This infrastructure has been made available in partnership with EPFLís Laboratory for Neural Micro Circuitry. Accomplishment of Blue Brain Project hinges on very high volumes of standardised and high-quality experimental data that encompasses all likely levels of brain organisation. Data is sourced from the literature (via the projectís automatic information extraction tools), from Human Brain Project, from large data-acquisition initiatives outside the project, and from EPFLís Laboratory for Neural Micro Circuitry.
Blue Brain workflow generates a huge need for computational power that falls in high-performance computing arena. In Blue Brain cellular-level models, depiction of detailed electrophysiology and communication of one neuron is estimated to require as high as 20,000 differential equations. While there are modern multi-core workstations, it is very challenging to solve such a high number of equations in biological real time.
Blue Brain Projectís simulation of the neocortical column incorporates detailed representations of a minimum of 30,000 neurons. Generally, in order to facilitate valid boundary conditions, N times this number is required.
In early stages, IBM BlueGene/L supercomputer running on 8192 processors was being used. BlueGene/L system is a totally new approach in supercomputer design optimised for bandwidth, scalability and the ability to handle large amounts of data while consuming a fraction of power and floor space required by some of the leading supercomputing systems.
The system needs the floor space of about four large fridges, and has a peak processing speed of a minimum of 22.8 trillion floating-point operations per second (22.8 teraflops).
This means that the supercomputer can theoretically carry out 22.8 trillion calculations per second. By mapping one or two simulated brain neurons to each processor, the computer becomes a silicon replica of 10,000 neurons communicating back and forth.
BlueGene/L System was designed to simulate high-speed atomic interactions, which also provide the optimal architecture to simulating neural interactions. Simulations optimsed for clusters using MPI messaging can easily be ported to run on Blue Gene. BlueGene/L Prototype System allows parallel processing of virtually any number of processors to meet the memory and speed demands of a simulation. It can be scaled up enormously to meet further computational demand, and has provided the foundation for further development on BlueGene/P, the next-generation IBM supercomputer that constitutes a quantum leap in memory capacity, processing speed and whole-brain simulations.
Till recently, the project was powered by a 16,384-core IBM BlueGene/P supercomputer that had a memory about eight times more than the memory of IBM BlueGene/L. This makes IBM BlueGene/P capable of touching pet flops speeds and quadrillions of calculations per second.
Presently, IBM BlueGene/Q supercomputer with 65,536 cores and extended memory capabilities hosted by Swiss National Supercomputing Center in Lugano has been deployed.
The Blue Brain Software
The human neocortex region has millions of microcircuits called neocortical columns and, hence, it is important to create a molecular-level modelling of a neocortical column using sophisticated software. This software version is transformed into a hardware versionóa molecular-level neocortical column chipówhich can then be duplicated. While there are many software for simple/point neurons, there are no optimised software programs that can carry out very large-scale, that is, tens of thousands, simulations of morphologically-complex neurons. The software for such simulations consists of a hybrid between two powerful software approaches: one for large-scale neural network simulations called Neocortical Simulator and the other is a well-established program called NEURON.
Microcircuit databases are critical as Laboratory of Neural Micro Circuitry has attained a huge quantum of data on the composition and connectivity of the neocortical column. Microcircuit data from many other labs all over the globe will also be included. The new database, NEOBASE, is constructed on ROOT platform and modelled on the lines of CERNís innovative work that has facilitated thousands of researchers to work as one team on the same project.
The ultimate objectives are to enable full-scale researcher interactions through NEOBASE for further construction of microcircuit database, partnered visualisations and planned simulations. Microcircuit visualisation is possible since BlueBuilder has been built to design, upload and connect thousands of model neurons. BlueBuilder uploads Neurolucida files with complete morphological data.
In parallel, the model neurons need to be modelled in NEURON to include the physiological properties. The export from BlueBuilder references NEURON files for BlueColumn simulations, while BlueBuilder pulls out neurons directly from NEOBASE. Connections are formed according to established connectivity rules. Phenomenological and biophysical models of synapses are assigned to the connections within BlueBuilder. It generates two types of output files: first for simulations by NCS or Neurodamus on Blue Gene and second for visualisation in BlueVision.
Visualisation consists of diverse graphic formats ranging from neurolucida reconstructions, lines, triangles, particles to NURBS. Ultra-high-resolution graphics are designed by deploying interactive walk inside and navigation technologies. 2D, 3D and 3D immersive visualisation systems are also likely to be used.
BlueImage is a software module connected to BlueVision that permits in silico imaging of the activity produced in the neocortical column. All values that emerge as an output of the simulation can be imaged. Microcircuit analysis is possible as Blue Gene simulations can generate terabytes of data in a span of minutes of simulations.
BlueAnalysis is used to graphically display, analyse, discard and archive data at a speed as high as feasible. Microcircuit simulations are feasible since the neocortical microcircuit is simulated via NeoCortical Simulator, which is capable of large-scale simulations.
NeoCortical Simulator is optimised for parallel simulations with MPI messaging and permits easy expansion of the complexity of the simulation.
NeoCortical Simulator is deployed for all large-scale simulations using the least neuronal models. As many as ten compartmental models can be used in NeoCortical Simulator.
Detailed neuron simulations are conducted via NEURON developed and implemented for simulations on Blue Gene. A merged NeoCortical Simulator-NEURON simulator called Neurodamus carries out large-scale NEURON simulations. Neurodamus is likely to evolve further for optimal large-scale Blue Gene simulations of multi-compartmental neuron models.
BlueStim is a software interface to Neurodamus that enables mapping of external input into any one of the 100 million synapses in the column. Stimulus Generator permits connecting of columns with the external world or with other columns and other brain regions.
BlueRead, a software interface to NeoCortical Simulator and Neurodamus, enables us to define values that are to be read out of the simulator for visualisation, display and analysis.
What the future holds
The brain carries out several analogue operations that computers are not capable of performing and, in many cases, it manages to carry out hybrid digital-analogue computing. The biggest differentiating factor between the brain and computers is that the brain is constantly changing with time. Imagine, if components such as integrated circuits and transistors started changing, then a computer would actually end up becoming unusable.
You can imagine the brain as a dynamically-morphing computer, considerably different from other organs like the heart or lungs.
Understanding the brain is vital, not just to understand the biological mechanisms that give us our thoughts and emotions, and that make us human, but for practical reasons. Understanding how it processes information can make a fundamental contribution to the development of new computing technologiesóneurorobotics and neuromorphic computing. More important still, understanding the brain is essential to understand, diagnose and treat brain diseases that are imposing a rapidly-increasing burden on the worldís aging population.
The present configuration of Blue Gene, that is, BlueGene/Q, is capable of touching pet flops speeds and quadrillions of calculations per second. This supercomputer has as many as 65,536 cores and extended memory capabilities. The next generation of Blue Gene supercomputers is expected to be able to deliver an even higher level of computing power and, hence, be able to simulate even more neurons with significant complexity.
However, sky is the limit, as today scientists are thinking beyond all this and conducting challenging research studies. They are hoping to create an artificial brain that can think, respond, take decisions and store any type of data in its memory. Key objective is to upload a human brain into a machine, which will enable man to think and take decisions effortlessly.
After the death of the human body, the virtual brain will actually be able to act as the deceased. Hence, there will be no loss of knowledge, intelligence, personalities, feelings and memories of humans. In a way, man is on his way to becoming immortal.
Deepak Halan is associate professor at School of Management Sciences, Apeejay Stya University