Research

The PKU-EMBL Lab develops a high-throughput pipeline that transforms environmental microbial "dark matter" (微生物“暗物质”) into computable knowledge. Focusing on extreme habitats characterized by high salinity, high alkalinity, or low temperature, we develop molecular biology technologies and intelligent analytical methods for systematic microbial genome reconstruction and functional annotation.

The resulting structured omics data provide training resources for domain-specific artificial intelligence models while supporting the development of intelligent algorithms and predictive models for microbiome research. The pipeline currently integrates three connected levels, progressing from genes to reactions to networks.

Technical Roadmap

An integrated computational and experimental route from extreme habitats and multi-omics data to gene discovery, pathway prediction, community control, and biotechnology applications.

Research Directions

1. Functional genomics
   We develop highly sensitive, deep-learning-enabled methods and supporting algorithms to recover and characterize low-abundance functional guilds and their genomes that are readily missed by conventional workflows. By combining protein and genome language models, we identify novel enzymes, distantly related functional genes, metabolic gene clusters, and environmental adaptation elements that are difficult to detect through traditional homology searches.
2. Metabolic reaction networks
   Using minimal biochemical reaction units as building blocks, we construct knowledge graphs that connect genes, enzymes, substrates, products, and reactions. We integrate knowledge-graph reasoning with vector-database-based semantic retrieval of multi-omics evidence and develop a collaborative multi-agent reasoning framework for the automated discovery, feasibility assessment, and experimental validation of novel metabolic pathways.
3. Microbial community interactions and control
   Starting from metagenome-resolved community structure, we integrate functional genes, metabolite exchange, and environmental factors to build computable world models of microbial communities. These models are designed to predict community interactions, state transitions, and responses to targeted interventions.

Together, these three levels form a progressive gene–reaction–network technology chain and establish a new paradigm for the intelligent analysis of complex microbiomes. In parallel, we are developing prototype microbiome AI models for core tasks including rare-taxon identification, metabolic pathway reasoning, and community-regulation simulation, creating a closed loop from data annotation to intelligent decision-making.

Data Foundation and Translation

The lab has conducted large-scale metagenomic and metatranscriptomic sequencing, together with environmental physicochemical measurements, across extreme aquatic ecosystems in northwestern China, including salt, acidic, and alkaline lakes. The resulting datasets account for more than 80% of comparable data available globally. This unique data foundation supports both AI-driven discovery and rigorous training and validation of our in-house models.

Combining these data with our analytical methods, we have systematically uncovered previously unknown microbial lineages with potential functions and novel biosynthetic gene clusters for bioactive compounds. Cultivation and biosynthetic technologies are then used to advance these discoveries toward applications in environmental engineering, including pollutant bioremediation and biomass-to-energy conversion, and in pharmaceutical development through the discovery of novel bioactive natural products.

Software

The lab's official GitHub organization is PKU-EMBL. The core software project is BASALT, a Nature Communications toolkit for metagenomic binning and refinement. Related software includes BASALT-Air.

Representative Impact

Research led by Tenured A/Prof. Ke Yu includes more than 100 SCI-indexed papers and over 6,000 citations, with first-author or corresponding-author publications in broad-interest, microbiology, environmental science, and environmental engineering journals including Nature Communications, Microbiome, Environmental Science & Technology, and Water Research. His first-author metatranscriptomics study on activated sludge microbiomes was recognized by Prof. Per H. Nielsen as one of the landmark methodological studies for understanding activated-sludge microbial ecology over the past five decades. Another multi-omics study was highlighted by Prof. Marvin Whiteley for advancing methods to study interspecies microbial interactions.

Resources

• Google Scholar: Tenured A/Prof. Ke Yu's publication profile
• GitHub organization: PKU-EMBL
• Core software: BASALT
• Pipeline: BASALT-Air
• Website: PKU-EMBL Lab

WeChat official account

PKU EMBL