🧬 Building biological foundation models across microbial genomes and microbiomes.

Ph.D. Student @ School of Life Science and Technology, Huazhong University of Science and Technology (HUST)

I develop biological foundation models that learn transferable representations across microbial genomes and microbiomes, aiming to build more generalizable, interpretable, and biologically meaningful AI systems for microbiome research.

My research lies at the intersection of Microbiome × Foundation Models × AI for Biology × Bioinformatics.

• 🧬 Biological Foundation Models
• 🦠 Microbiome Representation Learning
• 🧫 Microbial Genome Representation Learning
• 🤖 AI for Biology
• 🧠 Mechanistic Interpretability

MGM2 is a unified foundation model that transforms microbial composition and relative abundance into contextualized feature representations and fixed-size community embeddings.

It jointly models microbial identity and abundance through an abundance-aware Transformer, supporting both sequence-aware (OTU/ASV) and taxonomy-based microbiome profiles.

Highlights

• 🧬 Joint modeling of microbial identity and abundance
• 🌐 Context-aware microbial community representations
• 🔀 Support for both sequence-aware and taxonomy-only inputs
• 📊 Transferable community embeddings for downstream microbiome analysis
• 📦 Multiple model scales for flexible deployment

Microbiome · Foundation Model · Representation Learning

MicroVQVAE is a genome foundation model that converts ordered prokaryotic protein sequences into discrete, context-aware genome tokens.

Built upon PAIR-esm2 protein embeddings and vector-quantized representation learning, it captures genomic context while enabling compact, interpretable genome representations.

Highlights

• 🧩 Discrete genome tokenization
• 🔗 Context-aware genome representations
• 🌳 Phylogeny-aware representation learning
• 🔬 Functional discovery beyond sequence similarity
• ⚡ Scalable inference for newly sequenced genomes

Genome · Foundation Model · Vector Quantization

MGM is the first generation of our microbiome foundation models, learning transferable representations directly from microbial community composition for diverse downstream microbiome tasks.

Highlights

• 🌐 Context-aware microbial representations
• 🔄 Transferable embeddings across downstream tasks
• 🧪 Generalization across environments and host-associated microbiomes
• 🚀 Foundation model for microbiome representation learning

Microbiome · Foundation Model · Transfer Learning

• 🧬 Scaling microbiome foundation models
• 🧫 Genome representation learning with biological foundation models
• 🤖 AI agents for autonomous biological discovery
• 🧠 Mechanistic interpretability of biological foundation models

• 📧 Email: haohongzh@gmail.com
• 🐦 X: https://x.com/ludens_zhang
• 💼 LinkedIn: https://www.linkedin.com/in/haohong-zhang-549036242

Building foundation models to understand the microbial world.