An AI system, ChromoGen, can predict thousands of 3D structures of chromatin [kroh-muh-tin] - the mixture of DNA and proteins packed into chromosomes - in minutes. Developed at the Massachusetts Institute of Technology (MIT), this deep learning approach aims to accelerate research into how chromatin structures affect gene expression in individual cells, crucial for understanding genetic diseases and developing gene-editing treatments. Chromatin allows the DNA in the genome to fold up and fit into the nucleus of each cell. Its building blocks, nucleosomes [noo-klee-uh-sohms], comprise sections of DNA wound around histone [his-tohs] proteins. These form chromatin fibers that fold into chromosomes. Folded chromatin structures regulate gene expression by controlling the proximity of promotor [proh-moh-ter] and enhancer regions on the DNA. ChromoGen 'reads' DNA sequences to predict chromatin structures, providing data to answer questions relating chromatin structure and gene expression, according to Bin Zhang of MIT. ChromoGen uses diffusion modeling, generating new data by adding and reversing random noise in the training dataset. The model was trained with over 11 million known 3D genome structures and taught to associate chromatin structures with DNA sequences. The system can generate a thousand structures for a DNA region in 20 minutes on one GPU. Aleksandr Sahakyan at the University of Oxford comments that this brings genomic spatial interactions closer to their three-dimensional representation, outlining the role of the underlying DNA sequence. He predicts that genome folding will soon be solved like protein folding through AlphaFold. Separately, a Columbia University team developed the General Expression Transformer (GET), an AI model predicting gene expression in cells. Trained on data from over 1.3 million cells covering 213 human cell types, GET can make accurate predictions about cell types like astrocytes [as-troh-sites]. Raul Rabadan, Director of the Program for Mathematical Genomics at Columbia University, describes GET as a revolution in biology, enabling predictive science based on datasets for gene regulation. Researchers hope GET will aid in developing gene therapies to correct mutations affecting specific cell types. GET could also facilitate decisions on experiments by identifying relevant genetic combinations in diseases like cancer, where numerous mutations may occur. Xi Fu, a doctoral student in Rabadan's lab, trained GET using information from normal human tissue cells, differing from approaches focusing on abnormal cells.
AI Predicts Chromatin Structure and Gene Expression with Unprecedented Speed and Accuracy
閱讀更多有關此主題的新聞:
发现错误或不准确的地方吗?
我们会尽快处理您的评论。