Molecular self-assembly offers a 'bottom-up' route to fabrication with subnanometre precision of complex structures from simple components. DNA has proved to be a versatile building block for programmable construction of such objects, including two-dimensional crystals, nanotubes, and three-dimensional wireframe nanopolyhedra. Templated self-assembly of DNA into custom two-dimensional shapes on the megadalton scale has been demonstrated previously with a multiple-kilobase 'scaffold strand' that is folded into a flat array of antiparallel helices by interactions with hundreds of oligonucleotide 'staple strands'. Here we extend this method to building custom three-dimensional shapes formed as pleated layers of helices constrained to a honeycomb lattice. We demonstrate the design and assembly of nanostructures approximating six shapes-monolith, square nut, railed bridge, genie bottle, stacked cross, slotted cross-with precisely controlled dimensions ranging from 10 to 100 nm. We also show hierarchical assembly of structures such as homomultimeric linear tracks and heterotrimeric wireframe icosahedra. Proper assembly requires week-long folding times and calibrated monovalent and divalent cation concentrations. We anticipate that our strategy for self-assembling custom three-dimensional shapes will provide a general route to the manufacture of sophisticated devices bearing features on the nanometre scale.

/pdf/self-assembly-of-dna-into-nanoscale-three-dimensional-shapes-2g73511y4b.pdf

Self-assembly of DNA into nanoscale three-dimensional shapes

We describe an autonomous DNA nanorobot capable of transporting molecular payloads to cells, sensing cell surface inputs for conditional, triggered activation, and reconfiguring its structure for payload delivery. The device can be loaded with a variety of materials in a highly organized fashion and is controlled by an aptamer-encoded logic gate, enabling it to respond to a wide array of cues. We implemented several different logical AND gates and demonstrate their efficacy in selective regulation of nanorobot function. As a proof of principle, nanorobots loaded with combinations of antibody fragments were used in two different types of cell-signaling stimulation in tissue culture. Our prototype could inspire new designs with different selectivities and biologically active payloads for cell-targeting tasks.

http://science.sciencemag.org/content/sci/335/6070/831.full.pdf

A logic-gated nanorobot for targeted transport of molecular payloads

We demonstrate the ability to engineer complex shapes that twist and curve at the nanoscale from DNA. Through programmable self-assembly, strands of DNA are directed to form a custom-shaped bundle of tightly cross-linked double helices, arrayed in parallel to their helical axes. Targeted insertions and deletions of base pairs cause the DNA bundles to develop twist of either handedness or to curve. The degree of curvature could be quantitatively controlled, and a radius of curvature as tight as 6 nanometers was achieved. We also combined multiple curved elements to build several different types of intricate nanostructures, such as a wireframe beach ball or square-toothed gears.

/pdf/folding-dna-into-twisted-and-curved-nanoscale-shapes-16pdm71fsl.pdf

Folding DNA into Twisted and Curved Nanoscale Shapes

For over half a century, the biological roles of nucleic acids as catalytic enzymes, intracellular regulatory molecules, and the carriers of genetic information have been studied extensively. More recently, the sequence-specific binding properties of DNA have been exploited to direct the assembly of materials at the nanoscale. Integral to any methodology focused on assembling matter from smaller pieces is the idea that final structures have well-defined spacings, orientations, and stereo-relationships. This requirement can be met by using DNA-based constructs that present oriented nanoscale bonding elements from rigid core units. Here, we draw analogy between such building blocks and the familiar chemical concepts of "bonds" and "valency" and review two distinct but related strategies that have used this design principle in constructing new configurations of matter.

http://science.sciencemag.org/content/sci/347/6224/1260901.full.pdf

Programmable materials and the nature of the DNA bond

We describe a simple and robust method to construct complex three-dimensional (3D) structures by using short synthetic DNA strands that we call “DNA bricks.” In one-step annealing reactions, bricks with hundreds of distinct sequences self-assemble into prescribed 3D shapes. Each 32-nucleotide brick is a modular component; it binds to four local neighbors and can be removed or added independently. Each 8–base pair interaction between bricks defines a voxel with dimensions of 2.5 by 2.5 by 2.7 nanometers, and a master brick collection defines a “molecular canvas” with dimensions of 10 by 10 by 10 voxels. By selecting subsets of bricks from this canvas, we constructed a panel of 102 distinct shapes exhibiting sophisticated surface features, as well as intricate interior cavities and tunnels.

