Search

Masked Image Modeling (MIM) achieves outstanding success in self-supervised representation learning. Unfortunately, MIM models typically have huge computational burden and slow learning process, which is an inevitable obstacle for their industrial applications. Although the lower layers play the key role in MIM, existing MIM models conduct reconstruction task only at the top layer of encoder. The lower layers are not explicitly guided and the interaction among their patches is only used for calculating new activations. Considering the reconstruction task requires non-trivial inter-patch interactions to reason target signals, we apply it to multiple local layers including lower and upper layers. Further, since the multiple layers expect to learn the information of different scales, we design local multi-scale reconstruction, where the lower and upper layers reconstruct fine-scale and coarse-scale supervision signals respectively. This design not only accelerates the representation learning process by explicitly guiding multiple layers, but also facilitates multi-scale semantical understanding to the input. Extensive experiments show that with significantly less pre-training burden, our model achieves comparable or better performance on classification, detection and segmentation tasks than existing MIM models., Comment: CVPR 2023

Sequential immunization elicits antibodies against the V3-glycan patch of HIV-1 Env in rhesus macaques and also induces off-target antibody responses.

Receptor activity-modifying proteins (RAMPs) modulate the activity of many Family B GPCRs. We show that RAMP2 directly interacts with the glucagon receptor (GCGR), a Family B GPCR responsible for blood sugar homeostasis, and broadly inhibits receptor-induced downstream signaling. HDX-MS experiments demonstrate that RAMP2 enhances local flexibility in select locations in and near the receptor extracellular domain (ECD) and in the 6th transmembrane helix, whereas smFRET experiments show that this ECD disorder results in the inhibition of active and intermediate states of the intracellular surface. We determined the cryo-EM structure of the GCGR-Gs complex at 2.9 Å resolution in the presence of RAMP2. RAMP2 apparently does not interact with GCGR in an ordered manner; however, the receptor ECD is indeed largely disordered along with rearrangements of several intracellular hallmarks of activation. Our studies suggest that RAMP2 acts as a negative allosteric modulator of GCGR by enhancing conformational sampling of the ECD.

Current state-of-the-art results in computer vision depend in part on fine-tuning large pre-trained vision models. However, with the exponential growth of model sizes, the conventional full fine-tuning, which needs to store a individual network copy for each tasks, leads to increasingly huge storage and transmission overhead. Adapter-based Parameter-Efficient Tuning (PET) methods address this challenge by tuning lightweight adapters inserted into the frozen pre-trained models. In this paper, we investigate how to make adapters even more efficient, reaching a new minimum size required to store a task-specific fine-tuned network. Inspired by the observation that the parameters of adapters converge at flat local minima, we find that adapters are resistant to noise in parameter space, which means they are also resistant to low numerical precision. To train low-precision adapters, we propose a computational-efficient quantization method which minimizes the quantization error. Through extensive experiments, we find that low-precision adapters exhibit minimal performance degradation, and even 1-bit precision is sufficient for adapters. The experimental results demonstrate that 1-bit adapters outperform all other PET methods on both the VTAB-1K benchmark and few-shot FGVC tasks, while requiring the smallest storage size. Our findings show, for the first time, the significant potential of quantization techniques in PET, providing a general solution to enhance the parameter efficiency of adapter-based PET methods. Code: https://github.com/JieShibo/PETL-ViT, Comment: Accepted to ICCV 2023

Cells sense and respond to mechanical stimuli for activation, proliferation, migration, and differentiation. The associated mechanosensing and biomechanical properties of cells and tissues are significantly implicated in the context of cancer, fibrosis, dementia, and cardiovascular diseases. To gain more mechanobiology insights, dynamic force spectroscopies (DFSs), particularly optical tweezers (OT), have been further advanced to enable in situ force measurement and subcellular manipulation from the outer cell membrane to the organelles inside of a cell. In this review, we first explain the classic OT-DFS rationales and discuss their applications to protein biophysics, extracellular biomechanics, and receptor-mediated cell mechanosensing. As a non-invasive technique, optical tweezers’s unique advantages in probing cytoplasmic protein behaviors and manipulating organelles inside living cells have been increasingly explored in recent years. Hereby, we then introduce and highlight the emerging OT rationales for intracellular force measurement including refractive index matching, active–passive calibration, and change of light momentum. These new approaches enable intracellular OT-DFS and mechanical measurements with respect to intracellular motor stepping, cytosolic micro-rheology, and biomechanics of irregularly shaped nuclei and vesicles. Last but not least, we foresee future OT upgrades with respect to overcoming phototoxicity and system drifting for longer duration live-cell measurements; multimodal integration with advanced imaging and nanotechnology to obtain higher spatiotemporal resolution; and developing simultaneous, automated, and artificial intelligence–inspired multi-trap systems to achieve high throughput. These further developments will grant unprecedented accessibility of OT-DFS and force measurement nanotools to a wider biomedical research community, ultimately opening the floodgates for exciting live-cell mechanobiology and novel therapeutic discoveri

