Researchers at Henan Polytechnic University have developed TM-Loop, a new computational framework for identifying chromatin loops—critical regulatory structures that bring distant genomic elements together to control gene transcription. The method utilizes 10 kb Hi-C (a technique to study 3D genome architecture) contact matrices as primary data, integrating ATAC-seq and CTCF ChIP-seq signals to construct a weighted feature system that mitigates sample imbalance. By employing a Transformer-based deep learning model, TM-Loop leverages multi-head attention mechanisms to capture both global and local feature dependencies within the genome.

The framework addresses significant limitations in existing 3D genomics tools, particularly in managing contact matrices characterized by low signal-to-noise ratios and high data sparsity. To improve accuracy, the system incorporates a dual-threshold filtering process and anchor-guided clustering, which effectively eliminate false signals and reduce the false discovery rate. Validation experiments demonstrate that TM-Loop outperforms current state-of-the-art methods across benchmarks including APA, protein enrichment, and 3D structural consistency. The findings were published in Scientific Reports on June 11, 2026, with the source code made available to the research community via GitHub. This project received funding support from the Henan Provincial Department of Science and Technology Research Project under Grant No. 252102210007.

Researchers at Henan Polytechnic University have developed TM-Loop, a new computational framework for identifying chromatin loops—critical regulatory structures that bring distant genomic elements together to control gene transcription. The method utilizes 10 kb Hi-C (a technique to study 3D genome architecture) contact matrices as primary data, integrating ATAC-seq and CTCF ChIP-seq signals to construct a weighted feature system that mitigates sample imbalance. By employing a Transformer-based deep learning model, TM-Loop leverages multi-head attention mechanisms to capture both global and local feature dependencies within the genome.

The framework addresses significant limitations in existing 3D genomics tools, particularly in managing contact matrices characterized by low signal-to-noise ratios and high data sparsity. To improve accuracy, the system incorporates a dual-threshold filtering process and anchor-guided clustering, which effectively eliminate false signals and reduce the false discovery rate. Validation experiments demonstrate that TM-Loop outperforms current state-of-the-art methods across benchmarks including APA, protein enrichment, and 3D structural consistency. The findings were published in Scientific Reports on June 11, 2026, with the source code made available to the research community via GitHub. This project received funding support from the Henan Provincial Department of Science and Technology Research Project under Grant No. 252102210007.