Visualising the information
With the proliferation of the data that underpins human biology and disease, Garvan is embracing new ways of understanding and exploring all this information through state-of-the-art visualisation and animation tools. Stay tuned for more about the Centre for Biodata Visualisation in the next issue of breakthrough .
enough, millions of individual genome sequences with accompanying clinical records, supplemented by data from the Internet of Things [the huge range of connected electronic devices and sensors]. The more data we have, the more we can make correlations between our individual genetic idiosyncrasies and our health. That’s why we’re building capacity in analysis and artificial intelligence to prepare us for this new generation of medical research and medicine. Our job is to lead the way. Our job is to invent the future.” The more data we have, the more we can make correlations between our individual genetic idiosyncrasies and our health Mr Tansel Ersavas, Principal Software Engineer, Deep Learning Initiative at the Kinghorn Centre for Clinical Genomics, is an artificial intelligence (AI) expert tasked with realising this heady ambition. “The data is just exploding,” he says. “Be it in sequencing, single cell genomics or another data type from outside Garvan, traditional statistical methods are insufficient to make sense of it adequately. The AI approach of deep learning is a good candidate because we now have the computing power, we have the quantity of data and we have people who can actually make it happen.” Deep learning is based on a mathematical model of a neuron, which, when multiplied many times over, becomes a simplified type of brain. Its strength lies in its capacity to modify algorithms based on machine feedback, without the intervention of a human teacher or programmer, and thus become self-learning. This technology lies behind some of the most advanced feats of computing that mimic – even exceed – the human experience of the world, such as speech- to-text technology, instant voice translation and the image recognition software in self-driving cars.
The deep learning approach has much promise in interrogating the genome, which is, as Ersavas says, “an imperfect art at the moment”, with only about two per cent reasonably well understood. Dr Kaplan adds that there is greater potential still: “Where we think deep learning is going to become particularly valuable is that, in addition to the genome, there are many other ‘omes’ that could be computed, such as the epigenome, the transcriptome, the metabolome, the proteome, the exposome, and then on top of all of that are personal devices such as Fitbits, your social media presence and so on. How do we take all of this highly variable data and find meaning from it? This is where opportunities around artificial intelligence will become very interesting and useful.” Currently Ersavas is working towards a process to interpret mitochondrial DNA as a pilot study for ultimate application to the whole human genome. This project is being undertaken in parallel with the Garvan-Deakin Program in Advanced Genomic Investigation, a collaboration with Deakin University’s Centre for Pattern Recognition and Data Analytics (PRaDA) established in March 2017. “This partnership aims to harness two great strengths: Garvan’s excellence in genomics and human biology, and PRaDA’s ability to search for patterns in complex data sets, with the mutual ambition of better understanding human biology, diversity and disease,” says Professor Mattick. With the ongoing development of analytical methods, bigger data means better data, and Garvan’s thousands-strong library of genomes is recast as relatively modest in scale. As the volume grows, however, so do the avenues for transforming information into insight, and healthcare in turn.
Tansel Ersavas
Warren Kaplan
John Mattick
February 2018 | 7
Made with FlippingBook HTML5