Published: December 2025 | Author: Maryam Rahimi Movassagh
The brain cannot be understood at a single scale. Neural activity, cognition, and disease emerge from interactions between genes, cells, circuits, and environments. Yet much of neuroscience has historically examined these layers in isolation. Single-cell and multi-omics technologies are changing this landscape, generating unprecedented biological detail. The remaining challenge lies in interpretation.
Artificial intelligence offers a way forward—but only if models are designed to respect biological structure and remain interpretable. My research explores how interpretable AI systems can integrate single-cell data, multi-omics profiles, and neuroimaging to uncover early biomarkers and mechanistic insights into neurological disorders.
Brain tissue is highly heterogeneous. Even neighboring cells can differ dramatically in gene expression, metabolic state, and functional role. Traditional bulk analyses average these signals, often obscuring early pathological changes.
Single-cell technologies reveal this hidden diversity. They allow us to observe how specific cell populations respond to stress, disease progression, or genetic perturbations. These early cellular deviations may serve as critical biomarkers long before clinical symptoms appear.
Gene expression alone does not tell the whole story. Epigenetic regulation, protein abundance, and metabolic activity all influence how cells function within neural circuits. Multi-omics profiling captures these complementary layers, providing a more complete picture of brain biology.
Integrating these data modalities is challenging. They differ in scale, noise characteristics, and biological meaning. Interpretable AI models can act as unifying frameworks, learning shared structure while preserving modality-specific information.
Neuroimaging and electrophysiology reveal how the brain functions at scale, but often lack molecular specificity. By linking imaging features to single-cell and multi-omics data, researchers can associate functional brain patterns with underlying biological mechanisms.
This integration enables a shift from descriptive biomarkers toward mechanistic understanding. Instead of asking where abnormalities occur, we begin to ask which molecular pathways drive them.
In biomedical research, black-box predictions are rarely sufficient. Clinicians and neuroscientists require explanations that align with biological knowledge. Interpretability ensures that AI models can be trusted, validated, and translated into real-world impact.
By embedding biological constraints and structured representations, interpretable AI can highlight meaningful relationships between genes, cell types, and brain networks. These insights form the foundation for hypothesis generation and experimental validation.
Neurological disorders often develop silently over years. Detecting early molecular changes is critical for intervention and disease management. Single-cell and multi-omics AI models provide a pathway toward identifying these early warning signals.
By integrating biological data across scales, this approach supports a future of precision neuroscience— where diagnosis and treatment are informed by an individual’s unique molecular and neural profile.
From: Maryam Rahimi Movassagh
Ph.D. Student, University of Central Florida (UCF)