FibreCastML : an open web platform for predicting electrospun nanofibre diameter distributions for biomedical applications

Introduction Electrospinning is a scalable technique for generating fibrous scaffolds with tunable micro- and nanoscale architectures for tissue engineering, drug delivery, and wound care. Machine learning (ML) has emerged as a powerful tool to accelerate process optimisation; however, existing models typically predict only mean fibre diameters, overlooking the entire diameter distribution that governs scaffold functionality and biomimicry. This study introduces FibreCastML, the first open-access, distribution-aware ML framework that predicts full fibre diameter spectra from routinely reported processing parameters and provides interpretable insights into parameter influence. Methods A comprehensive meta-dataset of 68,538 fibre-diameter measurements from 1,778 studies across 16 biomedical polymers was curated. Six standard input parameters (solution concentration, voltage, flow rate, tip-to-collector distance, needle diameter, and rotation speed) were used to train 7 ML learners (linear model, elastic net, decision tree, multivariate adaptive regression splines, k-Nearest Neighbours, random forest, and radial-basis Support Vector Machine) under nested cross-validation with leave-one-study-out external folds to ensure generalisable performance. Model interpretability combined variable importance, SHapley Additive exPlanations (SHAP), correlation matrices, and 3D parameter maps. The FibreCastML web app integrates these capabilities with out-of-range detection, solvent suggestions, and automated Excel reports. Results Non-linear and local learners consistently outperformed linear baselines, achieving R 2 > 0.91 for polymers such as cellulose acetate, Nylon-6, Polyacrylonitrile, polyD,L-lactide, Polymethyl methacrylate, Polystyrene, Polyurethane, Polyvinyl alcohol and Polyvinylidene fluoride. Concentration emerged as the most influential variable globally. The FibreCastML app returns polymer-specific distribution plots, predicted-vs-observed diagnostics, feature importance and correlations, and transparent metrics ( R 2 , RMSE, mean absolute error) for user-defined settings. In an experimental validation case using different electrospinners and microscopies, predicted diameter distributions closely matched experimental measurements (Kolmogorov–Smirnov p > 0.13 and overlap coefficient of 84%). Discussion By shifting from mean-centric to distribution-aware modelling, this work establishes a new paradigm for electrospinning design. FibreCastML enables reproducible, sustainable, and data-driven optimisation of scaffold architecture, bridging experimental and computational domains. Openly available, it empowers laboratories worldwide to perform faster, greener, and more reproducible electrospinning research, advancing sustainable nanomanufacturing and biomedical innovation.