALB"),
                               c("3SQJ", "A", 396, "GLU", "LYS", "")))
pred_res <- predatoR(info_df =  input_df, n_threads = 8, gene_name_info = TRUE)

predatoR() function returns a data.frame which contains additional two columns; 'Prediction' and 'Probability'. 'Prediction' represents the result of the impact prediction and 'Probability' represents the probability that the mutation classified as Pathogenic or Neutral.

example_result <- as.data.frame(rbind(c("3SQJ", "A", "196", "GLN", "LEU", "ALB", "Neutral", "0.6521832"),
                                      c("3SQJ", "A", "396", "GLU", "LYS", "ALB", "Neutral", "0.6009792")))
colnames(example_result) <- c("PDB_ID", "Chain", "Position", "Orig_AA", "Mut_AA", "Gene_Name", "Prediction", "Probability")
knitr::kable(example_result, align = c("c", "c", "c", "c", "c", "c", "c", "c"))

Exploratory Usage

Network properties can be calculated by using different distance cutoffs. In this approach, predatoR() does not make any prediction about the mutation, but returns a data frame contains all 24 features annotated to the dataset. Both network formalisation approaches also can be used.

# networks build using all atoms and 7.6Å cutoff
prediction_result <- predatoR(input_df, distance_cutoff = 7.6, network_approach = "all") 

# networks build using only Cα atoms and 8Å cutoff
prediction_result <- predatoR(input_df, distance_cutoff = 8, network_approach = "ca")