ERFMTDA

We propose a new computational model based on rotative factorization machine to predict tsRNA-disease associations. This is the implementation of ERFMTDA:
Wei Lan, Dong Wang, Wenyi Chen, Xuhua Yan, Qingfeng Chen, Shirui Pan, Yi Pan.
ERFMTDA: Predicting tsRNA–disease associations using an enhanced rotative factorization machine.
bioRxiv. https://doi.org/10.64898/2026.03.20.713298

Environment Requirement

• Python 3.12.7
• PyTorch 2.5.1
• NumPy 1.26.4
• pandas 2.2.2
• matplotlib 3.9.2

Dataset

The dataset used in this study was manually curated from published literature. We searched the relevant studies and collected experimentally validated tsRNA–disease associations. The curated dataset is provided as tsRNA-disease.xlsx and is included in the ERFMTDA, case study, and denovo folders for different experimental settings.

Usage

1. Data preprocessing
   First run generate_dataset.py to preprocess the tsRNA–disease association data.
2. Model training
   Then run train.py to train the model and output the prediction performance metrics.

Parameter Settings

Dataset preprocessing

The following parameters are configured in generate_dataset.py:

• Number of principal components in the association matrix：32
• Top-k parameter used in the negative sampling module：20

Model training

The following parameters are specified in train.py:

• Feature embedding dimension: 32
• Hidden units in the attention layers: 32
• Dropout rate (attention mechanism and amplification network): 0.1
• Optimizer: Adam
• Learning rate: 1×10⁻³
• L2 regularization weight decay: 1×10⁻⁵
• Batch size: 32
• Training epochs: 200