Big Neuron – Towards Big Data in Neuroscience

Summary

This year Hanchuan Peng (Allen Institute for Brain Science) began the next phase of the effort called Big Neuron. However, rather than a competition, this time the project would be a collaboration. The key idea is to create a single platform on which all algorithms can be run, compared, and their results combined to form reconstructions better than any one could achieve alone.

This year Hanchuan Peng (Allen Institute for Brain Science) began the next phase of the effort called Big Neuron. However, rather than a competition, this time the project would be a collaboration.

The key idea is to create a single platform on which all algorithms can be run, compared, and their results combined to form reconstructions better than any one could achieve alone.

Several years ago a joint project was started in order to determine just what the state of the art was in automatically generating digital reconstructions of neurons. The effort stands between computer science, specifically machine learning and computer vision, and neuroscience. The neuronal reconstructions are currently used for various morphological analyses and computational modeling, helping scientists better understand the relationships between morphology, electrophysiology, gene and protein networks, and disease states. In the future, it could also be highly informative for understanding the connectivity within and between brain regions which underlies neural computation and ultimately how our minds work. Producing a single reconstruction can take many hours, even days or weeks depending on the size and complexity, which keeps the field from being able to exploit big data approaches and making discoveries it would make possible.

Losavio et al. Live Neuron Morphology Automatically Reconstructed From Multiphoton and Confocal Imaging Data. J Neurophysio, 2008, 100:2422-9.

Losavio et al. Live Neuron Morphology Automatically Reconstructed From Multiphoton and Confocal Imaging Data. J Neurophysio, 2008, 100:2422-9.

 

diadem_logo_splashThe previous project, called the DIADEM ChallengeDIgital and Axonal reconstruction of DEndritic Morphology — was a multi-stage competition run over a year and ending in a conference and final competition phase. No team beat the 20 fold increase in speed over current manual and semi-automated (i.e. computer assisted) reconstruction that was the goal, but it was clear that the community was making great strides and that with the lessons learned and new resources the promise of full automation was close.

This year Hanchuan Peng (Allen Institute for Brain Science) began the next phase of the effort called Big Neuron. However, rather than a competition, this time the project would be a collaboration. The key idea is to create a single platform on which all algorithms can be run, compared, and their results combined to form reconstructions better than any one could achieve alone. The scattered nature of the field has made sharing and comparing algorithms difficult, however with a standard platform the process becomes much easier. Peng’s Vaa3D, started while he was at HHMI’s Janelia Research Campus, is a visualization and analysis tool for anatomical imaging data which can be easily extended with plugins. Several hackathons have already taken place all over the world (China, the UK, and the US) to get scientists together to port their algorithms as plugins for Vaa3D. One hackathon focussed on producing manually reconstructed test data and another focussed on analytical tools.

Creating a Consensus

I joined Hanchuan, his team, and a variety of neuroscientists and computer scientists at the Janelia Research Campus (Ashburn, VA) hackathon in June in order to build a tool for merging reconstructions into a consensus. The algorithm is still in progress and should be completed and tested within the next couple of months. The challenge of producing a consensus comes primarily from the uncertainty of how various branches are connected to one another. The uncertainty may arise out of dark spots due to inconsistent labeling of the neuron, signal noise, low resolution, or high branch density within a region. The resulting errors can be as simple as a missed branch or subtree and as complex as branches connected incorrectly and possibly with their polarity (i.e. which end is towards the soma vs towards terminal tips) flipped. In other cases, two branches might cross each other or bend near each other without crossing, making it very difficult at times for an algorithm to distinguish the cases.

Zhao et al. Automated Reconstruction of Neuronal Morphology Based on Local Geometrical and Global Structural Models. Neuroinformatics, 2011, 9.

Zhao et al. Automated Reconstruction of Neuronal Morphology Based on Local Geometrical and Global Structural Models. Neuroinformatics, 2011, 9.

The core of the algorithm I am working on with Giorgio Ascoli, with generous assistance from several others involved in Big Neuron, focuses on identifying branches first, then allowing the various algorithms’ reconstructions to vote on how the connect to one another. In addition to creating a more reliable reconstruction than any one algorithm is currently capable, it will allow for additional analysis showing which algorithms are best in what conditions which can be fed back in to better weight each algorithm’s contribution to the final consensus reconstruction.

Partners

Mason’s Neural Informatics Center Joins International Brain Research Project

By Molly Brauer from Mason News March 31, 2015

George Mason University researchers have joined a consortium of scientists from Berkeley to Beijing to collect worldwide data on neurons—the brain cells that form the building blocks of the human nervous system.

The Seattle-based Allen Institute for Brain Science spearheads the project, dubbed “BigNeuron.” The effort will allow scientists to examine and share mountains of data, gain insights into brain functions and turn the data into knowledge.

Neurscientist Giorgio Ascoli. Creative Services photo

Neurscientist Giorgio Ascoli. Creative Services photo

Leave a Reply

Skip to toolbar