(174x) Prediction of Thawing Rates and Stability of Large Scale Frozen Biopharmaceutics Using Machine Learning

Freezing and thawing (F/T) of bulk protein solutions is a common processing step used to maintain stability and quality of biopharmaceutical products during development and production. Although F/T of proteins offers numerous advantages, the stresses from freezing and thawing can result in the loss of protein stability and functionality. Therefore, F/T of proteins need to be better understood to maintain stability and quality of proteins during manufacturing. The overall objective of this research is to use a validated machine learning model to predict the thawing rate and the stability of our model protein, BSA after every F/T cycle. Protein stability is evaluated quantitatively using protein aggregation studies such as dynamic light scattering (DLS) and high-performance liquid chromatography (HPLC). Thawing rates and parameters evaluated from protein aggregation studies are obtained experimentally and are then used in the training of the developed ML model to make predictions.

Methods:

An Artificial Neural Network (ANN) model is used to predict the thawing rates and stability of model protein, BSA, after every F/T cycle. Data driven ML models such as the ANN models, have been chosen as they perform well in the case of non-linear experimental data, and they also have the capability of extrapolating data to overcome the complexity and cost burden of experimental work. The input parameters such as protein solution concentration, surface area to volume ratio, loading distance between bottles while thawing, etc. were used as input features in the developed ANN model and the thawing rate and stability parameters were extracted from the model as predictions. Multiple-input-single-output models were implemented in Matlab using the Levenberg-Marquardt (LM) algorithm. In the experimental side, a protein surrogate solution and a protein solution were used to experimentally determine the thawing rate. Both the surrogate and the protein solutions were frozen for 24 hours in a -80°C freezer and then thawed. Thawing rates (°C/min) were calculated based on temperature differences between the starting and ending points of experiments and the time taken to reach the end points from the starting points.

Results:

The goal of the experiments was to calculate the thawing rates and quantify stability in terms of particle aggregation or monomer recovery for the protein solution. The experimentally obtained values were then used to train the developed AI-ML model to make predictions of protein thawing rate, protein surrogate thawing rate and protein stability as functions of F/T process parameters and material attributes. The AI-ML model performance was evaluated using Mean Relative Error (MRE) between the ground truth values and predicted values.