Digital Twin Brain: a simulation and assimilation platform for whole human brain

Wenlian Lu,Longbin Zeng,Xin Du,Wenyong Zhang,Shitong Xiang,Huarui Wang,Jiexiang Wang,Mingda Ji,Yubo Hou,Minglong Wang,Yuhao Liu,Zhongyu Chen,Qibao Zheng,Ningsheng Xu,Jianfeng Feng

2023-08-02

Abstract:In this work, we present a computing platform named digital twin brain (DTB) that can simulate spiking neuronal networks of the whole human brain scale and more importantly, a personalized biological brain structure. In comparison to most brain simulations with a homogeneous global structure, we highlight that the sparseness, couplingness and heterogeneity in the sMRI, DTI and PET data of the brain has an essential impact on the efficiency of brain simulation, which is proved from the scaling experiments that the DTB of human brain simulation is communication-intensive and memory-access intensive computing systems rather than computation-intensive. We utilize a number of optimization techniques to balance and integrate the computation loads and communication traffics from the heterogeneous biological structure to the general GPU-based HPC and achieve leading simulation performance for the whole human brain-scaled spiking neuronal networks. On the other hand, the biological structure, equipped with a mesoscopic data assimilation, enables the DTB to investigate brain cognitive function by a reverse-engineering method, which is demonstrated by a digital experiment of visual evaluation on the DTB. Furthermore, we believe that the developing DTB will be a promising powerful platform for a large of research orients including brain-inspiredintelligence, rain disease medicine and brain-machine interface.

Neural and Evolutionary Computing,Neurons and Cognition

What problem does this paper attempt to address?

### Problems the paper attempts to solve

This paper aims to build a computational platform named "Digital Twin Brain" (DTB), which can simulate spiking neural networks on the scale of the full - sized human brain, and further, can simulate personalized biological brain structures. Different from most brain simulations with homogeneous global structures, the authors emphasize the important influence of sparsity, coupling, and heterogeneity in brain sMRI (structural magnetic resonance imaging), DTI (diffusion tensor imaging) and PET (positron - emission tomography) data on the efficiency of brain simulation. Through large - scale experimental verification, the DTB for human brain simulation is a communication - intensive and memory - access - intensive computing system, rather than a computationally - intensive system.

In order to balance and integrate the computational load and communication traffic from heterogeneous biological structures to general - purpose GPU - based high - performance computing (HPC), the authors adopt a variety of optimization techniques to achieve leading - edge simulation performance of spiking neural networks on the scale of the full - sized human brain. In addition, combined with the mesoscale data assimilation method, DTB can study brain cognitive functions through the reverse - engineering method, which has been verified in digital experiments of visual evaluation. The authors believe that the DTB under development will become a powerful platform applicable to multiple research directions including brain - like intelligence, brain - disease medicine, and brain - machine interfaces.

### Key problems and challenges

1. **Biological structures of large - scale data**:
- The adult brain contains about 100 billion neurons and 1000 trillion synapses, of which the cerebral cortex and cerebellum account for 99%.
- There are hundreds of functional regions in the brain, and the types of neurons, synapse types, and the number of connections per neuron vary in different regions.

2. **Simulation speed**:
- The complexity of the brain synaptic connection structure leads to difficulties in high - efficiency simulation. For example, based on sMRI, DTI, and PET data, the sparsity of brain connections is about 7.2 per thousand of full connections, and there is significant coupling.
- The average connection ratios in different brain regions are as follows: cortex 2/7, subcortex 6/25, brainstem 14/25, and cerebellum 16/125.
- The ratio of the maximum number of outgoing connections to the minimum number of outgoing connections is as high as 35, which leads to extreme imbalance in computation and communication traffic.

3. **Brain function**:
- Existing technologies cannot run large - scale brain simulations at real - time or near - real - time speeds, especially for near - full - sized human brain simulations with real biological structures.
- The authors propose a new data assimilation method, and through comparison with the actual biological brain, it is verified that the Pearson correlation coefficient of DTB in the visual evaluation task exceeds 0.65.

### Solutions

1. **Optimization algorithms**:
- A partitioning algorithm has been developed to minimize the communication traffic between different GPUs and improve the simulation efficiency.
- By balancing the communication load between each GPU, network congestion is avoided, thereby improving the latency performance.

2. **Data structures**:
- Generate neuron property tables and connection tables for brain network simulation. Each GPU corresponds to a neuron property table, which stores properties such as the membrane potential equation parameters of neurons.
- The connection table stores the GPU ID, neuron ID, synapse type, and synapse weight of the presynaptic neurons in each GPU.

3. **Data assimilation methods**:
- A Hierarchical Mesoscopic Data Assimilation (HMDA) framework is proposed for estimating model parameters, especially the synaptic conductivity of each neuron.
- By tracking the BOLD signal, the brain - scale neural network is statistically inferred. In the specific implementation, the hyperparameters are first estimated in a smaller - scale model (20 million neurons and 20 billion synapses), and then the parameters of the full - sized human brain model (8.6 billion neurons and 47.8 trillion synapses) are sampled from these hyperparameters.

### Summary

This paper solves the computational and communication bottleneck problems in large - scale human brain simulation by building the DTB platform, and verifies the effectiveness of the model through the data assimilation method. This work provides new tools and methods for understanding brain functions and developing brain - like intelligence technologies.