QTFPred: robust high-performance quantum machine learning modeling that predicts main and cooperative transcription...

Area of Science:

• Genomics
• Computational Biology
• Quantum Machine Learning

Background:

• Deep learning is crucial for identifying transcription factor (TF) binding sites.
• Conventional methods face challenges with limited training data for specific TFs.

Purpose of the Study:

• Introduce QTFPred, a quantum-classical hybrid framework for base-resolution TF binding prediction.
• Address data scarcity issues in TF binding site identification.

Main Methods:

• Integrate quantum convolutional layers into neural networks.
• Utilize quantum circuits for exponential feature space expansion.
• Train the model from scratch using GPU simulation.

Main Results:

• Achieve robust performance in data-sparse scenarios.
• Deliver state-of-the-art accuracy on 49 ENCODE ChIP-seq datasets (92% binary, 96% signal prediction).
• Outperform conventional models in precision and stability, revealing TF motif representations and cooperative binding insights.

Conclusions:

• Quantum machine learning, via QTFPred, offers a powerful solution for genomics modeling.
• QTFPred overcomes limitations of traditional deep learning in TF binding prediction.
• The framework provides insights into TF binding mechanisms and cooperative binding.