This repository contains the code and data for training the RNAcompete experiment classifier used in Ray et al. 2022. To reproduce the results in the paper, follow the instructions below.
RNAcompete_LR_classifier.py: Contains the Python script that trains the classifier, tests the classifier, and applies the classifier to all ucRBP experiments.
training_set.txt: Tab-delimited file containing the feature matrix for training. Each line is an individual RNAcompete experiment. The first column (class) indicates whether it was a passed (1) or failed (0) experiment. Subsequent columns contain values for each of the features as described in the paper.
testing_set.txt: Tab-delimited file with the same format as training_set.txt. This file contains the 40 experiments used to test the classifier.
ucrbp_experiment_features.txt: Tab-delimited file with the same format as training_set.txt and testing_set.txt, EXCEPT for the 'class' column. Contains feature values for all ucRBP experiments (full-length ucRBPs and truncated constructs).
train_IDs.txt: Contains the RNAcompete experiment identifiers for all experiments in training_set.txt.
test_IDs.txt: Contains the RNAcompete experiment identifiers for all experiments in testing_set.txt.
ucrbp_IDs.txt: Contains the RNAcompete experiment identifiers for all experiments in ucrbp_experiment_features.txt. IDs in this file correspond to IDs in Supplementary Table 1.
For all *IDs.txt files, the IDs correspond line-by-line to the values in *_set.txt and ucrbp_experiment_features.txt. For example, to directly view the feature values for ucRBP experiments by ID, the files can be merged as follows:
paste ucrbp_IDs.txt ucrbp_experiment_features.txt
Clone the repository and ensure the following requirements are installed:
- python3 version 3.8
- numpy version 1.19.2
- sklearn version 0.23.2
- skopt version 0.8.1
- pickleshare version 0.7.5
Run the following from within the directory:
python RNAcompete_LR_classifier.py
LR_model.sav: Contains a "pickled" version of the logistic regression model trained in RNAcompete_LR_classifier.py. This model can be loaded in python by running the following:
import pickle
lr = pickle.load(open('LR_model.sav', 'rb'))
LR_coefficients.txt: Tab-delimited file containing one line for each feature along with the weight of the feature in the model.
test_set_AUROC.txt: Contains the AUROC for the test set.
test_set_probability_estimates.txt: Contains one line per test set experiment with the probability estimate that the experiment belongs to class 1 (passed experiments).
ucrbp_probability_estimates.txt: Contains one line per ucRBP experiment with the probability estimate that the experiment belongs to class 1 (passed experiments).
For both *_probability_estimates.txt files, the probability estimates correspond line-by-line to the associated *IDs.txt files (and thus the files containing the features as well). For example, to directly view the probability estimates for ucRBP experiments by ID, the files can be merged as follows:
paste ucrbp_IDs.txt ucrbp_probability_estimates.txt