Science
An AI digital twin for reaction-centric drug design
Drug systems that undergo chemically controlled activation, transformation, or release need a framework that considers molecular structure, reaction, kinetics, and biological context together - so predictions stay mechanistically relevant.
The thesis
Traditional discovery pipelines are often disconnected. They do not always represent how a drug will act as a dynamic chemical system in a human biological context. SPADE Twin - Sequence-Programmed Active Drug Engineering - is proposed as an AI digital twin for reaction-centric design that closes that gap.
At the molecular level
Drug systems are represented with structure-based encodings, augmented with chemically informed descriptors that capture steric properties, local bonding environments, and related electronic effects. These features feed a deep learning architecture that learns non-linear relationships between molecular profile and reaction behavior. The framework is designed to predict reaction-relevant results: activation tendency, intermediate formation, and transformation pathways.
At the reaction level
Learned molecular representations are combined with reaction-engineering principles to predict kinetic behavior, identify rate-limiting steps, and describe conversion over time. That lets the twin simulate how a therapeutic system changes under particular conditions - including when, where, and how activation may happen. By embedding mechanistic pathway reactions in an AI-driven framework, the twin retains physical understanding while expanding predictive scope.
Biological relevance
The framework connects chemical transformations to downstream processes such as transport, release kinetics, and functional activity. In prodrug systems, this supports analysis of how molecular modifications influence activation timing and therapeutic effects.
Motivating application
Cancer-relevant prodrugs are a primary motivating application: activation chemistry, timing, and tissue context all matter at once. Well-known public examples (such as oral prodrugs of established cytotoxic agents) illustrate the class of design questions - without publishing our private candidate packs or internal worked pipelines.
What this contributes
In summary, SPADE Twin is framed as a rapid virtual screening platform for reaction-centric drug design, aiming to reduce reliance on purely costly, labor-intensive trial-and-error validation. By integrating deep learning with chemical kinetics and systems-level understanding, the work aims to enhance - rather than replace - classical reaction engineering, and to accelerate data-driven design for transformable therapeutic systems.
Related research direction
Separately, our group also explores a multiscale AI research direction for precision medicine: connecting drug response across biological scales (from molecular transformation through cellular and metabolic context to whole-body disposition) in one predictive framing. Prodrugs are an important case study for that work because activation chemistry and system-level response both matter. That is a related research direction - not the same product scope as SPADE Twin's reaction-centric prodrug focus on this site.
Scientific spine
01
Molecule
Structure encodings and chemically informed descriptors
02
Reaction
Kinetics, rate-limiting steps, conversion over time
03
Biology
Transport, release, activation timing and relevance
SPADE Twin is developed by Sabyasachi Mohanty, Samarpan Mohanty, and Mona Bavarian, Department of Chemical & Biomolecular Engineering, University of Nebraska–Lincoln, in the context of AIChE 2026 research communication on digital twins and reaction engineering in pharmaceuticals.