Sequence2Vec: a novel embedding approach for modeling transcription factor binding affinity landscape.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Accurate characterization of transcription factor (TF)-DNA affinity is essential for understanding gene regulation.
• Predicting TF-DNA binding affinity remains a significant challenge despite advances in biotechnology.

Purpose of the Study:

• To propose a novel sequence embedding approach for modeling the TF-DNA binding affinity landscape.
• To develop a predictive model that accurately captures TF-DNA binding characteristics.

Main Methods:

• Representing DNA binding sequences using hidden Markov models (HMMs) to capture positional and long-range dependencies.
• Employing a message passing-like embedding algorithm, Sequence2Vec, to map HMMs into a nonlinear feature space.
• Integrating probabilistic graphical models, feature space embedding, and deep learning for predictive modeling.

Main Results:

• Sequence2Vec effectively models the TF-DNA binding affinity landscape.
• The method demonstrates superior performance compared to alternative machine learning and state-of-the-art binding affinity prediction methods.
• Comprehensive validation was performed on over 90 large-scale TF-DNA datasets from diverse high-throughput experimental technologies.

Conclusions:

• Sequence2Vec offers a powerful and accurate approach for predicting TF-DNA binding affinity.
• This method advances the quantitative understanding of molecular mechanisms in endogenous gene regulation.
• The Sequence2Vec program is publicly available for research use.