Detergents have been major contributors to membrane-protein structural study for decades. However, membrane proteins solubilized in conventional detergents tend to aggregate or denature over time. Stability of large eukaryotic membrane proteins with complex structures tends to be particularly poor, necessitating development of novel detergents with improved properties. Here, we prepared a novel class of detergents, designated 3,4-bis(hydroxymethyl)hexane-1,6-diol-based maltosides (HDMs). When tested on three membrane proteins, including two G-protein-coupled receptors (GPCRs), the new detergents displayed significantly better behaviors compared with DDM. Moreover, the HDMs were superior or comparable to LMNG, an amphiphile widely used for GPCR structural study. An optimal balance of detergent rigidity vs. flexibility of the HDMs is likely responsible for their favorable behaviors toward membrane-protein stability. Thus, the current study not only introduces the HDMs, with significant potential for membrane-protein structural study, but also suggests a useful guideline for designing novel detergents for membrane-protein research.

Contrastive learning between different views of the data achieves outstanding success in the field of self-supervised representation learning and the learned representations are useful in broad downstream tasks. Since all supervision information for one view comes from the other view, contrastive learning approximately obtains the minimal sufficient representation which contains the shared information and eliminates the non-shared information between views. Considering the diversity of the downstream tasks, it cannot be guaranteed that all task-relevant information is shared between views. Therefore, we assume the non-shared task-relevant information cannot be ignored and theoretically prove that the minimal sufficient representation in contrastive learning is not sufficient for the downstream tasks, which causes performance degradation. This reveals a new problem that the contrastive learning models have the risk of over-fitting to the shared information between views. To alleviate this problem, we propose to increase the mutual information between the representation and input as regularization to approximately introduce more task-relevant information, since we cannot utilize any downstream task information during training. Extensive experiments verify the rationality of our analysis and the effectiveness of our method. It significantly improves the performance of several classic contrastive learning models in downstream tasks. Our code is available at https://github.com/Haoqing-Wang/InfoCL., Comment: Accepted by CVPR 2022 as Oral presentation

Membrane proteins are of biological and pharmaceutical significance. However, their structural study is extremely challenging mainly due to the fact that only a small number of chemical tools is suitable for stabilizing membrane proteins in solution. Detergents are widely used in membrane protein study, but conventional detergents are generally poor at stabilizing challenging membrane proteins such as G protein-coupled receptors and protein complexes. In the current study, we prepared tandem triazine-based maltosides (TZMs) with two amphiphilic triazine units connected by different diamine linkers, hydrazine (TZM-Hs) and 1,2-ethylenediamine (TZM-Es). These TZMs were consistently superior to a gold standard detergent (DDM) in terms of stabilizing a few membrane proteins. In addition, the TZM-Es containing a long linker showed more general protein stabilization efficacy with multiple membrane proteins than the TZM-Hs containing a short linker. This result indicates that introduction of the flexible1,2-ethylenediamine linker between two rigid triazine rings enables the TZM-Es to fold into favourable conformations in order to promote membrane protein stability. The novel concept of detergent foldability introduced in the current study has potential use for rational detergent design and membrane protein applications.

