Predicting purification process fit of monoclonal antibodies using machine learning.

Area of Science:

• Biotechnology
• Protein Engineering
• Computational Chemistry

Background:

• Early-stage therapeutic monoclonal antibody (mAb) development faces challenges with extensive experimentation for purification assessment.
• Timeline pressures and material constraints limit the evaluation of numerous molecules and process conditions.
• High-throughput screening data and advanced molecular descriptors have enabled predictive modeling for mAb behavior.

Purpose of the Study:

• To introduce and validate a quantitative structure-property relationship (QSPR) strategy for *in silico* assessment of mAb purification process fit.
• To leverage sequence-based predictions for efficient selection of purification resins and operating conditions.
• To reduce the experimental workload in early-stage biopharmaceutical development.

Main Methods:

• Application of Principal Component Analysis (PCA) to reduce multi-dimensional batch-binding data into a single dimension representing chromatographic behavior.
• Utilizing Kernel Ridge Regression (KRR) to predict the principal component for novel mAb sequences.
• Benchmarking four distinct descriptor sets, including biophysical structural models and protein language models, for model development.

Main Results:

• A QSPR workflow was successfully demonstrated for 97 mAbs across five chromatography resins and varying buffer conditions (pH, salt concentration).
• The models accurately predicted the principal component of chromatographic binding behavior from amino acid sequences.
• Performance comparison of different descriptor sets highlighted the utility of various feature extraction methods.

Conclusions:

• QSPR models offer a powerful computational tool for *in silico* purification process fit assessment of therapeutic mAbs.
• Sequence-based prediction enables informed selection of chromatography resins and operating parameters early in development.
• This strategy significantly alleviates experimental demands, accelerating the development timeline for biologics.