TrumanPhd / f5C-finder

f5C-finder is the first nueral network-based model for the identification on 5‑Formylcytidine modifications on mRNA

f5C-finder

Introduction

f5C-finder is the first neural network-based model for the identification on 5‑Formylcytidine modifications on mRNA. This repository is the offical tensorflow vision of f5C-finder with all the codes used in the paper:
f5C-finder: combining language model with multi-head attention for predicting 5‑Formylcytidine modifications on mRNA

1. Webserver with GUI

Your can use f5C-finder for inference at our webserver , submitted data will be kept confidential.

2. env

make sure your tensorflow > 2.0

3. Installation

You can download this Repository with

git clone https://github.com/TrumanPhd/f5C-finder.git

or just download with Github GUI

4. Source code

1.Ablation experiments

Ablation.py and Ablation_evaluation.py are the scripts for Ablation experiments

And for the Ablation results, please reference to the Ablation_exp.xlsx and Ablation_expRNN.xlsx

2.The structure of 2 LSTM-based models and 3 attention-based models

please reference: Model_lib2.py for LSTM-based models

Fusion3feature.py, encoder1234_best.py Hash1_best.py for attention-based models

(the five encoder can be referenced from feature_selection.py utils.py and Hao_keras_toolkit.py)

3. model training 10-fold cross-valiation and independent test

train.py and model_evaluation.py for model training and evaluation

the results can be referenced from metrics.xlsx

4. RF SVM ADA the loser models VS f5C-finder

for RF ADA

can be referenced from ML.py ML_evaluation.py best_hyperparameters.xlsx

for SVM

can be referenced from SVM.py ML_evaluation.py SVMresults.xlsx

5. Figure

plot.py

and other files are the saved parameters from the model training.

6. Dataset

In the dataset, we provide the data trained and evaluated from all the models, to make it more usable we replace 'U' to 'T'，as shown in the dataset.

Citation

It`s my pleasure if can use f5C-finder for f5C identification or improve the accuarcy on the provided f5C dataset, and please cite at