산업 / 제약·바이오
APLUS Simulation CEO So Jin Lee “AI Prediction Engine Built on Data From 441 Antibodies Improves Human PK Predictions”
“What sets APLUS Simulation apart is that we have integrated machine learning directly into the pharmacokinetic prediction engine rather than using it merely as a supporting feature. We have substantially improved human pharmacokinetic predictions compared with conventional methods and successfully implemented the technology in working software.”So Jin Lee, CEO of APLUS Simulation and an adjunct professor at Sungkyunkwan University's School of Pharmacy, outlined the company's approach to developing artificial intelligence (AI)-based drug development software in an interview with Yakup News on Oct. 21 at its headquarters in Seoul's Seocho District. The company's strategy is to extend its expertise in commercial pharmacometrics modeling into AI-based prediction technologies and software development.APLUS Simulation has developed KINENTRIX, a platform designed to predict human pharmacokinetics (PK) using nonclinical data from therapeutic antibodies. The company is funding software development through revenue generated by its existing modeling services, without external investment, while building proprietary datasets and algorithms as technological assets. Within its first year, it has completed at least 25 projects and received a research award at an international scientific conference.▪ Commercial Pharmacometrics Expert Builds Clinical Development and Regulatory Experience Across Approximately 200 ProjectsLee joined Q-fitter in 2016 as one of its founding members. The company is widely recognized as South Korea's first specialized pharmacometrics service provider. She was among the country's early professionals to perform pharmacometric modeling and analysis as a full-time commercial service for pharmaceutical companies, at a time when such work was largely conducted in academic settings.She subsequently led a modeling team at AIMS BioScience, providing pharmacometric modeling and clinical pharmacology consulting for clinical development programs and global regulatory submissions.“At the time, modeling in Korea was primarily conducted at universities, but pharmaceutical companies needed more than academic analyses,” Lee explained. “An important part of commercial modeling is producing results suitable for regulatory submissions, conducting additional analyses during the review process and responding to questions from regulatory authorities.”In March 2026, Lee was the first author of a study published in the international journal Translational and Clinical Pharmacology. The study analyzed 192 projects in which she participated at Q-fitter and AIMS BioScience between 2016 and 2024.Of the 75 client companies, 45, or 60%, commissioned at least two projects. Oncology accounted for the largest share, with 89 projects, representing 46.35% of the total. Approximately 75% of all projects were either directly used in regulatory submissions or provided supporting evidence for filings with authorities including South Korea's Ministry of Food and Drug Safety (MFDS), the US Food and Drug Administration (FDA) and the European Medicines Agency (EMA).One notable example was the development of EPYSQLI (development code SB12), Samsung Bioepis' eculizumab biosimilar.While working at her previous organization, Lee participated in pharmacokinetic/pharmacodynamic (PK/PD) modeling using Phase 1 and Phase 3 clinical trial data and helped respond to regulatory inquiries. The analyses supported global regulatory approvals, including the US Biologics License Application (BLA). She also conducted first-in-human (FIH) dose predictions for bispecific antibodies, antibody-drug conjugates (ADCs) and gene therapies.▪ Integrating 441 Antibodies and 21 Prediction Methods With AI Directly Embedded in the PK EngineAllometric scaling extrapolates human PK from monkey PK data based on body-weight differences, but it may not adequately account for drug-specific characteristics such as molecular structure and target biology.APLUS Simulation collected nonclinical and clinical information on 441 approved and investigational therapeutic antibodies to build an integrated, machine learning-ready database. Using these data, the company developed a hybrid human PK prediction model that combines conventional modeling approaches with machine learning (ML).The research team organized the information into seven standardized datasets covering molecular characteristics, nonclinical and human PK, dosing regimens, in vitro experiments, population pharmacokinetic (PopPK) models, and pharmacodynamic and efficacy data.Researchers and interns spent more than six months manually reviewing published literature and using AI-based data extraction tools to cross-check numerical values and measurement units.The platform compares 21 prediction methods: one conventional allometric scaling approach, 10 hybrid methods that use machine learning to correct prediction errors from allometric scaling, and 10 pure machine learning methods that estimate human PK directly from drug characteristics.The pure machine learning methods are designed to predict human PK even when monkey PK data are unavailable.“Established commercial modeling software benefits from extensive accumulated datasets and validated models,” Lee said. “Our approach is different because we have integrated both hybrid and pure machine learning methods directly into the PK prediction engine.”The company filed a South Korean patent application related to the system in September.