State of Brain Emulation Report 2025 (2510.15745v1)

Abstract: The State of Brain Emulation Report 2025 provides a comprehensive overview of recent achievements in brain emulation. By analyzing current trends and the state of the art, this report aims to identify key opportunities and challenges facing the field.

• The paper provides a rigorous operationalization of simulation versus emulation, establishing essential benchmarks for evaluating whole brain models.
• It details significant advancements in neural data acquisition, connectomics, and computational modeling across four model organisms with quantitative insights.
• The report emphasizes practical implications for biomedical research, AI alignment, and the scaling challenges requiring exascale computing.

State of Brain Emulation: Technical Assessment and Research Trajectory

Introduction

The "State of Brain Emulation Report 2025" provides a comprehensive, empirically grounded synthesis of the current status, technical challenges, and research frontiers in whole brain emulation (WBE). The report systematically reviews advances in neural dynamics, connectomics, and computational neuroscience across four model organisms—C. elegans, larval zebrafish, Drosophila, and mouse—culminating in an analysis of prospects for human brain emulation. The document is notable for its rigorous distinction between simulation and emulation, its quantitative treatment of data and resource scaling, and its focus on the empirical bottlenecks that delimit progress in the field.

Conceptual Framework: Simulation vs. Emulation

A central contribution of the report is the operationalization of the simulation/emulation distinction. A simulation is defined as a model that reproduces observed outputs without necessarily recapitulating the internal causal dynamics of the biological system. In contrast, an emulation is a model that matches outputs by instantiating the same internal causal mechanisms at a specified level of biophysical detail. The report introduces the notion of "minimal brain emulation," specifying baseline criteria (e.g., accurate connectome, cell type diversity, plasticity, neuromodulation, temporal resolution) that must be met for a model to be considered a candidate emulation. This formalism is critical for establishing objective benchmarks and guiding empirical validation.

Advances in Data Acquisition and Connectomics

Neural Dynamics

The field has witnessed exponential improvements in neural recording capabilities, with the number of simultaneously recorded neurons increasing by three orders of magnitude since the 1980s. However, no organism has yet achieved whole-brain, single-neuron resolution recordings (>90% of neurons) with sufficient temporal fidelity and behavioral coverage. For example, C. elegans recordings reach ~50% of neurons, Drosophila up to 68%, and larval zebrafish ~80%, but all with significant trade-offs in temporal resolution, duration, and behavioral repertoire. In mammals, the best current approaches (e.g., light beads microscopy, Neuropixels) can record up to 1 million neurons in mice, still only ~1% of the brain.

Connectomics

Connectome reconstruction at synaptic resolution has become increasingly tractable for small brains. The cost per reconstructed neuron has dropped from ~$16,500 (1980s, *C. elegans*) to ~$100 (zebrafish larvae, 2025). For Drosophila, the entire CNS connectome is now available, with proofreading times reduced to ~19 minutes per neuron. In mammals, the MICrONS project has reconstructed 1 mm³ of mouse cortex, but scaling to whole-brain connectomics remains limited by manual proofreading costs and data storage requirements (petabyte to exabyte scale). The report highlights the potential of expansion microscopy (ExM) and barcoding to provide molecularly annotated connectomes, which would address the current lack of synaptic-level molecular information in EM-based reconstructions.

Computational Neuroscience: Modeling and Simulation

Model Complexity and Data Constraints

The report provides a detailed taxonomy of neuron and synapse models, from simple leaky integrate-and-fire (LIF) units to multi-compartment Hodgkin-Huxley models and molecular dynamics. It emphasizes that model complexity must be matched to available data: more detailed models require exponentially more parameters, which are not currently constrained by experimental measurements, especially in large brains. For small organisms, data-driven approaches (e.g., ANN-based models for C. elegans) are feasible, but for mammals, biophysical priors and generative models linking structure to function are essential.

Embodiment and Behavioral Validation

A key insight is the necessity of embodiment—linking brain models to virtual or physical bodies—to enable closed-loop validation against behavioral benchmarks. The report discusses the development of neuromechanical simulators (e.g., NeuroMechFly, OpenSim, MuJoCo) and the need for standardized, multi-modal benchmarks that integrate neural activity prediction, behavioral indistinguishability, and causal perturbation experiments.

Hardware and Scaling

Modern hardware (e.g., NVIDIA H100 GPUs) can simulate up to 1 million neurons, with memory as the primary bottleneck. Whole-brain emulation for C. elegans and Drosophila is feasible on a single GPU; mouse brain emulation requires small GPU clusters; human brain emulation would demand exascale clusters comparable to those used for frontier AI training. The report provides detailed scaling laws for compute, memory, and interconnect requirements, noting that the "memory wall" and bandwidth limitations are now the dominant constraints for large-scale simulations.

