Understanding how biological systems respond to infection, stress, and engineered interventions remains a central challenge in modern biotechnology and health science. My research develops robust, integrative computational frameworks to simulate the dynamic behavior of metabolism in eukaryotic (plant, human and animal) systems under pathogenic (e.g., viral, bacterial) and non-pathogenic (e.g., cancerous) perturbations. By integrating time-resolved multi-omic data with mechanistic and data-driven modeling, these frameworks transform noisy, sparse biological measurements—such as those from metabolomics—into interpretable regulatory insights.
At the core of this work are scalable algorithms and simulation platforms that model enzyme kinetics, host–pathogen interactions, and metabolic adaptation with high temporal fidelity and mutational sensitivity. These tools not only enhance our system-level understanding of metabolism but also support translational goals in precision medicine, therapeutic design, and biodefense. My broader interests span systems biology, AI-integrated simulation, and predictive modeling for complex biological systems, with a strong drive to connect computational methods to real-world applications in pharmaceutical R&D, biomanufacturing, and digital health.
A flagship platform I co-developed and led through integration is RealKcat, a predictive enzyme kinetics classifier trained on the deeply curated KinHub-27k dataset of enzyme–substrate pairs. RealKcat fuses transformer-based protein language models and substrate-aware chemical embeddings with structure-guided active site detection, enabling interpretable prediction of kinetic parameters (kcat and KM) ranges at mutation-level resolution. Unlike traditional black-box AI tools, RealKcat offers clear attributions to catalytic residues and molecular features—facilitating enzyme variant prioritization, regulatory review, and guided protein engineering.
Complementing this, I also led the development of a dynamic flux inference and predictive pipeline, capable of converting time-series and isotope-labeled metabolomics into high-resolution flux maps. Applied to sphingolipid biosynthesis in plant systems, this method revealed stress-responsive bottlenecks and rate-limiting steps and is now being extended to model pancreatic cancer metabolism by incorporating kinetic constraints into thermodynamically feasible frameworks to expose metabolic vulnerabilities for therapeutic targeting.
Collaborative applications could span:
Ultimately, it is the mutational precision, temporal resolution, and system-wide interpretability of these frameworks that enable teams to extract actionable insights without the burden of costly experimental iterations. Whether leading platform development, mentoring junior scientists, or collaborating across disciplines, I am deeply committed to cross-functional innovation that translates molecular understanding into tangible solutions. By advancing transparent, modular, and scalable computational platforms, my goal is to equip research and industry teams alike with the predictive power and flexibility needed to transform biological potential into therapeutic, agricultural, and industrial impact—accelerating progress from molecule to market.