Advancing microRNA Target Site Identification via Bias-Corrected Chimeric Datasets for Machine Learning Approaches

This repository contains all the code required to reproduce the work presented for my dissertation, submitted in partial fulfilment of the requirements for the degree of M.Sc. in Molecular Medicine (By Research) in July 2025.

Repository Structure

The directories are ordered chronologically, corresponding to the order of the analyses and experiments presented in the dissertation. Each analysis/experiment is designed to be run independently. In general, subsequent analyses build upon the outputs or results generated by preceding ones.

Each directory follows the same structure, as follows:

project-root/
│
├── 00_first_analysis/
│   ├── code/
│   ├── data/
│   ├── results/
│   ├── RUNME.sh
│   └── README.md
│
├── 01_second_analysis/
│   ├── code/
│   ├── data/
│   ├── results/
│   ├── RUNME.sh
│   └── README.md
│
└── README.md           # This file

The README.md provides a description of the analysis or experiment and instructions on how to run it in the respective directory. It generally follows a common structure, including a description, list of dependencies (including versions used), and further instructions or notes where necessary.

The RUNME.sh master script is an executable file for running the entire analysis or experiment. It executes other scripts from the code/ directory as needed.

The code/ directory contains Python, R or bash scripts needed to run the analysis or experiment. Each script contains docstrings including a description, usage, and arguments. These scripts are executed by the respective RUNME.sh master script.

Unless indicated otherwise in the respective README.md, the RUNME.sh master script should be run from the parent directory of the analysis/experiment as follows:

bash RUNME.sh

Most master scripts contain SBATCH directives (edit as necessary!) and can run on a HPC with SLURM using:

sbatch RUNME.sh

The data/ and results/ directories are placeholder input and output directories, respectively. These are created when the RUNME.sh master script is executed.

A RUNME.log file is outputted to the results/ directory.

Available Results

Some results are provided for reference; some are available in this repository, while others are available on the data-sharing platform Zenodo.

• 01_Pre_Process_ChimeCLIP
  • results/raw_chimeCLIP_file_list.txt: list of relevant file names to be processed in this work
• 02_Run_HybriDetector
  • results/preprocessed_chimeCLIP_file_list.txt: list of file names to be processed with HybriDetector following pre-processing
  • HybriDetector output files: selected output files to be used for further processing
• 03_Concatenate_HybriDetector_Output
  • AGO2 eCLIP Manakov2022 full dataset (positives): concetenated output of 19 sample files processed with HybriDetcetor
• 04_Post_Process_Biased
  • Biased datasets: biased datasets
• 05_Bias_Analysis
  • results/training: per dataset Decision Tree (DT) Classifier trained on 3-mers derived from miRNA sequences
  • results/evaluation: per dataset Average Precision Score (APS) for the DT Classifier and a Random Classifier
• 06_Post_Process_Unbiased
  • Final miRBench datasets: unbiased / bias-corrected datasets contributed to the miRBench Python package
• 07_Benchmarking
  • results/evaluation: per dataset APS of all miRBench predictors
  • results/pr_curves: per dataset Precision-Recall (PR) curve for all miRBench predictors
• 08_Tree_Based_Models
  • results/training: Bayesian optimisation with 5-fold cross-validation results per model
  • Final trained models: Decision Tree (DT), Random Forest (RF), and XGBoost (XGB) Classifier models
• 09_Retrain_CNN
  • results/training: training history, and training accuracy and loss plots per model
  • Retrained CNNs: CNNs retrained on bias-corrected Hejret2023 train set and unbiased Manakov2022 test set, using the Sequence-only data representation (50_20_1) and the Sequence & Co-folding data representation (50_20_2).

Publications

Most of the work presented here has been included in the following publication:

Sammut, S., Gresova, K., Tzimotoudis, D., Marsalkova, E., Cechak, D., & Alexiou, P. (2025b). miRBench: Novel benchmark datasets for microRNA binding site prediction that mitigate against prevalent microRNA frequency class bias. Bioinformatics, 41, i542–i551. https://doi.org/10.1093/bioinformatics/btaf233

Other publications most relevant to this work include:

Hejret, V., Varadarajan, N. M., Klimentova, E., Gresova, K., Giassa, I.-C., Vanacova, S., & Alexiou, P. (2023). Analysis of chimeric reads characterises the diverse targetome of AGO2-mediated regulation. Scientific Reports, 13. https://doi.org/10.1038/s41598-023-49757-z

Klimentová, E., Hejret, V., Krčmář, J., Grešová, K., Giassa, I.-C., & Alexiou, P. (2022). miRBind: A deep learning method for miRNA binding classification. Genes, 13(12). https://doi.org/10.3390/genes13122323

Manakov, S. A., Shishkin, A. A., Yee, B. A., Shen, K. A., Cox, D. C., Park, S. S., Foster, H. M., Chapman, K. B., Yeo, G. W., & Van Nostrand, E. L. (2022). Scalable and deep profiling of mRNA targets for individual microRNAs with chimeric eCLIP. bioRxiv. https://doi.org/10.1101/2022.02.13.480296