Bidirectional time-series state transfer network: a computational framework for target-directed control optimization of metabolic processes.

Engineered microbial cell factories enable efficient and sustainable biomanufacturing, yet their industrial performance remains constrained by the lack of state-aware process control. Existing strategies typically rely on static setpoints or pre‑optimized policies, which fail to accommodate nonlinear metabolic dynamics, irregular sampling, and batch‑to‑batch variability. Here, we introduce Tac‑BTSTN, a computational target‑directed control optimization framework that learns controlled system dynamics directly from irregular time-series data. Tac‑BTSTN explicitly models the coupled progression of system states and control inputs, enabling accurate trajectory prediction and gradient‑based optimization of multi‑stage control strategies toward predefined target states. Through computational evaluations across theoretical dynamical models and a real-world transcriptomic dataset, Tac‑BTSTN demonstrates superior predictive accuracy, robustness to missing and noisy data, and precise in silico target tracking. By unifying state inference and control optimization within a single data‑driven framework, Tac-BTSTN provides an algorithmic basis for the development of intelligent and adaptive biological-process control systems. Experimental validation in real-world closed-loop fermentation setups and demonstration of product-yield improvement remain to be established.