BrainDead - RNA classification via accessible k-mers

Setup

To install and run BrainDead, you should use a commandline interface like bash.

(1) Downloading BrainDead script files from Github

First, go to a folder where you want to "install" BrainDead, i.e. downloading the script files listed in this github repository.

# download all files from 
wget https://github.com/BackofenLab/BrainDead/archive/refs/heads/main.zip
# uncompress the archive (might need an additional installation of "unzip")
unzip main.zip
# enter the unpacked and created folder
cd BrainDead-main

Afterwards, you find in

• src the script files (see below for usage)
• data the data files used for training the models etc.

The recommended way for running BrainDead locally is via Conda.

(2) Installing miniconda (if not already done)

If you don't have conda on your system (check if conda --version gives any output or an error message), please follow the installtion instruction for the minimal conda setup available via the following link.

• miniconda installation

conda is a commandline tool to simplify the installation and local management of tools and their dependencies in linux-based systems.

(3) Importing the dependencies via conda

Next, we want to install all files and tools needed to run BrainDead via conda. To this end, we will create a "conda environment", i.e. kind of a folder that contains all tools not part of this github repository.

The needed files and tools are listed in the conda-environment.yml file within the BrainDead-main folder extracted in step (1).

The following command creates the environment and imports BrainDead's dependencies.

conda env create --file conda-environment.yml

(4) Starting the conda environment

The created environment can be activate with the following command

conda activate BrainDead

Afterwards you should be able to call BrainDead's scripts (see below).

*Note: the script files are located in the src subdirectory!`

At the end of the page, you find the example calls.

Command helps

Generate kmer features with generate_kmer_features.py

usage: generate_kmer_features.py [-h] --kmers KMERS --fasta FASTA [--report-counts]
                                 [--out-csv OUT_CSV] [--minE-subopt MINE_SUBOPT]
                                 [--minE-intarna MINE_INTARNA]

Generate kmer-based count features based on sequence plus RNAsubopt and IntaRNA position-
wise energies.. 
Sample calls: 
"python generate_kmer_features.py --kmers "AGA,GC,GGG" --fasta test.fa --out-csv counts.csv" 
"python generate_kmer_features.py --kmers "AGA,GC,GGG" --fasta test.fa --out-csv "stdout" --report-counts --minE-subopt -5 --minE-intarna -2"

optional arguments:
  -h, --help            show this help message and exit
  --kmers KMERS         List of kmers as a comma separated string e.g. "AGG,GA,GG"
  --fasta FASTA         Sequences to extract features from as a FASTA file
  --report-counts       Whether to report counts as integer, default is binary
                        nohit(0)-hit(1)
  --out-csv OUT_CSV     CSV File name to write counts, pass "stdout" for stdout
  --minE-subopt MINE_SUBOPT
                        Minimum free energy of the position on RNAsubopt result
  --minE-intarna MINE_INTARNA
                        Minimum free energy of the position on IntaRNA result

Fit and predict ML model with fit_predict.py

usage: fit_predict.py [-h] --features-train FEATURES_TRAIN
                      [--features-train-index-col FEATURES_TRAIN_INDEX_COL]
                      [--features-train-header FEATURES_TRAIN_HEADER]
                      --labels-train LABELS_TRAIN
                      [--labels-train-index-col LABELS_TRAIN_INDEX_COL]
                      [--labels-train-header LABELS_TRAIN_HEADER]
                      [--model-choice {SVM-rbf,SVM-linear,Logistic-liblinear,Logistic-lbfgs}]
                      [--save-model] [--load-model LOAD_MODEL]
                      [--out-model OUT_MODEL] [--predict]
                      [--features-predict FEATURES_PREDICT]
                      [--features-predict-index-col FEATURES_PREDICT_INDEX_COL]
                      [--features-predict-header FEATURES_PREDICT_HEADER]
                      [--out-predict-labels OUT_PREDICT_LABELS]
                      [--not-validate-indexes] [--not-unique-indexes]
                      [--store-weights] [--standardize-scaling]

Train a machine learning model for the tabular CSV inputs, and predict on
optional CSV input Sample call: "python fit_predict.py --features-train
test_features.csv --labels-train test_labels.csv --features-train-header 0
--model SVM-rbf""python fit_predict.py --predict --features-predict
test_features.csv --features-predict-header 0 --load-model model.pkl""python
fit_predict.py --features-train test_features.csv --labels-train
test_labels.csv --features-train-header 0 --save-model --out-model jj.pkl
--predict --features-predict test_features.csv --features-predict-header 0
--load-model model.pkl"

optional arguments:
  -h, --help            show this help message and exit
  --features-train FEATURES_TRAIN
                        Input features to train model in Comma separated
                        format (CSV)
  --features-train-index-col FEATURES_TRAIN_INDEX_COL
                        The column number with index keys are provided
  --features-train-header FEATURES_TRAIN_HEADER
                        Whether features CSV has a header line
  --labels-train LABELS_TRAIN
                        Input reference labels to train model as one-column
                        Comma separated format (CSV)
  --labels-train-index-col LABELS_TRAIN_INDEX_COL
                        The column number with index keys are provided
  --labels-train-header LABELS_TRAIN_HEADER
                        Whether labels CSV has a header line
  --model-choice {SVM-rbf,SVM-linear,Logistic-liblinear,Logistic-lbfgs}
                        Included classifiers from scikit, see https://scikit-
                        learn.org/stable/modules/svm.html#classification
  --save-model          Save the trained scikit model into file
  --load-model LOAD_MODEL
                        Load model from file, instead of training on CSV input
  --out-model OUT_MODEL
                        Save the trained scikit model into file
  --predict             Optionally predict using the trained model, requires
                        --features-predict option
  --features-predict FEATURES_PREDICT
                        Input features to predict in comma separated format
                        (CSV)
  --features-predict-index-col FEATURES_PREDICT_INDEX_COL
                        The column number with index keys are provided
  --features-predict-header FEATURES_PREDICT_HEADER
                        Whether features CSV has a header line
  --out-predict-labels OUT_PREDICT_LABELS
                        File to store the predicted labels
  --not-validate-indexes
                        Check keys are consistently the same in lables and
                        features
  --not-unique-indexes  Check keys are not duplicated
  --store-weights       Save SVM coefficient weights
  --standardize-scaling
                        Do not scale and standardize features

Sample calls on test data:

1. Generating k-mer sequence and structure features

python src/generate_kmer_features.py --kmers "AGA,GC,GGG" --fasta data/test.fa --out-csv data/counts.csv

2. Training and saving the ML model:

$ python src/fit_predict.py --features-train data/190411-mTLR7-features.csv  --labels-train data/190411-mTLR7-labels.csv --features-train-header 0  --save-model --out-model data/190411-mTLR7.pkl

3. Training and predicting:

$ python src/fit_predict.py --features-train data/190411-mTLR7-features.csv  --labels-train data/190411-mTLR7-labels.csv --features-train-header 0  --save-model --out-model data/190411-mTLR7.pkl --predict --features-predict data/190411-mTLR7-features.csv --features-predict-header 0 --out-predict-labels data/predict.cvs

4. Predicting from the saved model:

$ python src/fit_predict.py --predict --features-predict data/190411-mTLR7-features.csv --features-predict-header 0 --out-predict-labels predict.cvs --load-model data/190411-mTLR7.pkl