由4位牛津大学校友创办的津渡生科自主研发的生命科学垂类大模型GeneLLM已在《Nature Communications》和《Advanced Science》等顶级期刊发表研究成果[1]。作为全球首个直接使用组学原始数据进行预训练的多组学大模型，GeneLLM已完成15亿参数和3.5万亿碱基序列的预训练，其XLarge版本实现了300亿参数的模型预训练[1]。该模型在1Gb极浅深度测序下仍能保持AUC > 0.8的性能表现，相比之下传统方法需要依赖6Gb深度测序，成本因此可缩减83%[1]。
津渡生科开发的BioFord Harness系统将AI与物理实验室打通，实现从科学假设生成到自动化实验验证的完整闭环[1]。其中BioFord Agent平台包含文献检索、实验设计、科学、实验调度和数据分析五大智能体[1]，已在国内知名高校落地应用，将科研周期从原本的数月缩短至一周[1]。GeneLLM可应用于新药研发、精准医疗等多个领域[1]。公司创始人金泳成的导师是英国皇家学会会员、Oxford Nanopore创始人Hagan Bayley[1]。
2025年，津渡生科完成了4轮融资[1]。其中天使+轮获红杉中国种子基金领投，Pre-A+轮分别获创东方投资和南山战新的千万级投资，A轮则获高特佳投资近亿元领投[1]。
Jingdu Bioscience, founded by four University of Oxford alumni, has developed GeneLLM, a multi-omics large language model designed specifically for life sciences applications [1]. The model represents the first globally to conduct pretraining directly using raw omics data, completing pretraining on 15 billion parameters and 3.5 trillion base sequences, with an XLarge version achieving 300 billion parameters [1]. Research based on GeneLLM has been published in top-tier journals including Nature Communications and Advanced Science [1].
The model demonstrates significant practical advantages in genomic analysis. GeneLLM maintains an AUC greater than 0.8 even at extremely shallow sequencing depth of 1Gb, reducing costs by 83 percent compared to traditional methods that require 6Gb sequencing depth [1]. The company's founder, Kim Yongseong, studied under Hagan Bayley, a Fellow of the Royal Society and founder of Oxford Nanopore [1].
Beyond the language model itself, Jingdu Bioscience has developed the BioFord Harness system, which integrates AI with physical laboratory infrastructure to create a complete workflow from scientific hypothesis generation to automated experimental validation [1]. The BioFord Agent platform comprises five intelligent modules: literature retrieval, experimental design, scientific analysis, experiment scheduling, and data analysis [1]. After deployment at leading domestic universities, the platform has compressed research cycles from months to approximately one week [1].
The company completed four funding rounds in 2025: an angel-plus round led by Sequoia China Seed Fund, a Pre-A+ round with lead investment from Chuangdongfang Capital at the million-yuan level, a Pre-A+ round with million-yuan investment from Nanshan Zhangxin Capital, and an A round led by Gaotech Investment with nearly 100 million yuan [1].
