RF4SV: A Random Forest approach for accurate deletion detection

This is a machine learning model based for accurate deletion detection for RF4SV OSF project.

About RF4SV

Efficiently detecting genomic structural variants (SVs) is a key step to grasp the "missing heritability" underlying complex traits involved in major evolutionary processes such as speciation, phenotypic plasticity, and adaptive responses. We present a random forest ensemble method for accurate deletion identification. We called this approach RF4SV.

Getting Started

Requirements:

- python3
- numpy
- pandas
- scikit-learn
- keras
- tensorflow

Download Drosophila melanogaster genome

 wget ftp://ftp.ensemblgenomes.org/pub/metazoa/release-43/fasta/drosophila_melanogaster/dna/Drosophila_melanogaster.BDGP6.22.dna.chromosome.*

Simulate deletions (RSVSim)

simdata/simSVDel.R

Read simulation process

Read simulation

simdata/iss_ReadSim.sh

-- Remember to adapt the paths and outputs.

Pre-process

Create input prediction matrix

MappExtract.sh

Handling imbalanced Data

Resampling data to find "good" balance

python balance_dataset.py

Benchmarking 1

To benchmark classique (LR: Logistic Regression; LDA: Linear Discriminant Analysis; KNN: k-Nearest Neighbors; NB: Naive Bayes; CART: Classification and Regression Tree algorithm) and ensemble learning methods (RF: Random Forest; ADA: AdaBoost; GBM: Gradient Boosting Machines):

To run the benchmark:

python benchml.py

-- Remeber to change input file in the data folder

Benchmarking 2

To benchmark using Neural Network (Deep Learning)

python ml4sv_cnn.py

-- Remember to change input file in the data folder

Building the RF model

To build and save a RF model

python rf_model.py

Do prediction

To load a saved RF model (uploaded in OSF project --> Files --> Model --> 'RF_model.sav') and predict new data

python predict.py

Post-process

For friendly user prediction results

python path/PredictPostProcess052020.py path/predictionfile.csv

-- In this case, data must be with 1 or 0 in column "Variant"

Docker RF4SV

Train and predict models using Random Forest, Benchmarking Algorithms and Neural Network

To run RF4SV project using Docker:

docker run --rm -v /your/data/dir:/data robertoxavier/rf4sv:v1.0 bash