Speaker: Gašper Tkačik
Subject: Information processing in neural and gene regulatory networks
Location: TU Delft
Author: Renée van der Winden
Gašper Tkačik came to talk to us about his research on information processing in biological networks. His main goal is to predict what biological networks do from first principles and to quantify their function in this sense. He first gave us a brief introduction into Shannon’s information theory, which quantifies and optimizes information transmission. He also posed the question: ‘How can we recover the input at the end of a process?’. To illustrate his points, Tkačik explained two examples to us.
The first example was about the retina and how it encodes information. Through measuring the information flow into and out of the retina, it was predicted what modification the neurons make on the incoming light. Namely, that they perform center-surround filtering. After this prediction was made, it was confirmed by measurements. So in this case they succeeded in predicting the function of a network from first principles. Continuing with the retina, a different experiment was performed in which the pattern of neurons firing when a movie was shown was examined. Through measurements, the scientists found a probability distribution for these patterns. Looking at this distribution they found out that the neural output actually is not decorrelated, as was previously thought. In fact, each pair of neurons is weakly correlated. Moreover, they succeeded in decoding what movie had been shown by looking at the output information provided by the retina.
Figure 1: A shortened overview of how the movie was decoded from the retinal code
The second experiment was about how morphogen gradients convey information in early development. The question that was posed was: ‘How much information is stored in the pattern?’. It turned out that the answer is approximately 2 bits per gene. However, four genes store 4.3 bits of information. By finding these numbers, Tkačik formalized an established concept of positional information.
The idea of quantifying what happens in biological networks is very interesting to me. I am interested in how organisms work, but I also really like the certainty that mathematics and physics give you. This is a way to combine the two. The talk was relatively easy to understand, which is always nice. It was also the first seminar in which I recognized concepts that I have learned during my own courses.