Key points are not available for this paper at this time.
Ingredient substitution has become a multidimensional challenge in modern food systems, where sensory authenticity, functional performance, nutritional equivalence, and cultural or regulatory compliance must be satisfied simultaneously. This review examines how artificial intelligence (AI) can contribute to this problem by synthesizing current advances across four scientific domains relevant to substitution: flavor perception, matrix functionality, nutrient bioavailability, and socio-regulatory constraints. The review follows a narrative, domain-focused approach rather than a systematic or quantitative protocol, with literature selected from Scopus, Web of Science, and Google Scholar to capture both foundational food science studies and emerging AI applications. A modular framework for AI-enabled ingredient substitution is proposed and structured around four domains: (1) flavor and aroma modeling, (2) functional property prediction, (3) nutritional profiling, and (4) constraint-based filtering. The framework brings together a range of AI techniques-including machine learning, graph neural networks, natural language processing, and multi-objective optimization-and connects them to domain-specific datasets such as volatile compound libraries, rheological measurements, dietary reference databases, and regulatory ontologies. The review identifies three major gaps limiting real-world deployment: the lack of multimodal datasets linking composition, perception, and processing; limited explainability of current AI models; and weak integration between computational outputs and regulatory or industrial workflows. Addressing these barriers will be essential for developing transparent, scalable, and context-aware substitution systems that align with future directions in sustainable and personalized food innovation.
Öz et al. (Mon,) studied this question.