Cells sense and respond to mechanical stimuli for activation, proliferation, migration, and differentiation. The associated mechanosensing and biomechanical properties of cells and tissues are significantly implicated in the context of cancer, fibrosis, dementia, and cardiovascular diseases. To gain more mechanobiology insights, dynamic force spectroscopies (DFSs), particularly optical tweezers (OT), have been further advanced to enable in situ force measurement and subcellular manipulation from the outer cell membrane to the organelles inside of a cell. In this review, we first explain the classic OT-DFS rationales and discuss their applications to protein biophysics, extracellular biomechanics, and receptor-mediated cell mechanosensing. As a non-invasive technique, optical tweezers’s unique advantages in probing cytoplasmic protein behaviors and manipulating organelles inside living cells have been increasingly explored in recent years. Hereby, we then introduce and highlight the emerging OT rationales for intracellular force measurement including refractive index matching, active–passive calibration, and change of light momentum. These new approaches enable intracellular OT-DFS and mechanical measurements with respect to intracellular motor stepping, cytosolic micro-rheology, and biomechanics of irregularly shaped nuclei and vesicles. Last but not least, we foresee future OT upgrades with respect to overcoming phototoxicity and system drifting for longer duration live-cell measurements; multimodal integration with advanced imaging and nanotechnology to obtain higher spatiotemporal resolution; and developing simultaneous, automated, and artificial intelligence–inspired multi-trap systems to achieve high throughput. These further developments will grant unprecedented accessibility of OT-DFS and force measurement nanotools to a wider biomedical research community, ultimately opening the floodgates for exciting live-cell mechanobiology and novel therapeutic discoveri

Membrane proteins are of biological and pharmaceutical significance. However, their structural study is extremely challenging mainly due to the fact that only a small number of chemical tools is suitable for stabilizing membrane proteins in solution. Detergents are widely used in membrane protein study, but conventional detergents are generally poor at stabilizing challenging membrane proteins such as G protein-coupled receptors and protein complexes. In the current study, we prepared tandem triazine-based maltosides (TZMs) with two amphiphilic triazine units connected by different diamine linkers, hydrazine (TZM-Hs) and 1,2-ethylenediamine (TZM-Es). These TZMs were consistently superior to a gold standard detergent (DDM) in terms of stabilizing a few membrane proteins. In addition, the TZM-Es containing a long linker showed more general protein stabilization efficacy with multiple membrane proteins than the TZM-Hs containing a short linker. This result indicates that introduction of the flexible1,2-ethylenediamine linker between two rigid triazine rings enables the TZM-Es to fold into favourable conformations in order to promote membrane protein stability. The novel concept of detergent foldability introduced in the current study has potential use for rational detergent design and membrane protein applications.

A biomembrane force probe (BFP) has recently emerged as a native-cell-surface or in situ dynamic force spectroscopy (DFS) nanotool that can measure single-molecular binding kinetics, assess mechanical properties of ligand-receptor interactions, visualize protein dynamic conformational changes and more excitingly elucidate receptor mediated cell mechanosensing mechanisms. More recently, BFP has been used to measure the spring constant of molecular bonds. This protocol describes the step-by-step procedure to perform molecular spring constant DFS analysis. Specifically, two BFP operation modes are discussed, namely the Bead-Cell and Bead-Bead modes. This protocol focuses on deriving spring constants of the molecular bond and cell from DFS raw data.

A biomembrane force probe (BFP) has recently emerged as a native-cell-surface or in situ dynamic force spectroscopy (DFS) nanotool that can measure single-molecular binding kinetics, assess mechanical properties of ligand-receptor interactions, visualize protein dynamic conformational changes and more excitingly elucidate receptor mediated cell mechanosensing mechanisms. More recently, BFP has been used to measure the spring constant of molecular bonds. This protocol describes the step-by-step procedure to perform molecular spring constant DFS analysis. Specifically, two BFP operation modes are discussed, namely the Bead-Cell and Bead-Bead modes. This protocol focuses on deriving spring constants of the molecular bond and cell from DFS raw data.

