Prediction of chromatin looping using deep hybrid learning (DHL)
image: Workflow of deep hybrid learning for chromatin loop prediction.
Credit: Mateusz Chiliński , Anup Kumar Halder , Dariusz Plewczynski
Studying the spatial landscape of the genome can bring new insights into our understanding of the disease developments. Multiple methods of capturing the spatial landscape were introduced from 2002, but these experiments are very time and cost-consuming. As artificial intelligence has revolutionized the field of biological sciences, a series of algorithms have been created to decipher the language of noncoding regions of DNA.
Recently, Quantitative Biology published a study entitled “Prediction of chromatin looping using deep hybrid learning (DHL)”, which incorporated the DNABERT pre-trained model in deep hybrid learning (DHL) system and devised a fine-tuning method for chromatin loop prediction (Figure 1). The result showed DNABERT could be used to predict the ChIA-PET experiments with high precision, and the DHL approach had increased the metrics on CTCF and RNAPII sets by using 1D genomic sequence information. This proposed DHL is stable and robust. It will provide a convenient and effective way to improve our knowledge about the gene regulation effect of chromatin interactions.