Model Organism Analyses

C. elegans

• Pros: Complete connectome, stereotyped anatomy, low cost, computationally tractable.
• Cons: Limited behavioral repertoire, graded potentials dominate, limited generalizability to vertebrates.
• Gaps: Whole-brain voltage imaging, integration of neuropeptide data, standardized multi-modal datasets.

Larval Zebrafish

• Pros: Vertebrate circuitry, optical transparency, diverse behaviors, imminent whole-brain connectome.
• Cons: Developmental variability, limited behavioral complexity at larval stage, lack of standardized pipelines.
• Gaps: High-speed, whole-brain voltage imaging; comprehensive perturbation atlases; molecular annotation.

Drosophila

• Pros: Rich adult behaviors, mature genetic toolkit, complete CNS connectome, manageable scale.
• Cons: Optical challenges due to tracheae, evolutionary distance from mammals.
• Gaps: Whole-brain, single-neuron resolution imaging; molecularly annotated connectomes; integrated neuromechanical models.

Mouse

• Pros: Mammalian brain, extensive data, biomedical relevance, strong research infrastructure.
• Cons: Scale of data acquisition, lack of whole-brain functional data, high cost.
• Gaps: Automated connectome proofreading, structure-to-function translation, standardized behavioral benchmarks.

Human

• Pros: Ultimate target for neuroscience and AI, initial connectomic and modeling milestones achieved.
• Cons: Scale (86B neurons), ethical and technical barriers to data acquisition, lack of high-resolution functional data.
• Gaps: Scalable connectomics, structure-to-function inference, exascale computing, model validation.

Methodological Innovations

The report details the state-of-the-art in neural recording (calcium/voltage imaging, fMRI, EEG, MEA, ultrasound), connectomics (EM, ExM, X-ray, barcoding), and computational modeling. It emphasizes the need for multi-modal, standardized, and open-access datasets, and highlights the critical role of AI in automating segmentation, proofreading, and parameter inference. The integration of molecular, structural, and functional data is identified as a key enabler for next-generation emulations.

Evaluation and Verification

A major limitation in the field is the lack of standardized, multi-dimensional benchmarks for model evaluation. The report advocates for rigorous, portfolio-based evaluation frameworks that combine neural activity prediction (e.g., ZAPBench), embodied behavioral tests (e.g., embodied Turing test), and causal perturbation experiments. The development of such benchmarks is essential for objective measurement of progress and for aligning research efforts.

Implications and Future Directions

Practical Implications

• Biomedical Research: High-fidelity emulations could transform drug discovery, disease modeling, and personalized medicine by enabling in silico experimentation at unprecedented scale and detail.
• AI Alignment: Brain emulation offers a path to AI systems with human-like structure and potentially more interpretable or alignable behavior.
• Neurotechnology: Advances in recording, imaging, and simulation will drive innovation in brain-computer interfaces and neuromorphic hardware.

Theoretical Implications

• Structure-Function Mapping: The transition from data-rich small organisms to data-scarce large brains will test the sufficiency of structural and molecular priors for inferring function.
• Limits of Emulation: The field must empirically determine the minimal necessary level of detail for faithful emulation, balancing computational tractability with biological realism.
• Embodiment and Identity: The necessity and fidelity of embodiment for preserving identity and function in emulations remains an open empirical and philosophical question.

Future Developments

• Automation: Continued progress in AI-driven segmentation and proofreading is likely to reduce the cost and time required for large-scale connectomics by orders of magnitude.
• Molecular Annotation: Integration of ExM and barcoding will enable molecularly annotated connectomes, critical for accurate functional inference.
• Scaling: Exascale computing and advances in memory/interconnect architectures will be required for mammalian and human brain emulation.
• Organizational Models: The report suggests that focused research organizations and specialized startups may be better suited than traditional academic labs for the scale and integration required in WBE.

Conclusion

The "State of Brain Emulation Report 2025" provides a rigorous, quantitative, and empirically grounded assessment of the field. It identifies clear technical bottlenecks—data acquisition, molecular annotation, model validation, and computational scaling—and outlines tractable research opportunities, especially in small model organisms. The report's emphasis on objective benchmarks, open data, and interdisciplinary integration sets a clear agenda for the next phase of WBE research. The trajectory of the field will be determined by the interplay of advances in experimental neurobiology, computational modeling, and large-scale engineering, with implications that extend across neuroscience, artificial intelligence, and the philosophy of mind.