Profile hidden Markov models (profile-HMMs) are sensitive tools for remote protein homology detection, but the main scoring algorithms, Viterbi or Forward, require considerable time to search large sequence databases.

Dopaminergic neurons with distinct projection patterns and physiological properties compose memory subsystems in a brain. However, it is poorly understood whether or how they interact during complex learning. Here, we identify a feedforward circuit formed between dopamine subsystems and show that it is essential for second-order conditioning, an ethologically important form of higher-order associative learning. The Drosophila mushroom body comprises a series of dopaminergic compartments, each of which exhibits distinct memory dynamics. We find that a slow and stable memory compartment can serve as an effective “teacher” by instructing other faster and transient memory compartments via a single key interneuron, which we identify by connectome analysis and neurotransmitter prediction. This excitatory interneuron acquires enhanced response to reward-predicting odor after first-order conditioning and, upon activation, evokes dopamine release in the “student” compartments. These hierarchical connections between dopamine subsystems explain distinct properties of first- and second-order memory long known by behavioral psychologists.

The anterodorsal projection neuron lineage of Drosophila melanogaster produces 40 neuronal types in a stereotypic order. Here we take advantage of this complete lineage sequence to examine the role of known temporal fating factors, including Chinmo and the Hb/Kr/Pdm/Cas transcriptional cascade, within this diverse central brain lineage. Kr mutation affects the temporal fate of the neuroblast (NB) itself, causing a single fate to be skipped, whereas Chinmo null only elicits fate transformation of NB progeny without altering cell counts. Notably, Chinmo operates in two separate windows to prevent fate transformation (into the subsequent Chinmo-indenpendent fate) within each window. By contrast, Hb/Pdm/Cas play no detectable role, indicating that Kr either acts outside of the cascade identified in the ventral nerve cord or that redundancy exists at the level of fating factors. Therefore, hierarchical fating mechanisms operate within the lineage to generate neuronal diversity in an unprecedented fashion.

Single neurons in visual cortex provide unreliable measurements of visual features due to their high trial-to-trial variability. It is not known if this “noise” extends its effects over large neural populations to impair the global encoding of sensory stimuli. We recorded simultaneously from ∼20,000 neurons in mouse visual cortex and found that the neural population had discrimination thresholds of 0.3° in an orientation decoding task. These thresholds are ∼100 times smaller than those reported behaviorally in mice. The discrepancy between neural and behavioral discrimination could not be explained by the types of stimuli we used, by behavioral states or by the sequential nature of trial-by-trial perceptual learning tasks. These results imply that the limits of sensory perception in mice are not set by neural noise in sensory cortex, but by the limitations of downstream decoders.

The challenge of recovering the topology of massive neuronal circuits can potentially be met by high throughput Electron Microscopy (EM) imagery. Segmenting a 3-dimensional stack of EM images into the individual neurons is difficult, due to the low depth-resolution in existing high-throughput EM technology, such as serial section Transmission EM (ssTEM). In this paper we propose methods for detecting the high resolution locations of membranes from low depth-resolution images. We approach this problem using both a method that learns a discriminative, over-complete dictionary and a kernel SVM. We test this approach on tomographic sections produced in simulations from high resolution Focused Ion Beam (FIB) images and on low depth-resolution images acquired with ssTEM and evaluate our results by comparing it to manual labeling of this data.

We present a compressed domain scheme that is able to recognize and localize actions at high speeds. The recognition problem is posed as performing an action video query on a test video sequence. Our method is based on computing motion similarity using compressed domain features which can be extracted with low complexity. We introduce a novel motion correlation measure that takes into account differences in motion directions and magnitudes. Our method is appearance invariant, requires no prior segmentation, alignment or stabilization, and is able to localize actions in both space and time. We evaluated our method on a benchmark action video database consisting of 6 actions performed by 25 people under 3 different scenarios. Our proposed method achieved a classification accuracy of 90%, comparing favorably with existing methods in action classification accuracy, and is able to localize a template video of 80 x 64 pixels with 23 frames in a test video of 368 x 184 pixels with 835 frames in just 11 seconds, easily outperforming other methods in localization speed. We also perform a systematic investigation of the effects of various encoding options on our proposed approach. In particular, we present results on the compression-classification trade-off, which would provide valuable insight into jointly designing a system that performs video encoding at the camera front-end and action classification at the processing backend.

