TRAFICA: an open chromatin language model to improve transcription factor binding affinity prediction

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• In silico transcription factor and DNA (TF-DNA) binding affinity prediction is crucial for understanding gene regulation.
• Current tools often overlook in vivo TF binding preferences in open chromatin regions, relying solely on in vitro data.
• This limitation hinders accurate prediction of TF-DNA interactions in their native cellular context.

Purpose of the Study:

• To develop an advanced computational model for predicting TF-DNA binding affinity.
• To incorporate sequence characteristics of in vivo open chromatin regions into binding affinity prediction.
• To improve the accuracy and biological relevance of TF-DNA binding predictions.

Main Methods:

• Developed TRAFICA, an open chromatin language model.
• Pre-trained TRAFICA on over 2.8 million nucleotide sequences from ATAC-seq experiments to learn in vivo TF binding preferences.
• Fine-tuned TRAFICA using low-rank adaptation (LoRA) on PBM and HT-SELEX TF-DNA binding profiles.

Main Results:

• TRAFICA significantly outperformed existing prediction tools and advanced DNA language models.
• Achieved state-of-the-art performance in predicting both in vitro and in vivo TF-DNA binding affinity.
• Demonstrated the importance of incorporating open chromatin sequence characteristics for improved prediction.

Conclusions:

• Considering sequence characteristics from open chromatin regions significantly enhances TF-DNA binding affinity prediction.
• TRAFICA represents a novel approach for more accurate TF-DNA interaction modeling.
• The findings pave the way for better understanding of gene regulation through improved TF-DNA binding predictions.