AI News AI资讯 6h ago Updated 3h ago 更新于 3小时前 41

Second complete map of a fruit fly brain completed 果蝇大脑第二张完整图谱完成

Researchers completed the first full connectome map of a male fruit fly brain, revealing over 300 million synaptic connections across roughly 150,000 neurons The project represents a landmark collaboration between biologists at Janelia Research Campus and computer scientists at Google, each providing essential expertise the other lacked Generative AI models were used to computationally stitch together sliced brain tissue, eliminating seam artifacts from physical sectioning A recurrent visual AI 研究人员完成了雄性果蝇大脑的完整神经元连接组图谱,发现超过3亿个突触连接 这是继雌性果蝇连接组之后的又一里程碑,为神经生物学研究提供关键工具 研究采用电子显微镜结合AI算法的技术路线,由Janelia Research Campus与Google合作完成 该技术有望应用于更复杂的神经系统,包括脊椎动物的脑图谱绘制

55
Hot 热度
65
Quality 质量
55
Impact 影响力

Analysis 深度分析

TL;DR

  • Researchers completed the first full connectome map of a male fruit fly brain, revealing over 300 million synaptic connections across roughly 150,000 neurons
  • The project represents a landmark collaboration between biologists at Janelia Research Campus and computer scientists at Google, each providing essential expertise the other lacked
  • Generative AI models were used to computationally stitch together sliced brain tissue, eliminating seam artifacts from physical sectioning
  • A recurrent visual AI model traces individual neurons through 3D space by converting electron microscopy voxels into spatial neuron outlines
  • This work builds on an earlier female Drosophila connectome and refines methodologies expected to scale toward increasingly complex nervous systems, including vertebrates

Why It Matters

This connectome represents a foundational resource for neurobiology, providing an exhaustive catalog of neural wiring that can accelerate understanding of how brain circuits process sensory input, generate behavior, and form memories. The integration of generative AI and recurrent visual models for large-scale brain mapping demonstrates a powerful new paradigm where AI is not merely辅助 but central to enabling scientific discovery at biological scales previously intractable. For AI practitioners, this showcases real-world applications of generative and spatial reasoning models in solving complex 3D reconstruction problems.

Technical Details

  • Sample preparation: A dissected fruit fly brain (plus ventral nerve cord) was sectioned into evenly spaced slices, preserving 3D architecture while enabling electron microscopy at synaptic resolution
  • Generative AI for seam repair: A generative model computationally fills gaps and distortions at slice boundaries, creating seamless tissue reconstructions that simplify all downstream processing
  • Recurrent 3D neuron tracing: A specialized visual model operates recurrently through spatial voxels, tracing cellular membranes to reconstruct individual neuron morphology across three-dimensional space
  • Synapse detection and classification: Additional models identify and classify synapse types, with tunable sensitivity parameters ("greediness") allowing adjustment of detection thresholds
  • Human-in-the-loop validation: Human proofreaders provide critical feedback to refine model accuracy, particularly in synapse classification
  • Scale: The male fruit fly brain contains approximately 150,000 neurons forming over 300 million synaptic connections

Industry Insight

  • The successful collaboration between biology and AI teams at Janelia and Google validates a hybrid workflow model where domain experts and ML engineers co-develop tools iteratively—organizations should invest in cross-disciplinary teams rather than treating AI as a post-hoc analysis step
  • The methodology demonstrated here—generative repair of physical artifacts combined with recurrent spatial reasoning—establishes a transferable pipeline that could accelerate connectome mapping in other model organisms and inform approaches to larger, more complex brains
  • As AI models for 3D biological reconstruction mature, expect increased investment in neuromorphic computing and brain-inspired architectures, since detailed connectomes provide both the data and the architectural blueprints for next-generation AI systems

TL;DR

  • 研究人员完成了雄性果蝇大脑的完整神经元连接组图谱,发现超过3亿个突触连接
  • 这是继雌性果蝇连接组之后的又一里程碑,为神经生物学研究提供关键工具
  • 研究采用电子显微镜结合AI算法的技术路线,由Janelia Research Campus与Google合作完成
  • 该技术有望应用于更复杂的神经系统,包括脊椎动物的脑图谱绘制

为什么值得看

这项研究展示了AI在神经科学领域的实际应用价值,为理解大脑信息处理机制提供了前所未有的工具。跨学科合作模式的成功验证了AI+科学发现的可行性,对神经科学和人工智能领域均有重要意义。

技术解析

  • 研究团队将果蝇大脑切成大量等间距切片,使用电子显微镜获取高分辨率图像,重建三维神经元结构
  • 使用生成式AI模型修复切片边缘的变形和材料缺失,使组织在计算拼接后看起来无缝
  • 开发了一种递归式视觉模型,将显微镜体素转换为神经元三维表示,沿空间追踪细胞膜边界
  • 其他模型用于识别和分类突触类型,可通过调整灵敏度参数进行微调,并依赖人类校对反馈

行业启示

  • AI在科学发现中的工具价值正在被验证,跨学科合作(生物学+计算机科学)是解决复杂科学问题的有效路径
  • 神经科学领域的突破可能为类脑计算和神经网络架构设计提供新的灵感
  • 该技术路线的可扩展性意味着未来有望绘制更复杂的神经系统图谱,包括脊椎动物大脑

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Research 科学研究