Event Abstract Back to Event Stochastic Adaptive Sampling in Sensory Information Encoding Zhuoyi Song1, 2, Marten Postma3, Stephen Billings1, Daniel Coca1*, Roger Hardie4* and Mikko Juusola2, 5* 1 University of Sheffield, Department of Automatic Control and Systems Engineering, United Kingdom 2 University of Sheffield, Department of Biomedical Science, United Kingdom 3 University of Amsterdam, Swammerdam Institute for Life Sciences, Netherlands 4 University of Cambridge, Department of Physiology, Development and Neuroscience, United Kingdom 5 Beijing Normal University, State Key Laboratory of Cognitive Neuroscience and Learning, China How are vastly varying light changes in natural environment efficiently represented by the limited dynamic range of photoreceptors? How do photoreceptors effectively encode light intensity changes in millisecond scale? These important questions are at the core of the quest to understand rapid and reliable sensory information encoding mechanisms. We recently uncovered such mysteries by generating biophysically realistic computational models of fly photoreceptors. The models explain how the cell’s ultrastructure and transduction machinery shape response dynamics and coding properties over a broad range of light intenisties [1]. A Drosophila photoreceptor integrates light information by adaptive stochastic sampling rule1. Its photo-sensitive waveguide (rhabdomere) consists of ~30,000 microvilli, each of which is capable of generating single photon responses (quantum bumps). Because a microvillus can only generate a bump at a time and become briefly refractory thereafter [2], a fly photoreceptor is essentially a photon counter, and its information coding capability is ultimately set by the size and number of samples it generates in a time unit. Efficient encoding emerges from adapting the sample sizes (bump waveforms) and numbers to ambient light intensity [3]. The refractory period of each microvillus opposes saturation; together these dynamically and stochastically adjust availability of microvilli (bump production rate: sample rate), whilst intracellular calcium and voltage adapt their bump amplitudes and waveforms (sample size). These adapting sampling principles result in robust encoding of natural light changes, which both approximates contrast constancy and enhances novel/surprising light changes at different light levels. Importantly, rather than treating stochasticity in transduction machinery simply as noise, our modelling framework enabled us to analyse its merits and costs within the signalling processes. The results revealed benefits of stochastic reactions and their compartmentalisation for generating reliably responses. While adaptation adjusts a photoreceptor’s signaling performance to light conditions, stochastic sampling of microvilli contributes to the prevention of saturation and equalizes the usage of available encoding range. Adaptive stochastic sampling accurately predicts information processing across a range of fly species with different visual ecologies, supporting its general role in encoding sensory information. It will be interesting to examine whether this encoding principle is also applicable to other sensory neurons, in which reaction cascades often share similar molecular constituents with compartmentalised sub-structures. This principle could also apply to information sampling/encoding by population of synapses in neurons/networks.