Video explaining the Brain Emulation Challenge
The Brain Emulation Challenge aims to accelerate neural circuit reconstruction and brain emulation by creating standardized reference brains with fully understood circuit structure and functional representations. With the advent of high-throughput electron microscopy (EM), expansion microscopy (ExM), Calcium and voltage imaging, and co-registered combinations of these techniques, we can now acquire high-resolution data sets that span multiple brain regions or entire small animal brains.
Current neuroscience research relies on correlational studies, which are indirect methods for studying a complex system. Large-scale, high-resolution reconstruction of brain circuitry is needed to achieve mechanistic explanations and predictions of cognitive function. However, there is a lack of published and validated attempts to reconstruct neuronal circuit function that successfully reproduces the full range of information processing.
Video demonstrating Autoassociative Growing
We propose creating a multi-tiered training data set generated from carefully crafted in-silico models of virtual "ground-truth" brain tissue. This approach has been successful in artificial intelligence, where standardized data sets and challenges have driven algorithmic improvements.
Our challenge series consists of successively more sophisticated data sets, each with associated abstractions and simplifications. By gradually increasing the complexity of the data sets, we can aid testing and improvement of analysis and translation methods. For more information on the challenge levels, please visit our website at https://braingenix.org/Challenge/Overview/.
Why take the brain emulation challenge? Why take a challenge that is providing virtual brain data from generated neural tissue?
If your system identification and reconstruction method successfully discovers the neural circuit and translates its meaningful cognitive function, which was hidden in the data your method analyzed, and about which we know everything, for which we can verify and validate exactly how well the reconstructed result performs a specific function, then we have much stronger reason to believe claims about reconstructions and discovered function from unknown biological neural tissue.
It is a way to test qualitatively and quantitatively if a proposed method can indeed discover and extract what it is meant to find, establishing trust that it is able to deliver a specific and correct working model based on collected brain data.
To participate in the challenge, please review the challenge levels and associated data sets on our website. Implement your analysis and translation methods on the provided data sets and submit your results.
This project is built on the BrainGenix platform for WBE research. We specifically use the following components from BrainGenix:
This project is licensed under the AGPLV3 license.
For questions or feedback, please contact us at [email protected].