Random scattering and aberrations severely limit the imaging depth in optical microscopy. We introduce a rapid, parallel wavefront compensation technique that efficiently compensates even highly complex phase distortions. Using coherence gated backscattered light as a feedback signal, we focus light deep inside highly scattering brain tissue. We demonstrate that the same wavefront optimization technique can also be used to compensate spectral phase distortions in ultrashort laser pulses using nonlinear iterative feedback. We can restore transform limited pulse durations at any selected target location and compensate for dispersion that has occurred in the optical train and within the sample.

We describe a procedure for designing proteins with backbones produced by varying the parameters in the Crick coiled coil-generating equations. Combinatorial design calculations identify low-energy sequences for alternative helix supercoil arrangements, and the helices in the lowest-energy arrangements are connected by loop building. We design an antiparallel monomeric untwisted three-helix bundle with 80-residue helices, an antiparallel monomeric right-handed four-helix bundle, and a pentameric parallel left-handed five-helix bundle. The designed proteins are extremely stable (extrapolated ΔGfold > 60 kilocalories per mole), and their crystal structures are close to those of the design models with nearly identical core packing between the helices. The approach enables the custom design of hyperstable proteins with fine-tuned geometries for a wide range of applications.

Two-photon microscopy together with fluorescent proteins and fluorescent protein-based biosensors are commonly used tools in neuroscience. To enhance their experimental scope, it is important to optimize fluorescent proteins for two-photon excitation. Directed evolution of fluorescent proteins under one-photon excitation is common, but many one-photon properties do not correlate with two-photon properties. A simple system for expressing fluorescent protein mutants is colonies on an agar plate. The small focal volume of two-photon excitation makes creating a high throughput screen in this system a challenge for a conventional point-scanning approach. We present an instrument and accompanying software that solves this challenge by selectively scanning each colony based on a colony map captured under one-photon excitation. This instrument, called the GIZMO, can measure the two-photon excited fluorescence of 10,000 colonies in 7 hours. We show that the GIZMO can be used to evolve a fluorescent protein under two-photon excitation.

Membrane proteins play a tremendously important role in cell physiology and serve as a target for an increasing number of drugs. Structural information is key to understanding their function and for developing new strategies for combating disease. However, the complex physical chemistry associated with membrane proteins has made them more difficult to study than their soluble cousins. Electron crystallography has historically been a successful method for solving membrane protein structures and has the advantage of providing a native lipid environment for these proteins. Specifically, when membrane proteins form two-dimensional arrays within a lipid bilayer, electron microscopy can be used to collect images and diffraction and the corresponding data can be combined to produce a three-dimensional reconstruction, which under favorable conditions can extend to atomic resolution. Like X-ray crystallography, the quality of the structures are very much dependent on the order and size of the crystals. However, unlike X-ray crystallography, high-throughput methods for screening crystallization trials for electron crystallography are not in general use. In this chapter, we describe two alternative methods for high-throughput screening of membrane protein crystallization within the lipid bilayer. The first method relies on the conventional use of dialysis for removing detergent and thus reconstituting the bilayer; an array of dialysis wells in the standard 96-well format allows the use of a liquid-handling robot and greatly increases throughput. The second method relies on titration of cyclodextrin as a chelating agent for detergent; a specialized pipetting robot has been designed not only to add cyclodextrin in a systematic way, but to use light scattering to monitor the reconstitution process. In addition, the use of liquid-handling robots for making negatively stained grids and methods for automatically imaging samples in the electron microscope are described.