▪ Agreement in Human Drug Exposure Predictions Improves From 42% to 65%; Second-Place Award at ISSX 2026APLUS Simulation presented its KINENTRIX development research at the 27th North American Meeting of the International Society for the Study of Xenobiotics (ISSX), held in San Francisco from Oct. 11 to 14, and won second place in the Best Presentation Award category.The researchers evaluated prediction performance for human drug clearance (CL) using data from 138 antibodies and volume of distribution (V) using data from 147 antibodies.The best-performing approach for clearance was a hybrid random forest model, while a pure machine learning random forest model performed best for volume of distribution.Both methods achieved lower geometric mean fold error (GMFE) values than conventional allometric scaling. AI- and machine learning-based methods also produced lower prediction errors than conventional allometry in 20 of the 24 evaluable subgroups.In simulations involving 40 drugs, the proportion of predicted steady-state area under the plasma concentration–time curve over a dosing interval (AUCτ,ss) values falling within twofold of the corresponding PopPK reference values increased from 42% to 65%, an improvement of 23 percentage points.KINENTRIX uses predicted clearance and volume of distribution parameters to simulate plasma drug concentrations and systemic drug exposure.The platform enables comparisons across doses and dosing intervals and supports virtual patient analyses incorporating characteristics such as body weight, age and serum albumin levels.These capabilities can support dose selection in clinical trial design within a model-informed drug development (MIDD) framework. However, independent external clinical validation using data from new drug candidates remains an additional step in the continued development of the platform.“Our goal was not necessarily to create a perfect model, but to improve clinical predictive performance compared with conventional approaches,” Lee said. “Now that we have implemented the platform's underlying architecture, we can continue refining it by incorporating new data and validation results.”▪ 25 Projects Completed Within the First Year as Service Revenue Funds Proprietary Software DevelopmentAPLUS Simulation is developing AI software while continuing to provide its established drug development modeling services.According to the company, it completed 25 projects within approximately one year of its founding, including a modeling assignment for a company based overseas. Revenue from these projects is reinvested in research and development, allowing the company to continue operating without external investment.APLUS Simulation also conducted a machine learning-based clinical biomarker analysis for a bispecific antibody development program with a biopharmaceutical developer ranked first in the 2025 HIT evaluation.The team quantified multiple immune-cell biomarkers and measurements collected at different time points. Machine learning was then used to identify key biomarker candidates capable of distinguishing treatment responders from nonresponders.The biological significance of the results was evaluated in collaboration with the client's immunology and biomarker research teams.Building on this experience, APLUS Simulation plans to develop customized AI platforms that standardize pharmaceutical companies' proprietary nonclinical and clinical data and are optimized for specific therapeutic indications or drug candidate classes.The strategy is to expand beyond individual analytical assignments into customized software development. The company is also pursuing research and talent-development collaborations with Duksung Women's University and data-driven technology collaborations with OrganoPlus, a company specializing in organoid technology.“What pharmaceutical companies need is not simply data analysis, but predictive technology that can support drug development decisions,” Lee emphasized. “APLUS Simulation's strength lies in accurately understanding the development challenges its clients face and providing practical solutions through data and modeling, thereby supporting scientifically grounded decision-making and regulatory interactions.”▪ Expanding Into ADCs, Bispecific Antibodies and PD Prediction While Targeting the US MarketAPLUS Simulation plans to expand its predictive capabilities from human PK to pharmacodynamic (PD) responses.Its development plans include clinical PD prediction models incorporating oncology efficacy endpoints such as objective response rate (ORR) and progression-free survival (PFS), as well as specialized modules for ADCs and bispecific antibodies.For international business expansion, the company is targeting US biotechnology firms as a key customer segment.Its business model combines existing pharmacometric modeling services, which provide client relationships and revenue to support research and development, with KINENTRIX and customized AI platforms to expand its software business.The company's future enterprise value will depend on how extensively these proprietary technologies are adopted in clients' drug development programs and whether they generate independent revenue.“Our goal is to establish a competitive position in the global drug development market by combining the pharmacometrics modeling expertise we have accumulated in Korea with AI-based predictive technologies,” Lee said. “We intend to expand beyond conventional modeling services and develop an AI prediction platform that pharmaceutical and biotechnology companies worldwide can use in their drug development programs.”
Hyeokjin Kwon, Jisu Heo
2026.10.22