PlantBind: An attention-based multi-label neural networks for predicting transcription factor binding sites in plants

Introduction

Identify of transcription factor binding sites (TFBSs) are essential for the analysis of gene regulation. Here, we present PlantBind, a method for the integrated prediction and interpretation of TFBS events from DNA sequences and DNA shape profiles. Built upon an attention-based multi-label deep learning framework, PlantBind not only simultaneously predicts the binding sites of 315 TFs. As shown in Figure 1, the hybrid model consists of data processing module, embedding module, feature extraction module, and multi-label output module.

Fig. 1 The model workflow

The model provides researchers with tools to:

1. Identify the TFBSs of transcription factors.
2. Identify DNA-binding motfi of transcription factors

Tutorials

These are a work in progress, so forgive incompleteness for the moment. If there's a task that you're interested in that I haven't included, feel free to post it as an Issue at the top.

1. Software Requirements

We recommend that you use conda to install all of the following software.

software list

• python v3.8.12
• pytorch v1.10.2
• numpy v1.16.4
• pandas v1.4.1
• sklearn v1.0.2
• scipy v1.5.3
• matplotlib v3.5.1

2. Data information and processing

In this part, we will first introduce the data information used in this model, then introduce the training and predicting data formats, and finally introduce how to create a data set that meets the model requirements for prediction.
All data is in the data directory:

• Ath-TF-peaks: the TFBS peak info of 315 Ath TFs, and one neg.bed file
• Maize-TF-peaks: the TFBS peak info of 4 maize TFs for trans-specise
• model: The file that holds the model, which can be loaded to predict new datasets

2.1 Training and Predicting data formats

For training, the data mainly consists of three files: (1)DNA sequence file; (2)DNA shape file; (3)data label file
For predicting, the data mainly consists of three files: (1)DNA sequence file; (2)DNA shape file

• Data Format Details Introduction
  • DNA Sequence File
  • DNA Shape File
  • Data Label File

2.2 Construction of the dataset

Next, we will mainly introduce how to create the files mentioned above.

• Data processing methods
  • DNA Sequence Data
  • DNA Shape Data
  • Data Label

3. Train and Test the model

• Training
  Input: train_sequence.table,train_DNA_shape.npy,train_label.txt.
  All data files need to be placed in the same folder before starting training, such as data_folder Output: trained_model_101_seqs.pkl

python -m torch.distributed.launch --nproc_per_node=8 \
        main.py --inputs data_folder/ --length 101 \
        --OHEM True --focal_loss True \
        --batch_size 1024 --lr 0.01 \
        --multi_GPU True

• Testing
  Input: test_sequence.table,test_DNA_shape.npy,test_label.txt.
  All data files need to be placed in the same folder before starting testing, such as data_folder
  Model:trained_model_101_seqs.pkl
  Output: metrics_b.json,metrics_m.json,Gmean_threshold_b.txt,Gmean_threshold_m.txt,precision_recall_curve_TFs_315.pdf,roc_prc_curve_TFs_315.pdf

python main.py --inputs data_folder/ --length 101 --mode test

4. Use the model to predict new data

Input: sequence.table,DNA_shape.npy
Output: accuracy
The analysis is done by using the inference.py script and modifying the relevant parameters.