Machine Learning-Driven Techno-Economic Uncertainty Analysis in Batch Pharmaceutical Manufacturing

Active pharmaceutical ingredients (APIs) are produced in batch multiproduct facilities, where efficient scale-up is crucial for reducing costs and improving productivity. Traditional uncertainty analyses rely on extensive simulations, making them time-consuming and resource-intensive. This study introduces a machine learning (ML)-based framework to estimate techno-economic uncertainties in batch API production, streamlining cost and profitability assessments. Using a pharmaceutical production simulation model for the synthesis of an API via the condensation of quinaldine and hydroquinone, an annual API output of 33,000 kg was estimated, with a baseline unit production cost (UPC) of USD 175/kg. ML predictions showed UPC variations between USD 140/kg and USD 240/kg, impacting the minimum product selling price (MPSP), which ranged from USD 262/kg to USD 525/kg to achieve an internal rate of return (IRR) of at least 30%. An uncertainty analysis revealed that MPSP fluctuates between USD 200/kg and USD 900/kg within a 90% confidence interval. The probability of profitability is only 5% at lower prices but rises to 90% when the MPSP approaches USD 900/kg. The proposed ML framework reduces the computational burden associated with traditional sensitivity analyses by using distinct ML models, including tree-based algorithms, instance-based methods, linear and polynomial regressors, and kernel-based models. This approach provides a fast and effective tool for estimating the UPC and MPSP distributions. With the dataset in hand, the ML-driven techno-economic assessment (TEA) can be used independently of any additional software, making it highly valuable for cost assessment in batch production across pharmaceutical, chemical, and biochemical industries.