How advanced is computational power when it comes to simulating biological systems? This year, with the planned release of “transcendence” (an absolutely awesome looking film based on the technological singularity) and the first ever Nobel Prize in chemistry awarded for the “development of multiscale models for complex chemical systems” in 2013 such as the interactions between pharmaceuticals and proteins, I was curious to see just how much computational power will influence not only scientific advancements in the future but how it is being employed right now.
So, everyone always tells you that your brain is the most complex computational machine ever conceived, and understanding it is a task akin to fathoming the formation of the universe we live in, but can you solve this?
Well you’ll have to forgive me for making my point so early (I know I certainly can’t) and yet the truth is this: your laptop, phone and even car could calculate it in less than a tenth of a second. So what differentiates the unfathomable intelligence of humans from the equally abysmal equation? Well it turns out that while capacity for multiple mathematical computations is limited in the brain, what makes it special is the modality by which it achieves it, and it is because of this that our brains and computers are so powerful yet so radically different.
There are several key differences between current computers and human brains:
Perhaps the biggest difference is that computers are digital. While the binary 1’s and 0’s of computers bear similarity to the “all-or-nothing” nature of electrical signals within our brains, this is where the similarity stops. It is now recognised that the rate at which neurons fire is more important than their strength. By being able to detect the frequency of firing neurons are able to fire in synchrony or disarray. This leads to the strengthening and weakening of neurons in response to stimulation and is believed to underpin memory, learning and many higher cognitive functions that elude the static nature of computers and allow brains to remain dynamic.
Once these memories and facts are stored in the brain we also differ in how we recall this information. Opening a file on a computer is based on recalling the exact information stored at a specific site within hard-drive, called “byte addressable memory.” Unlike this, human memory is essentially stored within other memories that are recalled based on similarity, and as such a few keywords can result in full recall. Rather than remembering the exact information, much like a massively parallel google, but based on your own experiences. Pretty cool huh?
Consider then, an artificial brain generated from computational power, where the fixed modular nature of computer circuitry is overcome by simulating the activity of neurons. This is a tall order considering that currently a single laptop is required to simulate a single neuron and our brain has around 1000 billion of these. Nevertheless this is currently a reality. The Blue Brain Project, consisting of over 10 European universities, is a project that aims to build a full simulation of a human brain. Not only will this form of computational neuroscience allow researchers to simulate and predict cognitive functions in ways previously to unethical or difficult to achieve, but also it will enable us to exploit the unique nature of the brain for advancements in computer science. Imagine a supercomputer within which sits a simulated brain, capable of massively parallel calculation without the need for huge amounts of stored data. This comes without the limitations of relatively slow or limited neurotransmission in the brain compared to the rapid electrical signalling of a computer. The prospect is both fascinating and terrifying. This is considered by many to be the technological singularity, and is by no means a new idea; in fact it was first proposed by mathematician John Von Neuman in 1958.
But why stop there? Simulating the output of a cell (such as an electrical transmission) is as the brain project shows, possible on a large scale. However the simulation of cellular growth involves many complex and multi-layered inputs. The first ever attempt at this was a cellular automaton based on very basic principles, the inspiration for this entire post. A simulation called “Conway’s game of life” invented by mathematician John Horton Conway in the 1970’s, is a game that allows “cells” or squares on an infinite chessboard to be “alive” or dead” based on the number of cells alive around them. While biologists will likely scoff at such a simple representation of a biological system what is fascinating is the complexity of emergent processes that manifest within a system based on only two rules. Interested? Check it out here: https://www.youtube.com/watch?v=XcuBvj0pw-E
Nerdy-ness quenched? Good, because more recent biological simulations of cells internal environments have started to inform our understanding on complex signalling pathways as they occur within actual cells, so that rather than generate models based on what we know the models are themselves informing our understanding of cellular communication. Researchers at MIT attempting to generate a model of the cellular signalling pathways that interact to control cell death, such as growth factors, cytokines and inflammation, DNA damage and apoptosis, found that while the model was functional, better analysis could be carried out when they applied increasingly extreme cellular prompts until the model broke and failed to predict cellular outcome. This system, called “model break-point analysis” highlighted that the dynamic range of signal intensity that a cell receives is a greater outcome of cellular fate that the intensity alone. This form of analysis is a step toward modelling complex intracellular signalling pathways, as previous models of cell signalling assume only basal or maximal activation of signal relays. By quantifying the degree of inaccuracy in the model after reaching the breakpoint, the researchers could determine the molecular events leading to the models failure just before it happened. Remarkably in so doing, new protein interactions that respond to dynamic range to facilitate pro-apoptotic interactions where found for several key proteins including Akt, ERK and MK2. This type of analysis was validated by experiments in live cells, and therefore represents a powerful tool for finding new targets for drugs in diseases like cancer.
I wonder if simulations such as the matrix would have need to be so complex? If not I guess the existence of science to study this complexity in some way proves we’re not acting as batteries for evil robots. On the other hand one project, called “OpenWorm” has set out to do just this, by generating an entirely simulated nematode worm C. Elegans. This worm contains only 959 cells, and the type of every individual one is known, making it the best start-point for simulating and entire organism. Of these, 302 are neurons, giving us a complete picture of its nervous system, and 92 as muscle cells. The system will model electrical activity within the nervous system and as such attempt to model soft body physics required for movement. If by any chance you’re imagining having your own entirely simulated desktop nematode worm, dream no more – the future is here: http://browser.openworm.org/
That damn equation again:
So now it’s probably time to deal with that equation. The computer scientists and statisticians among you will recognise this as a Hidden Markov Model algorithm or HMM. These statistical models are used for describing the evolution of observable events or “symbols” based on unobservable hidden factors called “states”. These hidden states form a Markov chain that controls the probability distribution of the observed symbol. While this might sound completely useless to a blog based largely on biology, these statistical models have been used extensively for both genome sequencing and determination of protein structure. For instance, many proteins have distinct functional domains, who’s interaction and function is controlled by statistical properties. Using HMMs we can predict the domains and their boundaries within the amino acid sequence based on a series of observed symbols located in other protein structures and the probability of the states that control these by analysing the amino acid properties, pretty neat.
I would like the to finish by adding that while the technological singularity is by no means close (despite what you might want to think), it might be time to get more involved in computer science. I myself have decided to learn java despite significant discouragement from fellows on computer science courses.
Janes KA, Reinhardt HC, & Yaffe MB (2008). Cytokine-induced signaling networks prioritize dynamic range over signal strength. Cell, 135 (2), 343-54 PMID: 18957207
Yoon, B. (2009). Hidden Markov Models and their Applications in Biological Sequence Analysis Current Genomics, 10 (6), 402-415 DOI: 10.2174/138920209789177575