/pdf/three-dimensional-structures-self-assembled-from-dna-bricks-31mzkon68k.pdf

Three-Dimensional Structures Self-Assembled from DNA Bricks

DNA nanotechnology exploits the programmable specificity afforded by base-pairing to produce self-assembling macromolecular objects of custom shape. For building megadalton-scale DNA nanostructures, a long 'scaffold' strand can be employed to template the assembly of hundreds of oligonucleotide 'staple' strands into a planar antiparallel array of cross-linked helices. We recently adapted this 'scaffolded DNA origami' method to producing 3D shapes formed as pleated layers of double helices constrained to a honeycomb lattice. However, completing the required design steps can be cumbersome and time-consuming. Here we present caDNAno, an open-source software package with a graphical user interface that aids in the design of DNA sequences for folding 3D honeycomb-pleated shapes A series of rectangular-block motifs were designed, assembled, and analyzed to identify a well-behaved motif that could serve as a building block for future studies. The use of caDNAno significantly reduces the effort required to design 3D DNA-origami structures. The software is available at http://cadnano.org/, along with example designs and video tutorials demonstrating their construction. The source code is released under the MIT license.

/pdf/rapid-prototyping-of-3d-dna-origami-shapes-with-cadnano-332a4wb24o.pdf

Rapid prototyping of 3D DNA-origami shapes with caDNAno

Deep neural networks are state of the art methods for many learning tasks due to their ability to extract increasingly better features at each network layer. However, the improved performance of additional layers in a deep network comes at the cost of added latency and energy usage in feedforward inference. As networks continue to get deeper and larger, these costs become more prohibitive for real-time and energy-sensitive applications. To address this issue, we present BranchyNet, a novel deep network architecture that is augmented with additional side branch classifiers. The architecture allows prediction results for a large portion of test samples to exit the network early via these branches when samples can already be inferred with high confidence. BranchyNet exploits the observation that features learned at an early layer of a network may often be sufficient for the classification of many data points. For more difficult samples, which are expected less frequently, BranchyNet will use further or all network layers to provide the best likelihood of correct prediction. We study the BranchyNet architecture using several well-known networks (LeNet, AlexNet, ResNet) and datasets (MNIST, CIFAR10) and show that it can both improve accuracy and significantly reduce the inference time of the network.

BranchyNet: Fast inference via early exiting from deep neural networks

We propose distributed deep neural networks (DDNNs) over distributed computing hierarchies, consisting of the cloud, the edge (fog) and end devices. While being able to accommodate inference of a deep neural network (DNN) in the cloud, a DDNN also allows fast and localized inference using shallow portions of the neural network at the edge and end devices. When supported by a scalable distributed computing hierarchy, a DDNN can scale up in neural network size and scale out in geographical span. Due to its distributed nature, DDNNs enhance sensor fusion, system fault tolerance and data privacy for DNN applications. In implementing a DDNN, we map sections of a DNN onto a distributed computing hierarchy. By jointly training these sections, we minimize communication and resource usage for devices and maximize usefulness of extracted features which are utilized in the cloud. The resulting system has built-in support for automatic sensor fusion and fault tolerance. As a proof of concept, we show a DDNN can exploit geographical diversity of sensors to improve object recognition accuracy and reduce communication cost. In our experiment, compared with the traditional method of offloading raw sensor data to be processed in the cloud, DDNN locally processes most sensor data on end devices while achieving high accuracy and is able to reduce the communication cost by a factor of over 20x.

/pdf/distributed-deep-neural-networks-over-the-cloud-the-edge-and-5129uy60s9.pdf

Distributed Deep Neural Networks Over the Cloud, the Edge and End Devices

Distributed Deep Neural Networks over the Cloud, the Edge and End Devices

We study embedded Binarized Neural Networks (eBNNs) with the aim of allowing current binarized neural networks (BNNs) in the literature to perform feedforward inference efficiently on small embedded devices. We focus on minimizing the required memory footprint, given that these devices often have memory as small as tens of kilobytes (KB). Beyond minimizing the memory required to store weights, as in a BNN, we show that it is essential to minimize the memory used for temporaries which hold intermediate results between layers in feedforward inference. To accomplish this, eBNN reorders the computation of inference while preserving the original BNN structure, and uses just a single floating-point temporary for the entire neural network. All intermediate results from a layer are stored as binary values, as opposed to floating-points used in current BNN implementations, leading to a 32x reduction in required temporary space. We provide empirical evidence that our proposed eBNN approach allows efficient inference (10s of ms) on devices with severely limited memory (10s of KB). For example, eBNN achieves 95% accuracy on the MNIST dataset running on an Intel Curie with only 15 KB of usable memory with an inference runtime of under 50 ms per sample. To ease the development of applications in embedded contexts, we make our source code available that allows users to train and discover eBNN models for a learning task at hand, which fit within the memory constraint of the target device.

Surat Teerapittayanon

Papers

Rapid prototyping of 3D DNA-origami shapes with caDNAno

BranchyNet: Fast inference via early exiting from deep neural networks

Distributed Deep Neural Networks Over the Cloud, the Edge and End Devices

Distributed Deep Neural Networks over the Cloud, the Edge and End Devices

Embedded Binarized Neural Networks