LPItabformer

Installation

conda create -n lpitabformer-env python=3.11
conda activate lpitabformer-env
pip install lpitabformer
pip install -r requirement.txt

Usage

from lpitabformer.ft_transformer import *
import torch

model = FTTransformer(
    categories = (10, 5, 6, 5, 8),      # tuple containing the number of unique values within each category
    num_continuous = 10,                # number of continuous values
    dim = 32,                           # dimension
    dim_out = 1,                        # binary prediction, but could be anything
    depth = 1,                          # depth
    heads = 8,                          # heads
    attn_dropout = 0.2,                 # post-attention dropout
    ff_dropout = 0.2,                   # feed forward dropout
    p = 0.8,                            # the strength of DSU
    )

# This input just for example, not for LPIs.
x_categ = torch.randint(0, 5, (10, 5))  # category values, from 0 - max number of categories, in the order as passed into the constructor above
x_numer = torch.randn(10, 10)           # numerical value

pred = model(x_categ, x_numer) 
print(pred.shape)

Example

The example code for training LPItabformer is in example/example.zip, you can download and unzip it and use it according to README.txt. dsuTab_cz.ipynb is the main example. The example for using the trained LPItabformer model to predict new LPIs are in example/tool_example/, you could follow predict.ipynb to predict your LPI data.

Model weights

The saving model weights are in example/tool_example/data/model/, hsa and mus represent human and mouse, respectively. You also could get from the example/example.zip. Note you need ft_transformer.py when you load the model weights, and you also leverage example/tool_example/predict.ipynb.

Input format

For the input data format, please refer to the instructions in the provided example or case input.

Citation

@article{lin2025lpitabformer,
  title={LPItabformer: Enhancing generalization in predicting lncRNA-protein interactions via a tabular Transformer},
  author={Lin, Qin and Sheng, Jie and Zhou, Chang and Xiao, Tao and Meng, Yilei and Lu, Mingxin and Zhang, Junfang and Yan, Xueyun and Peng, Luying and Cao, Huaming and Li, Li},
  journal={Computational and Structural Biotechnology Journal},
  year={2025},
  publisher={Elsevier}
}