Membrane protein structures provide atomic level insight into essential biochemical processes and facilitate protein structure-based drug design. However, the inherent instability of these bio-macromolecules outside lipid bilayers hampers their structural and functional study. Detergent micelles can be used to solubilize and stabilize these membrane-inserted proteins in aqueous solution, thereby enabling their downstream characterizations. Membrane proteins encapsulated in detergent micelles tend to denature and aggregate over time, highlighting the need for development of new amphiphiles effective for protein solubility and stability. In this work, we present newly-designed maltoside detergents containing a pendant chain attached to a glycerol-decorated tris(hydroxymethyl)methane (THM) core, designated GTMs. One set of the GTMs has a hydrophobic pendant (ethyl chain; E-GTMs), and the other set has a hydrophilic pendant (methoxyethoxylmethyl chain; M-GTMs) placed in the hydrophobic-hydrophilic interfaces. The two sets of GTMs displayed profoundly different behaviors in terms of detergent self-assembly and protein stabilization efficacy. These behaviors mainly arise from the polarity difference between two pendants (ethyl and methoxyethoxylmethyl chains) that results in a large variation in detergent conformation between these sets of GTMs in aqueous media. The resulting high hydrophobic density in the detergent micelle interior is likely responsible for enhanced efficacy of the M-GTMs for protein stabilization compared to the E-GTMs and a gold standard detergent DDM. A representative GTM, M-GTM-O12, was more effective for protein stability than some recently developed detergents including LMNG. This is the first case study investigating the effect of pendant polarity on detergent geometry correlated with detergent efficacy for protein stabilization. Statement of significance This study introduces new amphiphiles for use as biochemical tools in membrane protein st

A biomembrane force probe (BFP) has recently emerged as a native-cell-surface or in situ dynamic force spectroscopy (DFS) nanotool that can measure single-molecular binding kinetics, assess mechanical properties of ligand-receptor interactions, visualize protein dynamic conformational changes and more excitingly elucidate receptor mediated cell mechanosensing mechanisms. More recently, BFP has been used to measure the spring constant of molecular bonds. This protocol describes the step-by-step procedure to perform molecular spring constant DFS analysis. Specifically, two BFP operation modes are discussed, namely the Bead-Cell and Bead-Bead modes. This protocol focuses on deriving spring constants of the molecular bond and cell from DFS raw data.

Membrane protein structures provide a fundamental understanding of their molecular actions and are of importance for drug development. Detergents are widely used to solubilize, stabilize, and crystallize membrane proteins, but membrane proteins solubilized in conventional detergents are prone to denaturation and aggregation. Thus, developing novel detergents with enhanced efficacy for protein stabilization remains important. We report herein the design and synthesis of a class of phenol-derived maltoside detergents. Using two different linkers, we prepared two sets of new detergents, designated maltose-bis(hydroxymethyl)phenol (MBPs) and maltose-tris(hydroxymethyl)phenol (MTPs). The evaluation of these detergents with three transporters and two G-protein coupled receptors allowed us to identify a couple of new detergents (MBP-C9 and MTP-C12) that consistently conferred enhanced stability to all tested proteins compared to a gold standard detergent (DDM). Furthermore, the data analysis based on the detergent structures provides key detergent features responsible for membrane protein stabilization that together will facilitate the future design of novel detergents.

Membrane protein structures provide atomic level insight into essential biochemical processes and facilitate protein structure-based drug design. However, the inherent instability of these bio-macromolecules outside lipid bilayers hampers their structural and functional study. Detergent micelles can be used to solubilize and stabilize these membrane-inserted proteins in aqueous solution, thereby enabling their downstream characterizations. Membrane proteins encapsulated in detergent micelles tend to denature and aggregate over time, highlighting the need for development of new amphiphiles effective for protein solubility and stability. In this work, we present newly-designed maltoside detergents containing a pendant chain attached to a glycerol-decorated tris(hydroxymethyl)methane (THM) core, designated GTMs. One set of the GTMs has a hydrophobic pendant (ethyl chain; E-GTMs), and the other set has a hydrophilic pendant (methoxyethoxylmethyl chain; M-GTMs) placed in the hydrophobic-hydrophilic interfaces. The two sets of GTMs displayed profoundly different behaviors in terms of detergent self-assembly and protein stabilization efficacy. These behaviors mainly arise from the polarity difference between two pendants (ethyl and methoxyethoxylmethyl chains) that results in a large variation in detergent conformation between these sets of GTMs in aqueous media. The resulting high hydrophobic density in the detergent micelle interior is likely responsible for enhanced efficacy of the M-GTMs for protein stabilization compared to the E-GTMs and a gold standard detergent DDM. A representative GTM, M-GTM-O12, was more effective for protein stability than some recently developed detergents including LMNG. This is the first case study investigating the effect of pendant polarity on detergent geometry correlated with detergent efficacy for protein stabilization. Statement of significance This study introduces new amphiphiles for use as biochemical tools in membrane protein st

