The Forbidden Library: Mapping the Connectome

Visual Generated by Google’s Gemini

Imagine discovering an ancient library, buried deep underground. The corridors twist endlessly, split and merge in unpredictable ways. Rooms connect through narrow passageways while other appear completely isolate. Every book and staircase leads to somewhere else, yet no one has ever mapped the full structure.

This structure exists in real life, in Connectomics, where neuroscientists aim to map the neural networks of the human brain.

For this project, I designed a small experiment to explore the role of human tracing in connectome reconstruction. With prior about Connectomics, research I started to wonder: Can automated neuron semantation in connectomics be made more accurate through human-guided neuron tracing?

To simulate this idea, I used two platforms used in neuroscience research. The first platform I used was EyeWire, which simulated the reconstruction phase by tracing neurons from raw imaging data. The second platform I used was MICrONS Explorer (Neuroglancer), which simulated the analysis phase by exploring completed neuron reconstructions.

Part I: Neuron Reconstruction

Website Used: https://eyewire.org

EyeWire presents users with small blocks of 3D neural tissue created from stacked electron micoscrop images. Within each block, a computer algorithm has already attempted to identify part of a neuron. The task of the user is to determine which part of the 3D block is part of the given neuron.

During the experiment/simulation, the computer-identified segments turn turquoise.

Selecting the Unidentified Parts of the Neuron

After submitting the trace, the system provides immediate feedback, where the green identifies correct selections, red identifies selections and yellow identifies missing regions.

After Submitting Newly Identified Parts of the Neuron

This feedback allowed me to evaluate how well human tracing matched neuron structures. While performing the tracing tasks, I noticed that neurons branch in irregular ways. In several cases, it was very difficult to select or see the identified parts accurately as the branches crossed or ran very closely to others.

At first, the images looked like irregular shapes, but after taking a little time to trace segments, the branching patterns became easier to recognize.

Part II: Circuit Vizualization

Website Used: MICrONS Explorer, neuroglancer

Neuroglancer can be accessed by pressing the “Explore” button in the MICrONS Explorer menu (circled in red in the image).

After tracing neurons from raw image data, I used MICrONS Explorer to observe how reconstructed neurons appear within a larger circuit of neurons.

The MICrONS project is a large-scale effort to reconstruct neural networks from high-resolution brain imaging data. The platform allows and provides the user to explore reconstructions in three dimensions.

Using Neuroglancer, I could rotate neurons freely, zoom into individual branches, and identify synaptic connections between neurons. Unlike fragmented segments in EyeWire, the neurons here appeared as complete structures with complex branching patterns.

Each neuron was shown in different colours, and I was able to select certain neurons to make only the selected on show. This made it easier to distinguish individual cells and observe how they connect with other neurons.

The Image shows a reconstructed neuron and it’s branches extending in different directions

Seeing the completed reconstructions helped connect small tracing tasks to the larger goal of mapping neural circuits. If EyeWire represented the process of reassembling fragments, MICrONS showed the final structure once those fragments were combined.

Debrief

The simulation revealed several observations about connectome reconstruction.

First, neuron tracing requires careful visual interpretation. Even in a simplified simulation, identifying the correct neuron boundaries required very close attention to branch patterns and spatial continuity.

Second, automated segmentation appears to struggle in areas where neurons overlap or intersect. The regions require human judgment to determine whether the neuron belongs to the same one or a neighbouring ones.

Finally, that small tracing decisions contribute to a larger network. Individual neurons reconstruct to become a part of a dense web of neural connections.

Limitations

Professional laboratories work with massive datasets containing thousands of neurons and petabytes of imagining data. Reconstructing these datasets require advanced computational systems and machine learning pipeline. The simulations used in this article is only a small portion of the workflow. However, they still do illustrate one of the central challenges of the field: transforming fragmented imagining data into accurate maps of neural connections.

Analysis

This simulation helped reveal the complexity of mapping the connectome.

At first, neural tissues appear irregular, but through careful reconstruction and visualization, patterns begin to emerge. The connectome resembles the underground library, it is vast, interconnected and largely unexplored. Each traced neurons add another corridor to the map. Over time, those corridors begin to reveal the structure of the neural systems.

Scientists have not yet mapped the human connectome fully, but through a combination of human insight and computational tools, the once-hidden library is growing clearer and clearer.