生命科学学院学术报告
题目: The structure of an olfactory code
讲座人：Guangwei Si，Ph.D.
Research Associate, 
Center for Brain Science and Physics Department, 
Harvard University
时间： 2019年4月3日（星期三）15:00-16:00
地点：金光生命科学大楼411
摘要:
The nervous system uses populations of diverse neurons to encode the environment. Population-level sensory codes are often constrained to low-dimensional manifolds as a consequence of invariant properties across neurons and the neuronal circuit, as well as by regular and statistical structures in environmental stimuli themselves. Characterizing the structure of sensory representations and their origin in neuronal and environmental properties is essential to understanding perceptual mechanisms. Here, I will discuss our recent progress in understanding insect olfaction.
Olfaction is a canonical example of collective computation. Olfaction requires a diversity of cell types that together produce odor perception and behavior. At the sensory level, a relatively small number of neurons encode environments harboring vast numbers of odorant types across broad concentration ranges. Odor identities and odor intensities are entangled in the activity of sensory neurons, each of which might be activated by many different odorant molecules at many different concentrations. We asked whether structure in the olfactory code might facilitate perceptual tasks such as distinguishing different odors, recognizing the same odor across different concentrations, or measuring the size of concentration changes. As our model, we used the particularly accessible olfactory system of the fruit fly larva. First, we discovered that odor identity and intensity are encoded in orthogonal features of the population code. Odor representations form vector-like trajectories in neuronal activity space. The distance along these trajectories correlates with concentration. The direction of trajectories correlates with odor molecular structure. Closer examination of activity curves revealed another surprising invariant pattern. The activity of any sensory neuron scales with odor concentration along a fixed activation curve with only one free variable, namely an odor dependent sensitivity. Moreover, these values of sensitivity also follow a statistical structure, a specific long-tailed distribution. Odorant-receptor sensitivities are also largely correlated with a single geometrical property of the molecular structure. The several quantitative structures that are exhibited by individual and populations of olfactory neurons contribute to a low-dimensional structure in the primary olfactory code. I will discuss how understanding the quantitative structure of this sensory code provides a starting point for understanding the progression of olfactory representations across successive deeper layers of brain circuitry. I will also discuss the implications of animal development and experience in shaping circuit structure and sensory representations.
欢迎各位老师同学积极参加！