5 Questions to… Hannu Toivonen
(October 26th, 2012) The power of bioinformatics can never be valued enough in biological sciences. It can help to uncover previously hidden connections between genes and proteins or give data a very personal, musical note.
Hannu Toivonen is a computer scientist from the University of Helsinki in Finland. One of his main goals is to use large networks such as biological or social networks to identify useful and non-obvious information. He was, for example, involved with the creation of Biomine, a repository that allows predictions of unknown relationships between certain genes or proteins (BMC Bioinformatics, 13:119). But Hannu is also interested in computer creativity. For this, he uses his data mining approach to identify information networks that could be applied in creative tasks. And some interesting results in this area have already been produced, namely a comic art book with computer-generated poems: Heikki Paakkanen: Tee se kotona. Kiitos. (Do it at home. Thank you) Only available in Finnish.
Lab Times: One of your research interests is to extract non-obvious information from large networks, can you explain a bit more how this works?
For instance, assume a life scientist is curious to know how the Vitamin D receptor (VDR) might relate to Alzheimer's disease. A query to our Biomine service extracts, from a wealth of biological knowledge, a small network that illustrates connections between them. The result shows, for instance, that VDR interacts with cellular tumor antigen p53, which in turn interacts with several genes known to affect Alzheimer's (PAXIP1, PLAU, ACE). It also shows that the VDR gene is located in a genomic region associated with Alzheimer's, and that a gene (TNF) related to Alzheimer's participates in the regulation of calcidiol 1-monooxygenase activity, just like VDR. While the system cannot reliably tell if these are actually relevant connections, it is an efficient tool for exploring potentially interesting non-trivial links.
What is “Sleep Musicalization”?
We just launched this novel service, sleepmusicalization.net. It is also inspired by data mining and computational creativity: we made the computer creative to help in mining data. What the service is: it takes your sleep measurements and composes a novel piece of music from them. The ultimate goal of this work is to help users improve their sleep and well-being. The aim of music in this context is to turn the data into a personal experience, perhaps even an emotional one. Traditional methods – tables, charts, even sonification – aim to convey information objectively. In contrast, we aim to give a subjective experience. This hopefully helps users track their sleep, and makes it more fun, too.
How does “Sleep Musicalization” work?
There are two main components in the implementation. The first one concerns sleep analysis. It derives sleep measurements from the signal of a sensitive force sensor under your mattress. The key measurements obtained are sleep stages, heart rate, respiration rate, and sleeper movements. The second component, on which I will focus here, takes these sleep measurements as input and composes a song. In a nutshell, the composition algorithm first performs the following steps to obtain a novel tune. It generates a chord sequence using a Markov chain [a discrete random process], and then the melody and note lengths with two more Markov chains. They have all been constructed to follow some common conventions of western music. Note lengths are related to the sleeper's respiration rate. In the next steps, sleep measurements really come into play. An accompaniment is generated, quite deterministically, to give each sleep stage a different pattern. The tempo of the music is regulated by the heart rate. The volume of the music is controlled by the amount of sleeper movements. As a result, deeper sleep results in calmer music and lighter sleep in more lively. With a bit of practise one can learn to recognize sleep measurements more accurately.
What is so new about this approach?
The main novelty in this work is the proposal that data can potentially be described through the feelings or emotions the user experiences when listening to music composed from the data. The goal is not to convey specific or detailed information, but rather to give users a new way of perceiving their data. Where this approach will have applications we don't know yet. It obviously best fits applications where the data has a temporal dimension, because music has one. Since our feelings are personal and often private, it seems to me that the best applications of data musicalization are also with some personal data.
What are your plans for the future?
Regarding data musicalization, we would like to properly evaluate how users experience data as music, and we need to obtain a better understanding of what we can express through it and how to best do it. We are also interested in new applications. An interesting idea that was just proposed to us is an aid for blind people, a ‘musical radar’ indicating distances and directions to obstacles, perhaps even their sizes and velocities. We have also discussed data musicalization with musicians, and I would love the idea of a live performance where musicians improvise to “musicalized” data.”
Interview: Karl Gruber
Photo: Hannu Toivonen