A Modular Mechanistic In Silico Model for In Vitro Transcription Process Yield and Product Quality Prediction

In vitro transcription (IVT) plays a critical role in the manufacture of mRNA vaccines and therapeutics. Optimizing mRNA yield and ensuring product quality, such as capping efficiency and integrity, are essential but mechanistically complex. This study presents a modular mechanistic model of the IVT process to advance scientific understanding and improve predictive capability. The IVT reaction network is decomposed into interconnected modules describing (1) initiation and capping, (2) elongation and truncation, (3) termination and read-through, (4) mRNA degradation, (5) magnesium pyrophosphate precipitation, and (6) enzymatic degradation of pyrophosphate. Guided by biochemical principles and experimental data, kinetic models were developed for each module, accounting for mass balances, molecular complexation, and enzyme activity, and were subsequently assembled to capture coupled IVT dynamics. Multivariate residual analysis and Shapley value-based sensitivity analysis, guided by domain knowledge, were applied to iteratively improve model fidelity. These machine learning-driven analytics enabled identification of key mechanisms, supported in silico experimentation, and facilitated root-cause analysis. Combined with Gaussian-process-based batch Bayesian optimization for efficient parameter estimation, this framework establishes a scalable hybrid (mechanistic + machine learning) modeling platform that integrates heterogeneous data, accelerates model calibration, and supports rational design and optimization of mRNA manufacturing processes.

💡 Research Summary

This paper presents a comprehensive mechanistic‑in‑silico framework for predicting both yield and critical quality attributes (CQAs) of in‑vitro transcription (IVT), a pivotal step in the manufacture of mRNA vaccines and therapeutics. Recognizing that existing models either focus narrowly on reaction kinetics or rely solely on data‑driven surrogates, the authors develop a modular kinetic model that decomposes the IVT reaction network into six interconnected sub‑models: (1) initiation and co‑transcriptional capping, (2) elongation and premature truncation, (3) termination and read‑through, (4) mRNA degradation, (5) magnesium‑pyrophosphate (Mg‑PPi) precipitation, and (6) enzymatic pyrophosphate (PPi) degradation. Each module is expressed through mass‑balance differential equations incorporating enzyme–substrate complex formation, Michaelis‑Menten/Hill kinetics, and fast‑equilibrium assumptions for precipitation reactions.

The modules are linked to capture the full dynamics of full‑length, truncated, and over‑extended transcripts, each possibly capped or uncapped. Parameter estimation is performed on a heterogeneous dataset that includes discrete batch runs, time‑course measurements, and data from multiple reactor formats (tube, 96‑well, AMBR, EasyMax). To identify systematic model deficiencies, the authors apply multivariate residual analysis, revealing bias under high Mg²⁺ or low pH conditions, which they trace to under‑parameterized Mg‑PPi precipitation kinetics. Sensitivity analysis based on Shapley values quantifies the contribution of each input variable (template concentration, RNAP activity, NTP ratios, Mg²⁺ concentration, pH, temperature, etc.) to output variance, highlighting the nonlinear dominance of Mg²⁺ and PPi levels on both yield and capping efficiency.

Because the model is computationally intensive and the parameter space is high‑dimensional, a Gaussian‑process (GP) based batch Bayesian optimization (BO) scheme is employed. This approach efficiently explores the parameter landscape, leveraging parallel simulations to accelerate convergence. The resulting calibrated model achieves a mean absolute error (MAE) of 0.95 g L⁻¹ for mRNA yield (Spearman ρ = 0.94), 4.01 % MAE for integrity (ρ = 0.86), and 7.06 % MAE for 5′‑capping efficiency (ρ = 0.84).

In silico experiments using the validated model elucidate key mechanistic insights: (i) Mg‑PPi precipitation markedly reduces free Mg²⁺, limiting RNAP activity; (ii) enzymatic PPi degradation mitigates precipitation and restores Mg²⁺ availability; (iii) optimal Mg²⁺ levels must balance enzyme activation against precipitation risk; (iv) co‑transcriptional capping competes with elongation for RNAP, making capping efficiency sensitive to RNAP concentration and NTP availability. The authors demonstrate that adjusting Mg²⁺ concentration together with pyrophosphatase dosing can simultaneously improve yield and capping efficiency, providing actionable guidance for process development.

The modular architecture ensures extensibility: new caps, modified nucleotides (Ψ, m¹Ψ, s²U), or alternative polymerases can be incorporated by updating the relevant module without rebuilding the entire model. This flexibility, combined with the hybrid mechanistic‑ML workflow, offers a scalable platform for rapid iteration, root‑cause analysis, and rational design of IVT processes across different mRNA products.

Overall, the study advances the state of the art by integrating detailed biochemical mechanisms with modern machine‑learning analytics and Bayesian optimization, delivering a predictive tool that bridges fundamental understanding and practical process optimization for next‑generation mRNA manufacturing.