Artificial neural network (ANN)-based prediction of depth filter loading capacity for filter sizing

Area of Science:

• Biochemical Engineering
• Process Systems Engineering
• Biopharmaceutical Manufacturing

Background:

• Depth filter loading capacity is critical for efficient monoclonal antibody (mAb) purification in commercial manufacturing.
• Accurate prediction of filter performance is essential for process optimization and cost reduction.
• Traditional methods may not fully capture the complex interactions of operating parameters affecting filter lifespan.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting depth filter loading capacity.
• To evaluate the impact of various operating parameters on filter loading capacity.
• To demonstrate cost savings through optimized filter area sizing using the developed ANN model.

Main Methods:

• Application of artificial neural network (ANN) modeling using inlet stream properties (turbidity, cell count, viability), flux, and time.
• Analysis of differential pressure (DP) changes over time to assess filter loading.
• Training the ANN model with 174 training, 37 validation, and 37 test points, with a sigmoidal activation function.
• Utilizing a pressure cut-off of 1.1 bar for filter area sizing.

Main Results:

• The ANN model demonstrated excellent agreement with experimental data, achieving a regression coefficient (R²) of 0.98.
• The model successfully predicted differential pressure (DP) and was used for variable depth filter sizing.
• Monte Carlo simulations indicated potential cost savings of 10% in the cost of goods.

Conclusions:

• ANN modeling provides a robust and accurate method for predicting depth filter loading capacity in mAb manufacturing.
• Optimized filter area sizing based on ANN predictions leads to significant economic benefits.
• This approach enhances the efficiency and cost-effectiveness of biopharmaceutical purification processes.