QTFPred: robust high-performance quantum machine learning modeling that predicts main and cooperative transcription factor bindings with base resolution

TL;DR

QTFPred is a quantum machine learning model that accurately predicts transcription factor bindings at base resolution, even with limited data.

Contribution

Introduces QTFPred, a quantum-classical hybrid framework for robust TF binding prediction with superior performance in data-sparse scenarios.

Findings

QTFPred achieved state-of-the-art accuracy in 92% of binary and 96% of signal prediction tasks on 49 ChIP-seq datasets.

The model outperformed conventional models in precision and stability, especially in data-sparse conditions.

QTFPred reveals TF motif representations and provides insights into cooperative binding mechanisms.

Abstract

Deep learning has become an essential tool for identifying transcription factor (TF) binding sites, yet conventional approaches often struggle with limited training data for specific TFs. Here, we introduce QTFPred (Quantum-based TF Predictor), a quantum-classical hybrid framework that integrates quantum convolutional layers within neural networks to predict TF binding at base resolution. By leveraging the exponential feature space offered by quantum circuits and training from scratch via GPU simulation, QTFPred achieves robust performance even in data-sparse scenarios. In benchmarks on 49 Encyclopedia of DNA elements ChIP-seq datasets, QTFPred delivered state-of-the-art accuracy in 92% of binary prediction and 96% of signal prediction tasks, outperforming conventional models in precision and stability. Moreover, the method reveals underlying TF motif representations, offering insights into cooperative binding mechanisms. These results highlight the potential of quantum machine learning to overcome the limitations of traditional deep learning in genomics modeling.

Figures