Machine Learning Models for Improved Cell Screening

In the cell engineering pharmaceutical industry, the quality of the cell screening process directly affects the quality of monoclonal cell lines, which in turn impacts the cost and efficiency of the entire production process. Current screening methods largely rely on time-consuming and costly manual operations, not only increasing the risk of subjective bias but also reducing the efficiency and quality of screening. To address this challenge, this paper proposes a method based on binary classification prediction, introducing two machine learning models: the Stacked Machine Learning Model (SMLM) and the Simple Linear Model (SLM). SMLM makes initial predictions using four different base learning models, which are then integrated through a shallow artificial neural network to produce a final prediction. In contrast, SLM employs a deeper artificial neural network for direct prediction. We analyzed these two models from the perspectives of generalizability and interpretability and conducted a quantitative evaluation using accuracy as the primary metric. The results show that while SMLM offers stronger interpretability, SLM performs better on multiple evaluation metrics and has a simpler structure. This research not only provides a new direction for the automation of cell screening but also opens up new pathways for the development of the cell engineering pharmaceutical industry.