FUNCTION OF RETINAL CIRCUITS FOR NOISE REDUCTION

我的长期目标是发现神经回路是如何解决 信号处理。关键的问题是光学图像有噪声 (因为自然场景对比度低)，视网膜回路 噪音也很大(因为他们使用的泊松过程相对较小 数字：发射器量子很少，通道很少)。一般战略，以 改善和保护信噪比(SNR)是已知，例如信号 平均、带宽压缩和增益控制。但这些是怎么回事？ 在特定的神经回路中实施的情况并不成立。我提议 猫和侏儒β(X)神经节细胞的环路研究 猕猴的神经节细胞。这些像元类型对于精细空间至关重要 视觉，分别占视神经轴突的50%和90%。 这些电路的“原理线路图”实际上是完整的， 包括会聚的杆、锥和双极细胞的数量， 每一阶段的突触、电耦合部位、外侧部位 神经递质和突触后受体的连通性、特性。 大多数单个神经元类型的反应是已知的，包括 一些信号和噪声的幅度。最后，计算模型 整个电路的几个部件的(隔间的)已经 已建立(视锥-水平细胞、双极神经节细胞)并“调整”至 重现已知的反应，如感受野范围和 幅度。这些是大型模型(最多50,000个车厢) 由已建立的模拟器(NeuronC)管理。 我建议使用现有的模式，并在有需要时加以扩展，以及 模拟不同级别的光子、突触和通道噪波。这就做 评估这些噪声源对每个噪声源的贡献 电路中的阶段和特定电路特性的影响 改善/保持信噪比。具体地说，我将：L)比较 棒信号通过棒双极在高能光子中传输的β细胞 噪声(星光)来自杆信号，通过耦合到 中等光子噪声(黄昏)中的锥形双极电路。2)比较 当由前馈或/和控制时，双极单元输入阶段的SNR 反馈抑制。3)比较双极单元输出级的信噪比 受抑制性无长突突触控制的双极和无长突突触 神经节细胞的输入是不相关的或相关的(如“二元组”) Synapse)。4)评估每个回路对神经节细胞的噪声贡献 组件，包括尖峰生成器。5)表演《敏感度 分析“，以评估电路中的哪些参数影响最大 关于视网膜输出的信噪比以及从以下方面提高信噪比的成本 速度、可靠性和视网膜厚度。这个项目将会有所帮助 了解神经回路如何影响人类的空间敏锐度 幻象。

My long term goal is to discover how neural circuits solve problems of signal processing. The key problems are that optical images are noisy (because natural scenes have low contrast) and that retinal circuits are also noisy (because they employ Poisson processes with relatively small numbers: few transmitter quanta, few channels). General strategies to improve and protect signal/noise ratio (SNR) are known, such as signal averaging, bandwidth compression, and gain control. But how these are implemented in specific neural circuits is not established. I propose to study circuits to the beta (x) ganglion cell in cat and the midget (P) ganglion cell in monkey. These cell types are critical to fine spatial vision, contributing respectively 50% and 90% of axons in the optic nerve. The "schematic wiring diagrams" for these circuits are virtually complete, including numbers of converging rods, cones, and bipolar cells, number Of synapses at each stage, sites of electrical coupling, sites of lateral connectivity, identity of neural transmitters and postsynaptic receptors. The responses of most individual neuron types are known, including for some the signal and noise amplitudes. Finally, computational models (compartmental) of several components of the overall circuits have been established (cone-horizontal cell, bipolar-ganglion cell) and "tuned" to reproduce the known responses such as receptive field extent and amplitude. These are large-scale models (up to 50,000 compartments) governed by an established simulator (NeuronC). I propose to use the existing models, extending them where necessary, and simulate different levels of photon, synaptic, and channel noise. I will evaluate the respective contributions of these noise sources for each stage in the circuits and the effects of specific circuit features in improving/maintaining SNR. Specifically, I will: l) compare the SNR at the beta cell from rod signals transmitted via rod bipolar in high photon noise (starlight) to that from rod signals transmitted by coupling to the cone bipolar circuit in moderate photon noise (twilight). 2) compare the SNR at bipolar cell input stage when controlled by feedforward or/and feedback inhibition. 3) compare SNRs of bipolar cell output stage when controlled by inhibitory amacrine synapses where the bipolar and amacrine inputs to ganglion cell are uncorrelated or correlated (as at "dyad" synapse). 4) evaluate noise in ganglion cell contributed by each circuit component, including the spike generator. 5) perform "sensitivity analysis" to assess which parameters in the circuit have greatest affect on SNR at the retinal output and the costs of improving SNR in terms of speed, reliability, and retinal thickness. The project will help understand how neural circuits contribute to spatial acuity in human vision.