MCI-frcnn: A deep learning method for topological micro-domain boundary detection

• DOI: 10.3389/fcell.2022.1050769
• Publication Date: 2022-11-30T11:38:36Z
• Authors: Simon Zhongyuan Tian; Pengfei Yin; Kai Jing; Yang Yang; Yewen Xu; Guangyu Huang; Duo Ning; Melissa J. Fullwood; Meizhen Zheng
• Keywords: 0301 basic medicine; domain boundary; QH301-705.5; :Biological sciences [Science]; Topological Micro-Domain; deep learning; faster R-CNN algorithm; Cell and Developmental Biology; 03 medical and health sciences; Deep Learning; 3D genome organization; topological micro-domain; Biology (General)

Chromatin structural domains, or topologically associated domains (TADs), are a general organizing principle in chromatin biology. RNA polymerase II (RNAPII) mediates multiple interactive loops, tethering together as RNAPII-associated interaction (RAIDs) to offer framework for gene regulation. RAID and TAD alterations have been found be with diseases. They can further dissected micro-domains (micro-TADs micro-RAIDs) by clustering single-molecule chromatin-interactive complexes from next-generation three-dimensional (3D) genome techniques, such ChIA-Drop. Currently, there few tools available micro-domain boundary identification. In this work, we developed the MCI-frcnn deep learning method train Faster Region-based Convolutional Neural Network (Faster R-CNN) detection. At training phase MCI-frcnn, 50 images of RAIDs Drosophila RNAPII ChIA-Drop data, containing 261 micro-RAIDs ground truth boundaries, were trained 7 days. Using well-trained detected micro-RAID boundaries input new images, fast speed (5.26 fps), high recognition accuracy (AUROC = 0.85, mAP 0.69), region quantification (genomic IoU 76%). We applied detect human micro-TADs using GM12878 SPRITE data obtained score (mean gIoU 85%). all, which work is tool