2005 Special issue Quantifying information and performance for flash detection in the blowfly photoreceptor * Peng Xu, Pamela Abshire * Department of Electrical and Computer Engineering and the Institute for Systems Research, University of Maryland, College Park, MD 20742, USA Abstract Performance on specific tasks in an organism’s everyday activities is essential to survival. In this paper, we extend information-theoretic investigation of neural systems to task specific information using a detailed biophysical model of the blowfly photoreceptor. We determine the optimal detection performance using ideal observer analysis and find that detection threshold increases with background light according to a power function. We show how Fisher information is related to the detection performance and compare Fisher information and mutual information in this task-specific context. Our detailed model of the blowfly photoreceptor enables us to detangle the components of phototransduction and analyze the sensitivity of detection performance with respect to biophysical parameters. The biophysical model of the blowfly photoreceptor provides a rich framework for investigation of neural systems. q 2005 Elsevier Ltd. All rights reserved. Keywords: Blowfly photoreceptor; Biophysical model; Ideal observer analysis; Flash detection; Fisher information; Sensitivity analysis 1. Introduction Biological sensory organs operate under severe con- straints of size, weight, structural composition, and energy resources. In many cases, the performance levels are near fundamental physical limits (Bialek, 1987). Nowhere is evolutionary pressure on information processing stronger than in visual systems, where speed and sensitivity can mean the difference between life and death. Consider fly photoreceptors, capable of responding to single photons, while successfully adapting to light up to w10 6 effectively absorbed photons per second (Hardie & Raghu, 2001). Relying on their visual input, flies can chase mates at turning velocities of more than 3000 s K1 with delay time of less than 30 ms (Borst & Haag, 2002). The marvellous efficiency and effectiveness of neural systems motivate both scientific research to elucidate the underlying principles of biological information processing and engineering efforts to synthesize microsystems that abstract their organization from biology (Abshire & Andreou, 2001a). It is crucial to quantify information processing in neural systems for both purposes. Developed in the 1940s (Shannon, 1948), information theory is the study of information transmission in communication systems. It has been successful in estimating the maximal information transmission rate of communication channels, information channel capacity, and in designing codes that take advantage of it. The usefulness of information theory in neural information processing was recognized early (Barlow, 1961; Atick, 1992; Steveninck & Laughlin, 1996). Information transmission rate has been measured in many neural systems (Borst & Theunissen, 1999), and information channel capacity has been estimated in fly photoreceptors (Steveninck & Laughlin, 1996). However, in most previous work, the system was treated as a black-box and the analysis was performed from input–output measurements. This approach provides little insight into the internal factors that limit information transmission. To address this issue, we decomposed the black-box of one extensively studied system, the blowfly photoreceptor, into its elementary biophysical components, and derived a communication model. Since information channel capacity is a fundamental property of a communication channel, we quantified the effect of individual components on information capacity in the blowfly photoreceptor (Abshire & Andreou, 2001b). Neural Networks 18 (2005) 479–487 www.elsevier.com/locate/neunet 0893-6080/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.neunet.2005.06.004 * An abbreviated version of some portions of this article appeared in (Xu & Abshire, 2005a), published under the IEEE copyright. * Corresponding author. Tel.: C1 301 405 8974. E-mail addresses: pxu@glue.umd.edu (P. Xu), pabshire@glue.umd. edu (P. Abshire).