Characterizing Layer-Specific Mechanical Properties of Soft Materials by Pipette Aspiration Using Transformer Model and SHapley Additive exPlanations

Characterizing the mechanical properties of soft materials is essential for understanding the relationship between structures and properties in biological systems, disease diagnosis, and tissue engineering applications. Many biomaterials possess complex, multilayered structures, posing significant challenges in characterizing the mechanical properties of individual layers. This study presents a novel pipette aspiration method based on an end-to-end transformer-based deep learning model to address these challenges. Dimensional analysis is employed to extract governing dimensionless parameters, enhancing model interpretability and data analysis efficiency. A large-scale dataset generated through finite element (FE) simulations is used to train the transformer model, which predicts the layer-specific mechanical properties directly from experimental curves and geometric parameters. To ensure transparency and provide physical insights, we invoke SHapley Additive exPlanations (SHAP), an advanced interpretability technique, to analyze feature contributions and highlight critical factors influencing predictions. Validation with experimental data demonstrates the robustness and practical applicability of our method. This work represents a significant advancement in mechanical property characterization, offering a data-efficient, interpretable, and experimentally validated approach to evaluating layer-specific properties of multilayered soft materials.