AI & Mechanistic Neuroscience: Hybrid Models Beyond Prediction

The intersection of artificial intelligence and mechanistic neuroscience is rapidly transforming our understanding of neural systems. While AI methodologies, particularly deep learning, have excelled in reproducing behavioral and neural data, a key challenge remains: moving beyond black-box prediction to achieve interpretable, mechanistically grounded models that support causal explanation and intervention. Recent progress highlights the potential of hybrid approaches that integrate the predictive strengths of AI with the explanatory power of mechanistic models, fostering deeper insights into neural function, computation, and pathology across scales ranging from molecular to behavioral. Despite these advancements, the field is grappling with questions about how best to blend these paradigms for maximal scientific and translational benefit.Cutting-edge studies have shown that hybrid models can capture complex neural dynamics and behaviors, offering a promising route to bridge data-driven and theory-driven research. Efforts combining machine learning with biophysically realistic circuits, or leveraging AI for hypothesis generation and model selection, have yielded new frameworks for studying the brain at multiple scales. In parallel, mechanistic and quantitative systems pharmacology frameworks are increasingly paired with AI to link biology to outcomes and accelerate model development. Yet, there remain open questions about interpretability, generalizability, identifiability, and how such models can be reliably used to advance mechanistic understanding, inspire novel experiments, or inform clinical applications. Further research is needed to refine hybrid model architectures, develop interpretability tools, and establish benchmarks for assessing predictive performance, mechanistic validity, and translational relevance (including robustness and intervention accuracy).This Research Topic aims to showcase pioneering work that advances hybrid AI-mechanistic modeling in systems neuroscience. We seek contributions that explore methodological innovations, address the limitations of current models, and propose strategies for integrating predictive accuracy with mechanistic insight. Submissions can delineate conceptual frameworks, present case studies, or provide tools and benchmarks that enhance the interpretability and utility of hybrid models. By emphasizing approaches that go beyond mere prediction, we hope to advance the dialogue between data-driven and mechanistic neuroscience.To gather further insights in this rapidly evolving area, we welcome articles that focus on innovative integrations of AI and mechanistic modeling, as well as critical evaluations of their implications and limitations. Relevant themes include, but are not limited to:o Architectures and frameworks for AI-mechanistic hybrid modelso Methods for interpreting and validating hybrid models in neuroscienceo Applications of hybrid models to neural coding, circuits, disease progression, and pharmacological interventionso Tools for transparency, explainability, uncertainty quantification, and hypothesis generationo Comparative analyses of predictive and mechanistic performanceo Hybrid AI–mechanistic and AI–quantitative systems pharmacology case studieso Case studies uniting machine learning with neuroscientific theoryo Ethical and philosophical considerations in hybrid modelingWe welcome diverse article types, including original research, reviews, methods, opinion pieces, and perspective articles. Topic editors: Christoph Metzner, Federico Reali Source: https://www.frontiersin.org/research-topics/78738/ai-mechanistic-neuroscience-hybrid-models-beyond-prediction