The work presented here pioneers the fusion of synthetic data generation and deep learning techniques to predict sensory perception in plant-based food models modified with hydrocolloids and novel proteins, such as fungal polysaccharides and insect proteins. The aim of this work is to demonstrate the capabilities of predictive modeling and transfer learning in deep neural networks for data derived from complex experiments in food and sensory science. We engineered a comprehensive, literature-informed dataset that mimics human ratings across 10 smoothie formulations, with each formulation either unmodified or systematically modified using fungal polysaccharides or insect proteins. By integrating rheology (shear, extensional), oral tribology, surface activity parameters, and human anatomical and physiological parameters, we trained a multilayer perceptron regression model - MLPR (Figure 1), capable of accurately predicting 10 sensory food texture attributes (MSE < 0.13; Test R² = 0.91). Our results reveal that fungal polysaccharides drive shear thinning and extensional resistance, while insect proteins amplify bulk and surface activity, mirroring the known behavior of hydrocolloids. The trained model not only learned complex sensory-rheology interactions but also demonstrated predictive modelling power with transfer learning on unseen conditions. Experiments like these demonstrate a significant gap in utilizing models for complex data. Finally, this framework provides a proof-of-concept tool to complement traditional sensory science by offering a scalable, rapid, and cost-efficient platform for simulating consumer responses, which enables the opening of a new frontier in intelligent food design, food ingredient screening, and novel food formulation.