Analysis of cytokine release assay data using machine learning approaches.

Area of Science:

• Immunology
• Biotechnology
• Computational Biology

Background:

• Cytokine Release Syndrome (CRS) is a critical safety concern in monoclonal antibody (mAb) therapeutic development.
• Assessing the cytokine-inducing potential of mAbs is essential for predicting and mitigating CRS.
• Existing methods for analyzing complex biological data can be enhanced by machine learning.

Purpose of the Study:

• To apply and evaluate multiple machine learning approaches for analyzing in vitro human blood assay data related to mAb-induced cytokine release.
• To identify which machine learning techniques are most effective in characterizing CRS.
• To determine key cytokine biomarkers associated with CRS and explore potential novel correlations.

Main Methods:

• Utilized Hierarchical Cluster Analysis (HCA) to understand data clustering.
• Employed Principal Component Analysis (PCA) followed by K-means clustering for sample classification and visualization.
• Applied Decision Tree Classification (DTC) to identify important cytokine predictors of CRS.

Main Results:

• All three machine learning approaches successfully identified treatments causing the most severe cytokine responses.
• HCA provided insights into the number of data clusters.
• PCA with K-means enabled sample-level classification and visualization of treatment clusters.
• DTC models highlighted the significance of IFN-γ, TNF-α, and IL-10 in CRS, and suggested a potential correlation with IL-17.

Conclusions:

• Tandem application of HCA, PCA/K-means, and DTC offers a robust and complementary strategy for analyzing mAb-induced cytokine release data.
• Machine learning facilitates improved parameter selection and data analysis for mAb therapeutic development.
• DTC analysis identified key cytokines (IFN-γ, TNF-α, IL-10) and suggested IL-17 as a potential CRS biomarker, warranting further investigation.