The icosahedron is the largest of the Platonic solids, and icosahedral protein structures are widely used in biological systems for packaging and transport. There has been considerable interest in repurposing such structures for applications ranging from targeted delivery to multivalent immunogen presentation. The ability to design proteins that self-assemble into precisely specified, highly ordered icosahedral structures would open the door to a new generation of protein containers with properties custom-tailored to specific applications. Here we describe the computational design of a 25-nanometre icosahedral nanocage that self-assembles from trimeric protein building blocks. The designed protein was produced in Escherichia coli, and found by electron microscopy to assemble into a homogenous population of icosahedral particles nearly identical to the design model. The particles are stable in 6.7 molar guanidine hydrochloride at up to 80 degrees Celsius, and undergo extremely abrupt, but reversible, disassembly between 2 molar and 2.25 molar guanidinium thiocyanate. The icosahedron is robust to genetic fusions: one or two copies of green fluorescent protein (GFP) can be fused to each of the 60 subunits to create highly fluorescent 'standard candles' for use in light microscopy, and a designed protein pentamer can be placed in the centre of each of the 20 pentameric faces to modulate the size of the entrance/exit channels of the cage. Such robust and customizable nanocages should have considerable utility in targeted drug delivery, vaccine design and synthetic biology.

We describe a general approach to designing two-dimensional (2D) protein arrays mediated by noncovalent protein-protein interfaces. Protein homo-oligomers are placed into one of the seventeen 2D layer groups, the degrees of freedom of the lattice are sampled to identify configurations with shape-complementary interacting surfaces, and the interaction energy is minimized using sequence design calculations. We used the method to design proteins that self-assemble into layer groups P 3 2 1, P 4 21 2, and P 6. Projection maps of micrometer-scale arrays, assembled both in vitro and in vivo, are consistent with the design models and display the target layer group symmetry. Such programmable 2D protein lattices should enable new approaches to structure determination, sensing, and nanomaterial engineering.

Electronic and biological systems both perform complex information processing, but they use very different techniques. Though electronics has the advantage in raw speed, biological systems have the edge in many other areas. They can be produced, and indeed self-reproduce, without expensive and finicky factories. They are tolerant of manufacturing defects, and learn and adapt for better performance. In many cases they can self-repair damage. These advantages suggest that biological systems might be useful in a wide variety of tasks involving information processing. So far, all attempts to use the nervous system of a living organism for information processing have involved selective breeding of existing organisms. This approach, largely independent of the details of internal operation, is used since we do not yet understand how neural systems work, nor exactly how they are constructed. However, as our knowledge increases, the day will come when we can envision useful nervous systems and design them based upon what we want them to do, as opposed to variations on what has been already built. We will then need tools, corresponding to our Electronic Design Automation tools, to help with the design. This paper is concerned with what such tools might look like.

Epilepsy afflicts 1-2% of the world’s population and often goes untreated; nearly 70% of those with a form of epilepsy fail to receive proper treatment. Therefore, there is great demand for the design of novel, effective anticonvulsants to combat epilepsy in its numerous forms. Previously, alpha-hydroxy-alpha-phenylcaprolactam was found to have rather potent antiepileptic activity [anti-maximal electroshock (MES) ED(50)=63 mg/kg and anti-subcutaneous Metrazol (scMet) ED(50)=74 mg/kg] when administered intraperitoneally in mice. We focused our attention on the development of this compound through traditional medicinal chemistry techniques-including the Topliss approach, isosteric replacement, methylene insertion, and rigid analogue approach-in the hopes of determining the effect of caprolactam alpha-substitution and other structural modifications on anticonvulsant activity. A number of the desired targets were successfully synthesized and submitted to the Anticonvulsant Screening Program of the National Institute of Neurological Disorders and Stroke (NINDS). Phase I results were quite promising for at least three of the compounds: alpha-ethynyl-alpha-hydroxycaprolactam (10), alpha-benzyl-alpha-hydroxycaprolactam (11), and alpha-hydroxy-alpha-(phenylethynyl)caprolactam (13). Phase II results for 11 strongly suggested it as a new structural class for further development, as it exhibited an anti-MES T.I. in excess of 4.0. Further, the potent activity of 13 in all models also pointed to the substituted alkynylcaprolactams as a new anticonvulsant structural class.

Cells form networks in animal tissues through synaptic, chemical, and adhesive links. Invertebrate muscle cells often connect to other cells through desmosomes, adhesive junctions anchored by intermediate filaments. To study desmosomal networks, we skeletonised 853 muscle cells and their desmosomal partners in volume electron microscopy data covering an entire larva of the annelid . Muscle cells adhere to each other, to epithelial, glial, ciliated, and bristle-producing cells and to the basal lamina, forming a desmosomal connectome of over 2000 cells. The aciculae - chitin rods that form an endoskeleton in the segmental appendages - are highly connected hubs in this network. This agrees with the many degrees of freedom of their movement, as revealed by video microscopy. Mapping motoneuron synapses to the desmosomal connectome allowed us to infer the extent of tissue influenced by motoneurons. Our work shows how cellular-level maps of synaptic and adherent force networks can elucidate body mechanics.

