TRAFICA: An Open Chromatin Language Model to Improve Transcription Factor Binding Affinity Prediction

The logo of TRAFICA was generated by using the application of StableDiffusionXL in Poe

---

(*) System requirement

1. OS: Linux
2. Nvidia GPU (CUDA support is need):

The pre-training of TRAFICA took over five days on a single Nvidia A100 GPU card. The fine-tuning of TRAFICA took about 30 minutes for an HT-SELEX dataset on the same hardware device.

3. Python and other dependencies: environment.yaml

---

(A) Installation: Conda environment

1. Creat a new conda environment with the provided '.yaml' file.

conda env create -f environment.yaml

2. Activate the conda environment

conda activate TRAFICA

(B) Quick start (with example data)

TRAFICA pre-training and fine-tuning (Click here for details)

• Pre-training from scratch and Fine-tuning the pre-trained using the example data

(C) The availability of the pre-trained moel and Data

• The weights of pre-trained model is available at the HuggingFace repository
• The datasets used for TRAFICA pre-training/fine-tuning and evaluation are available at Zenodo

---

(D) Contact:

Mr. Yu Xu, email: [email protected]; [email protected]
Dr. Eric Lu Zhang, email: [email protected]