Physicochemical-informed predictive modelling on small datasets for designing conductive polymer inks in soft bioelectronics

Designing conductive polymer inks for bioprinting presents challenges due to the interplay between polymer chemistry, conductive performance, and fabrication process. Herein, we built an AI-ready surrogate that integrates physicochemical feature engineering to predict and optimise the conductivity of inks for bioelectronic applications. The conductivity at 1000 Hz served as the targeted outcome for machine learning model evaluation. On a held-out test set, MAE values ranged from 0.14 to 0.26, with R ² values of 0.40–0.87. Stacked ensemble model (Ridge + XGBoost) emerged as the best-performing model, highlighting its ability to capture both linear and non-linear relationships. Our closed-loop workflow includes the exploration of over 800 samples in the virtual data space using Latin Hypercube Sampling to identify candidate formulations with high conductivities. Next, we assessed the processability of the predicted formulation for 3D bioprinting. The formulation is printable and biocompatible, demonstrating its potential for application in soft tissue bioelectronic interfaces. This workflow accelerates the discovery of high-performance conductive inks and provides mechanistic insights into polymer chemistry and electrochemical functions. It demonstrates the potential of physicochemical-informed AI in assisting formulation design with small datasets, enabling the rational development of 3D printable materials for bioelectronics and advanced biofabrication.