Few-shot classification aims to recognize unseen classes with few labeled samples from each class. Many meta-learning models for few-shot classification elaborately design various task-shared inductive bias (meta-knowledge) to solve such tasks, and achieve impressive performance. However, when there exists the domain shift between the training tasks and the test tasks, the obtained inductive bias fails to generalize across domains, which degrades the performance of the meta-learning models. In this work, we aim to improve the robustness of the inductive bias through task augmentation. Concretely, we consider the worst-case problem around the source task distribution, and propose the adversarial task augmentation method which can generate the inductive bias-adaptive 'challenging' tasks. Our method can be used as a simple plug-and-play module for various meta-learning models, and improve their cross-domain generalization capability. We conduct extensive experiments under the cross-domain setting, using nine few-shot classification datasets: mini-ImageNet, CUB, Cars, Places, Plantae, CropDiseases, EuroSAT, ISIC and ChestX. Experimental results show that our method can effectively improve the few-shot classification performance of the meta-learning models under domain shift, and outperforms the existing works. Our code is available at https://github.com/Haoqing-Wang/CDFSL-ATA., Comment: Accepted by IJCAI-21 (the 30th International Joint Conference on Artificial Intelligence) Main Track

HIV-1 Env, a trimer of gp120-gp41 heterodimers, mediates membrane fusion after binding host receptor CD4. Receptor binding displaces V1V2 loops from Env's apex, allowing coreceptor binding and opening Env to enable gp41-mediated fusion. We present 3.54 Å and 4.06 Å cryoelectron microscopy structures of partially open soluble native-like Env trimers (SOSIPs) bound to CD4. One structure, a complex with a coreceptor-mimicking antibody that binds both CD4 and gp120, stabilizes the displaced V1V2 and reveals its structure. Comparing partially and fully open Envs with closed Envs shows that gp41 rearrangements are independent of the CD4-induced rearrangements that result in V1V2 displacement and formation of a 4-stranded bridging sheet. These findings suggest ordered conformational changes before coreceptor binding: (1) gp120 opening inducing side-chain rearrangements and a compact gp41 central helix conformation, and (2) 4-stranded bridging-sheet formation and V1V2 displacement. These analyses illuminate potential receptor-induced Env changes and inform design of therapeutics disrupting viral entry.

HIV-1 Env, a trimer of gp120-gp41 heterodimers, mediates membrane fusion after binding host receptor CD4. Receptor binding displaces V1V2 loops from Env's apex, allowing coreceptor binding and opening Env to enable gp41-mediated fusion. We present 3.54 Å and 4.06 Å cryoelectron microscopy structures of partially open soluble native-like Env trimers (SOSIPs) bound to CD4. One structure, a complex with a coreceptor-mimicking antibody that binds both CD4 and gp120, stabilizes the displaced V1V2 and reveals its structure. Comparing partially and fully open Envs with closed Envs shows that gp41 rearrangements are independent of the CD4-induced rearrangements that result in V1V2 displacement and formation of a 4-stranded bridging sheet. These findings suggest ordered conformational changes before coreceptor binding: (1) gp120 opening inducing side-chain rearrangements and a compact gp41 central helix conformation, and (2) 4-stranded bridging-sheet formation and V1V2 displacement. These analyses illuminate potential receptor-induced Env changes and inform design of therapeutics disrupting viral entry.

Membrane proteins are widely studied in detergent micelles, a inembranc-mimetic system formed by amphiphilic compounds. However, classical detergents have serious limitations in their utility, particularly for unstable proteins such as eukaryotic membrane proteins and membrane protein complexes, and thus, there is an unmet need for novel amphiphiles with enhanced ability to stabilize membrane proteins. Here, we developed a new class of malonate-derived detergents with four glucosides, designated malonate-derived tetra-glucosides (MTGs), and compared these new detergents with previously reported octyl glucose neopentyl glycol (OGNG) and n-dodecyl-beta-D-maltoside (DDM). When tested with two G-protein coupled receptors (GPCRs) and three transporters, a couple of MTGs consistently conferred enhanced stability to all tested proteins compared to DDM and OGNG. As a result of favorable behaviors for a range of membrane proteins, these MTGs have substantial potential for membrane protein research. This study additionally provides a new detergent design principle based on the effect of a polar functional group (i.e., ether) on protein stability depending on its position in the detergent scaffold.

