[Edaily Reporter Hong Ju-yeon ] Advances in artificial intelligence (AI) are reshaping the process of new drug development. Whereas researchers used to read papers and formulate hypotheses, AI now reads hundreds of thousands of papers and proposes hundreds of hypotheses, after which researchers simply select the most promising ones. Going a step further, we are now on the verge of a stage where AI designs experimental protocols and robots directly synthesize compounds. This is why big tech companies like Google DeepMind and NVIDIA are rushing to enter the “AI scientist” market.

In South Korea, BioNexus has entered this competition with its “AI-based bio-data intelligence and scientist agent platform for new drug development.” In an interview with Pharm eDaily, E-Daily’s premium pharmaceutical and biotech content service, Kim Tae-hyung, CEO of BioNexus, stated, “AI-driven drug discovery has now moved beyond simply accelerating the identification of candidate compounds; it is evolving to integrate literature searches, omics data analysis, hypothesis generation, experimental design, and business development (BD) strategies.”

One hypothesis a week → Hundreds a day

CEO Kim Tae-hyung is considered a pioneer of the first generation of bioinformatics in Korea. Bioinformatics is a field that addresses biological problems using applied mathematics, statistics, computer science, and AI, with a particular focus on analyzing and interpreting molecular-level data such as genomics and transcriptomics. He previously served as a division head at Teragen Bio and participated in projects to develop Korea’s first human genome map and the first whale genome map.

Having devoted 20 years to data, he decided to launch his startup in November 2024 because he felt the balance had been disrupted. As recently as the early 2000s, the cost of bio data itself served as a barrier to entry, but technological advancements have drastically reduced production costs. Conversely, the capabilities of the people needed to interpret that data have remained stagnant. CEO Kim explained, “At the time of founding, data in the biofield was exploding, but researchers lacked the time and tools to interpret it,” adding, “I founded the company because I believed generative AI had finally reached a level where it could bridge that gap.”

BioNexus develops and operates an “AI research partner” designed to bridge that gap. Centered around “NexusScience,” a core engine that mimics a scientist’s way of thinking, the platform consists of “NexusRAG,” a customized research search tool; “Nexus Co-Scientist,” a multi-agent-based collaborative research platform; and the drug and target analysis platform “NexusDrugLab.” The company is also developing “K-BioLLM-30B,” a Korean-style language model specialized for the biotech field, with the goal of releasing it within the year.

Nexus Co-Scientist learns from hundreds of thousands of papers organized by disease or research topic; AI agents assigned different roles—such as generation, review, and ranking—collaborate to derive hypotheses and even propose data analysis and experimental designs. For example, when attempting to verify the mechanism of action of a GLP-1 analog, traditional searches would be limited to the keyword “GLP-1,” but the AI agents expand the context to include single-cell data related to obesity, diabetes, and adipose tissue; they then download publicly available data and complete the analysis and summarization. CEO Kim stated, “Tasks that used to take more than six months can now be completed in half a day.”

Nexus Drug Lab, which handles compound screening, analyzes structures to determine their suitability as therapeutic targets and predicts how the compounds will be expressed and function within cells. The system is designed to reduce the number of experiments required. He explained, “By utilizing AI, we can reduce the number of experiments from 1,000 to 100, thereby minimizing the validation process.” While a human researcher might formulate one hypothesis per week, AI can generate hundreds in a single day.

The next hurdle is experimentation… “A process that used to take up to two years will be shortened to at least one month.”

CEO Kim identified “experimentation” using AI as the next step. “With AI, hypotheses emerge quickly and data analysis can be completed in a day or two

,

but the challenge lies in validation,” he said, adding that he plans to boost efficiency through an automated laboratory.

BioNexus is currently preparing to validate its “AI Anti-Aging Research Institute,” located in Cheongsong County, North Gyeongsang Province, along with its automated laboratory. The system is designed so that AI generates hypotheses and experimental protocols and even performs protein synthesis. The goal is to automate everything from peptide and antibody production to cellular response experiments involving compounds and peptides, and plans for equipment setup have already been established. When the AI optimizes sequences to design hundreds of combinations, the lab synthesizes and measures them; the AI then interprets the results and proposes the next design. CEO Kim stated, “We will create a system where AI can handle everything from literature searches and data analysis to hypothesis generation, experimentation, and result interpretation,” adding, “Our goal is to shorten a process that used to take one to two years to just one to two months, and in the long term, to automate preclinical trials as well.”

Speed isn’t the only reason the lab is needed. Kim attributes the AI’s current inability to distinguish between valid and invalid data to its training data. Until now, the AI has been trained solely on successful experimental data published in academic papers. “With a lab in place, the AI can also learn from failure data,” Kim explained. “Currently, humans intervene to filter out the failures, but by learning from failure data, the AI can become as intelligent as an expert.” Currently, among the hundreds to thousands of hypotheses generated by AI, only about 10% are recognized as valid by experts; however, learning from failure data could significantly increase this ratio.

BioNexus’s ultimate vision is a digital twin. The company is currently carrying out the “Virtual Cell” project, a national research initiative with a budget of 8 billion won. Based on the premise that each cell reacts differently to drugs, the project involves accumulating single-cell sequencing data to create virtual cells that can predict results equivalent to those obtained from actual drug treatment without actually administering the drugs. The immediate priority is to generate and accumulate data from dozens of cell lines. The goal is to reach a stage where experimental results can be known without actually conducting the experiments.

Business Model: Platform and Compounds… Targeting an IPO by 2030

BioNexus has a two-pronged business

model

. The company simultaneously operates two lines of business: providing platforms to researchers and directly developing compound assets derived from those platforms. The company enters into approximately 10 joint research agreements per half-year, including large-scale research consortia worth over 10 billion won. The structure calls for the shared ownership of the compound assets derived from these agreements. BioNexus has signed agreements with organizations such as the Korea Basic Science Institute (KBSI) and the Korea Institute of Ocean Science and Technology (KIOST) under which BioNexus provides the AI hypotheses, the partner institutions cover the experimental costs, and the patents are jointly owned.

Results are already emerging. BioNexus has secured contracts with the Korea Disease Control and Prevention Agency, the National Institute of Health, the Korea Bioinformatics Center (KOBIC), and the Korea Research Institute of Bioscience and Biotechnology (KRIBB) for projects involving large language model (LLM)-based data curation, rare disease mutation analysis, and the construction of a bio-big data platform. The company is also collaborating with Upstage and Rebellion to develop an “NPU-based Korean-language bio-AI research platform.”

The company is receiving approximately 2 billion won annually through government projects, and it expects to generate 2 to 3 billion won in revenue this year from projects with institutions seeking to adopt AI-based bio technologies. The company explains that its business foundation is solid, with a net profit margin reaching 30–40% and GPU support provided through government grants. The company recently closed a 1 billion won seed investment round and plans to secure Series A funding in the second half of the year. It envisions sustaining annual growth of approximately threefold to achieve 100 billion won in revenue and over 30 billion won in operating profit by 2030. It also aims to go public that same year.

CEO Kim stated that the ultimate goal is to realize “popular science” through AI. To this end, the company is working to increase accessibility by adapting BioNexus’s solutions—currently used primarily in research laboratories—for use by individuals. CEO Kim said, “By collaborating with AI, writing a research paper—which previously took over a year—can now be done in just one week.” He added, “Within three years, we will enter an era where anyone can work like a researcher without having to pursue a master’s or doctoral degree, and an era will dawn where even high school students can develop new drugs and write research papers.”