Track I Large-Scale Population Cohorts: Subtrack 4. Genomic Foundation Model (Genos)(J28003)

Foundation Model

Establish a standardized tokenization framework based on multi-species, multi-omics datasets, integrating advanced techniques such as the Gengram memory module, multi-token prediction, and DNAChunk.

Develop a globally leading, human genome–centered ultra-long-context foundation model architecture capable of modeling genomic sequences at the megabase (million–base pair) scale with single-nucleotide resolution.

Construct Genos, a genomic foundation model that supports million–base pair–level contextual modeling with single-base precision.

Vertical Applications

Provide foundational model support for downstream domains such as chronic diseases, maternal and child health, brain health, and microbiome research.

Empower studies on variant pathogenicity interpretation, disease–phenotype association analysis, and related investigations.

Construct a genome semantic atlas based on natural population evolution, open model capabilities to the research community, and promote interdisciplinary integration in genomics as well as the development of a collaborative scientific ecosystem.

Affiliated Branches

• Guangdong–Hong Kong–Macao Branch (Shenzhen)
• Central China Branch (Wuhan)

Hangzhou Branch (Hangzhou)

(1) Preferred Background

Computer Science (LLMs/AI Systems), Bioinformatics, Computational Biology, Mathematics, or Statistics.

(2) Core Technical Skills

Strong understanding of Transformers and large-scale pretraining; distributed training experience; ability to design tokenization or long-context modeling strategies.

(3) Modeling Competency

Experience with foundation model development, MoE architectures, or long-sequence modeling; familiarity with genomic data representation.

(4) Preferred Qualifications

Published research in AI or computational genomics; experience training models at scale (multi-GPU/cluster environment).

(5) Personal Traits

Algorithmically rigorous, comfortable with abstraction, resilient under research uncertainty, and motivated by building foundational AI systems.