README.md

In ZW-xjtlu/WhistleR: Comprehensive Annotation of Predictive Features for Interval-based Genomic Data

WhistleR: extract genomic features for predictive modeling in genomics

2021-01-30

Introduction

This guide will offer an overview on a work flow of Bioconductor package WhistleR to annotate comprehensive genome-derived features on the range-based genomic objects. The package can be used to extract properties of region lengths, sequence contents, relative positions, clustering effects, and conservation scores on top of the commonly used transcript annotations including 5’UTR, CDS, 3’UTR, transcripts, genes, and promoters. The genome-derived features can be used to predict variety of RNA and gene related genomic assays, such as the RBP binding sites and the RNA modification peaks or sites. WhistleR also provide the annotation of new attributes over user-defined genomic regions, enabling the augmentation of feature engineering under the interactive framework between properties and regions.

The package currently implements 2 types of feature extraction modules: the genome-derived features and the sequence-derived features.The former features are obtained through the interaction between genomic properties and the genomic regions, and the later includes the one-hot encoded nucleotide sequences or the encoding through pseudo nucleotide compositions (PseTNC). The genome-derived features can be annotated with the function “genomeDerivedFeatures”, and the sequence features can be annotated using the function “sequenceDerivedFeatures”. In addition, more genome-derived features can be defined using functions with names “extractRegion” + the metric names. While the specific functions can generate the genomic properties over the user defined region types.

The design of package enables the one-step extraction of hundreds of genomic features based only on the GRanges of the target genome intervals. Wide class of the Bioconductor annotation objects, including Txdb, EnsDb, BSgenome, and GScores, are supported in the package to enable the massive enumeration of features that based on novel sources of genomic regions.

While the comprehensive feature input is vital to the success of a genomic data science project, the genome-derived features are often more informative and interpretive than the primary sequence features. Thus, the WhistleR package can provide reliable support to both the machine learning modeling and the predictive factor identification on the genomic response variable.

This user’s guide will first give an overview to the key functionalities of the package, then illustrate the utility of the predictive features using a case study of the full workflow of a genomic prediction task, from EDA to ML modeling.

Overview of functionalities

WhistleR provides fast and comprehensive feature extraction from the range-based genome annotation; the extracted features can be used as the input for a variety of modeling purposes such as supervised predictive modeling and unsupervised factor analysis. The target range-based annotation should be stored in GRanges object, which can be created using the constructor function defined in the GenomicFeatures package. Alternatively, the GRanges can be imported from the external annotation files in BED, GFF, or GTF formats using the import function defined in dplyr package.

There are 2 types of genomic features implemented in WhistleR: the region properties features, and the sequence-derived features. The former is extracted using an interactive design between the genomic properties and the genomic regions, the latter is the sequence-derived features extracted from the sequences of the flanking regions of the annotation. When extracting the sequence-derived features, different encoding methods can be specified, such as the one-hot method or the pseudo nucleotide composition method.

Feature extraction framework of WhistleR

Download

WhistleR can be installed from both bioconductor and github.

To install WhistleR from bioconductor, start R (version >“3.6”) and enter:

if(!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("WhistleR")

For order versions of R, please refer to the appropriate Bioconductor release.

Installation from github can be initiated with the following command:

devtools::install_github("zw-xjtlu/WhistleR")

Complete User's Guide

For getting more information about the usage of WhistleR, enter the following R command after installation to view the instruction file:

WhistleR::WhistlerUsersGuide()

ZW-xjtlu/WhistleR documentation built on March 13, 2021, 10:50 a.m.