Deciphering dark matter in human gut metagenome using advanced sequencing technology and machine learning algorithms

Project Description

 
The human gastrointestinal tract harbours the highest microbial density in the body, with thousands of species coexisting in a complex and ever-changing community. These microbes, often called the "second genome," have a profound impact on our health, diseases, and other physiological processes. Understanding how the gut microbiome changes in different conditions is critical, but most gut microbes cannot be readily isolated and cultured in a lab. Instead, scientists use metagenomic sequencing, which analyses the mixed DNA of gut microbes directly from human faecal samples. This approach allows researchers to assemble microbial genomes, known as metagenome-assembled genomes (MAGs). In our project, we used advanced sequencing technologies and designed machine learning tools to better study the gut microbiome (https://www.youtube.com/watch?v=voGnbAHkuDU&t=1s). Collaborating with industry partners, we identified linked-read sequencing as a cost-effective method for analysing gut microbes (https://www.youtube.com/watch?v=j6Yqt8viM8A&pp=ygUVbGlua2VkLXJlYWRzIGx1IHpoYW5n). This technique uses barcodes to group short-reads from the same long DNA fragments, making genome assembly more efficient. However, computational tools specifically designed to perform linked-read metagenome assembly were lacking. To fill this gap, we developed Pangaea, a tool that uses deep learning to group and assemble these co-barcode linked-reads (https://www.youtube.com/watch?v=yGaIZryvZwg&t=474s). It could improve the accuracy of microbial genome assembly, particularly for low-abundance microbes. To further refine the contigs, we developed DeepMAS, a tool that detects and corrects misassembly by transforming sequencing alignment data into images and analysing them with computer vision techniques. Because traditional tools often overlook short contigs in contig binning, we introduced DeepMetaBin and METAMVGL, which use advanced graph-based methods to link both short and long contigs to increase MAG completeness. To eliminate contamination in these MAGs, we developed Deepurify, which uses a multi-modal deep learning model to ensure only contigs from the same species are included. All these tools are integrated into LRTK, a versatile toolkit for linked-read metagenomic analysis. Based on these advanced tools, we analysed over 10,000 human gut microbiome samples and generated high-quality reference genomes. This work revealed significant differences in gut microbial genomic variations between Han Chinese and Western populations, which could influence diet, disease, and treatment responses. Our project not only advances gut microbiome research but also provides practical tools for scientists worldwide, paving the way for personalised medicine and a deeper understanding of human health.

Project Investigator

Dr ZHANG Lu (Department of Computer Sciene)

Project Collaboraters

• Professor BIAN Zhaoxiang (Vincent V.C. Woo Chinese Medicine Clinical Research Institute)
• Dr ZHAI Lixiang (Centre for Chinese Herbal Medicine Drug Development Limited)
• Professor Pavel Pevzner  (University of California, San Diego (UCSD))
• Professor Zhang Louxin (National University of Singapore (NUS))
• Professor Cai Yunpeng (Shenzhen Institutes of Advanced Technology (SIAT))

Funding/Award

• Research Grants Council - Young Collaborative Research Grant
• Health Bureau - Health and Medical Research Fund
• Innovation and Technology Commission - Guangdong-Hong Kong Technology Cooperation Funding Scheme
• Beijing Genome Institute
• Guangdong Science and Technology Department - Guangdong Basic and Applied Basic Research Foundation

Data Analytics and Artificial Intelligence

EXPLORE MORE