We describe new detachable floating glass micropipette electrode devices that provide targeted action potential recordings in active moving organs without requiring constant mechanical constraint or pharmacological inhibition of tissue motion. The technology is based on the concept of a glass micropipette electrode that is held firmly during cell targeting and intracellular insertion, after which a 100µg glass microelectrode, a "microdevice", is gently released to remain within the moving organ. The microdevices provide long-term recordings of action potentials, even during millimeter-scale movement of tissue in which the device is embedded. We demonstrate two different glass micropipette electrode holding and detachment designs appropriate for the heart (sharp glass microdevices for cardiac myocytes in rats, guinea pigs and humans) and the brain (patch glass microdevices for neurons in rats). We explain how microdevices enable measurements of multiple cells within a moving organ that are typically difficult with other technologies. Using sharp microdevices, action potential duration (APD) was monitored continuously for 15 minutes in unconstrained perfused hearts during global ischemia-reperfusion, providing beat-to-beat measurements of changes in APD. Action potentials from neurons in the hippocampus of anaesthetized rats were measured with patch microdevices, which provided stable base potentials during long-term recordings. Our results demonstrate that detachable microdevices are an elegant and robust tool to record electrical activity with high temporal resolution and cellular level localization without disturbing the physiological working conditions of the organ.

The prediction of pathological changes on single cell behaviour is a challenging task for deep learning models. Indeed, in self-supervised learning methods, no prior labels are used for the training and all of the information for event predictions are extracted from the data themselves. We present here a novel self-supervised learning model for the detection of anomalies in a given cell population, StArDusTS. Cells are monitored over time, and analysed to extract time-series of dry mass values. We assessed its performances on different cell lines, showing a precision of 96% in the automatic detection of anomalies. Additionally, anomaly detection was also associated with cell measurement errors inherent to the acquisition or analysis pipelines, leading to an improvement of the upstream methods for feature extraction. Our results pave the way to novel architectures for the continuous monitoring of cell cultures in applied research or bioproduction applications, and for the prediction of pathological cellular changes.

Population recordings of calcium activity are a major source of insight into neural function. Large dataset sizes often require automated methods, but automation can introduce errors that are difficult to detect. Here we show that automatic time course estimation can sometimes lead to significant misattribution errors, in which fluorescence is ascribed to the wrong cell. Misattribution arises when the shapes of overlapping cells are imperfectly defined, or when entire cells or processes are not identified, and misattribution can even be produced by methods specifically designed to handle overlap. To diagnose this problem, we develop a transient-by-transient metric and a visualization tool that allow users to quickly assess the degree of misattribution in large populations. To filter out misattribution, we also design a robust estimator that explicitly accounts for contaminating signals in a generative model. Our methods can be combined with essentially any cell finding technique, empowering users to diagnose and correct at large scale a problem that has the potential to significantly alter scientific conclusions.

In this work, we address the problem of precisely localizing key frames of an action, for example, the precise time that a pitcher releases a baseball, or the precise time that a crowd begins to applaud. Key frame localization is a largely overlooked and important action-recognition problem, for example in the field of neuroscience, in which we would like to understand the neural activity that produces the start of a bout of an action. To address this problem, we introduce a novel structured loss function that properly weights the types of errors that matter in such applications: it more heavily penalizes extra and missed action start detections over small misalignments. Our structured loss is based on the best matching between predicted and labeled action starts. We train recurrent neural networks (RNNs) to minimize differentiable approximations of this loss. To evaluate these methods, we introduce the Mouse Reach Dataset, a large, annotated video dataset of mice performing a sequence of actions. The dataset was collected and labeled by experts for the purpose of neuroscience research. On this dataset, we demonstrate that our method outperforms related approaches and baseline methods using an unstructured loss.

The interaural time difference (ITD) is a major cue to sound localization along the horizontal plane. The maximum natural ITD occurs when a sound source is positioned opposite to one ear. We examined the ability of owls and humans to detect large ITDs in sounds presented through headphones. Stimuli consisted of either broad or narrow bands of Gaussian noise, 100 ms in duration. Using headphones allowed presentation of ITDs that are greater than the maximum natural ITD. Owls were able to discriminate a sound leading to the left ear from one leading to the right ear, for ITDs that are 5 times the maximum natural delay. Neural recordings from optic-tectum neurons, however, show that best ITDs are usually well within the natural range and are never as large as ITDs that are behaviorally discriminable. A model of binaural crosscorrelation with short delay lines is shown to explain behavioral detection of large ITDs. The model uses curved trajectories of a cross-correlation pattern as the basis for detection. These trajectories represent side peaks of neural ITD-tuning curves and successfully predict localization reversals by both owls and human subjects.