I am attending a meeting of the ESMTB in Lisbon. I am happy that the temperatures are very moderate, much more moderate than those at home in Mainz. The highest temperatures I encountered so far were due to body heat, for the following reason. The first two sessions on reaction networks were in rooms much too small for the number of participants with people sitting on extra chairs brought in from outside, windowsills and the floor, as available. This caused a rise in temperature despite the open windows. In any case, the size of those audiences is a good sign for the field.
The first plenary talk, by Samuel Kou, was about something called Google Flu Trends. The idea of this initiative, which started in 2008, was to predict the development of epidemics such as influenza by using data on the number of Google searches for phrases related to cases of flu. The idea is that between the time someone goes to the doctor because they are suffering from an infectious disease and the time data finally come together on a national level there can be a lag of about two weeks, even in countries like the US with a highly developed reporting system. Thus the hope was that using data on Google searches could enable predictions two weeks earlier than other methods. The first test of this system looked very successful and obtained a lot of positive media coverage. After about a year the predictions started to be very wrong (perhaps due to the swine flu prevalent at that time being atypical). After modifications were made to the method the predictions were even further off and in the opposite direction. Then there was a lot of negative media coverage. Google did not make the raw data available in order to allow experts to look into what had gone wrong. However a paper published in Nature revealed some things. The number of search terms used was optimized and showed a clear local maximum. The search terms were not made public but one striking jump in the distribution was revealed to be a search term related to the Oscar ceremony. It was later suggested that what the analysis was predicting was not the incidence of flu but the beginning of winter. The mathematics used in the analysis was made public and was extremely elementary, although that by itself is not necessarily a bad thing. The group which Kou belonged to was able to make a low-grade reconstruction of the raw data using the freely available Google services called Google Trends and Google Correlate. They could then apply sophisticated statistical techniques to this data, creating a model called ARGO which performed a lot better than Google Flu Trends. Later Google stopped publishing their own analyses and made at least part of their data available to external researchers.
Another plenary talk, by Eörs Szathmáry, was about relationships between learning processes in the brain and evolution. I only understood a limited amount of the content but I nevertheless enjoyed the presentation, which combined a wide variety of different ideas. One specific point I did pick up concerns the Darwin finches. The speaker explained that the molecular (genetic) basis of the different beak shapes of these birds is now understood. This evolution takes place in a three-parameter space and it was indicated how this kind of space can be established. A similar process has been modelled in the case of RNA by Uri Alon. In the talk there was a nice partially implicit joke. One of the speaker’s main collaborators in this work is called Watson and he hinted at the idea of himself as Sherlock Holmes. Apart from the content I found the delivery of the talk very accomplished. I found it good in a way I cannot make precise.
Hydrobates 
Leave a Reply