The performance of meta-learning approaches for few-shot learning generally depends on three aspects: features suitable for comparison, the classifier ( base learner ) suitable for low-data scenarios, and valuable information from the samples to classify. In this work, we make improvements for the last two aspects: 1) although there are many effective base learners, there is a trade-off between generalization performance and computational overhead, so we introduce multi-class least squares support vector machine as our base learner which obtains better generation than existing ones with less computational overhead; 2) further, in order to utilize the information from the query samples, we propose two simple and effective transductive modules which modify the support set using the query samples, i.e., adjusting the support samples basing on the attention mechanism and adding the prototypes of the query set with pseudo labels to the support set as the pseudo support samples. These two modules significantly improve the few-shot classification accuracy, especially for the difficult 1-shot setting. Our model, denoted as FSLSTM (Few-Shot learning with LSsvm base learner and Transductive Modules), achieves state-of-the-art performance on miniImageNet and CIFAR-FS few-shot learning benchmarks., Comment: 9 pages,3 figures,3 tables

The HIV-1 envelope (Env) glycoprotein binds to host cell receptors to mediate membrane fusion. The prefusion Env trimer is stabilized by V1V2 loops that interact at the trimer apex. Broadly neutralizing antibodies (bNAbs) against V1V2 loops, exemplified by PG9, bind asymmetrically as a single Fab to the apex of the symmetric Env trimer using a protruding CDRH3 to penetrate the Env glycan shield. Here we characterized a distinct mode of V1V2 epitope recognition by the new bNAb BG1 in which two Fabs bind asymmetrically per Env trimer using a compact CDRH3. Comparisons between cryo-EM structures of Env trimer complexed with BG1 (6.2 Å resolution) and PG9 (11.5 Å resolution) revealed a new V1V2-targeting strategy by BG1. Analyses of the EM structures provided information relevant to vaccine design including molecular details for different modes of asymmetric recognition of Env trimer and a binding model for BG1 recognition of V1V2 involving glycan flexibility.

The HIV-1 envelope (Env) glycoprotein binds to host cell receptors to mediate membrane fusion. The prefusion Env trimer is stabilized by V1V2 loops that interact at the trimer apex. Broadly neutralizing antibodies (bNAbs) against V1V2 loops, exemplified by PG9, bind asymmetrically as a single Fab to the apex of the symmetric Env trimer using a protruding CDRH3 to penetrate the Env glycan shield. Here we characterized a distinct mode of V1V2 epitope recognition by the new bNAb BG1 in which two Fabs bind asymmetrically per Env trimer using a compact CDRH3. Comparisons between cryo-EM structures of Env trimer complexed with BG1 (6.2 Å resolution) and PG9 (11.5 Å resolution) revealed a new V1V2-targeting strategy by BG1. Analyses of the EM structures provided information relevant to vaccine design including molecular details for different modes of asymmetric recognition of Env trimer and a binding model for BG1 recognition of V1V2 involving glycan flexibility.

Broadly neutralizing monoclonal antibodies protect against infection with HIV-1 in animal models, suggesting that a vaccine that elicits these antibodies would be protective in humans. However, it has not yet been possible to induce adequate serological responses by vaccination. Here, to activate B cells that express precursors of broadly neutralizing antibodies within polyclonal repertoires, we developed an immunogen, RC1, that facilitates the recognition of the variable loop 3 (V3)-glycan patch on the envelope protein of HIV-1. RC1 conceals non-conserved immunodominant regions by the addition of glycans and/or multimerization on virus-like particles. Immunization of mice, rabbits and rhesus macaques with RC1 elicited serological responses that targeted the V3-glycan patch. Antibody cloning and cryo-electron microscopy structures of antibody-envelope complexes confirmed that immunization with RC1 expands clones of B cells that carry the anti-V3-glycan patch antibodies, which resemble precursors of human broadly neutralizing antibodies. Thus, RC1 may be a suitable priming immunogen for sequential vaccination strategies in the context of polyclonal repertoires.

Broadly neutralizing monoclonal antibodies protect against infection with HIV-1 in animal models, suggesting that a vaccine that elicits these antibodies would be protective in humans. However, it has not yet been possible to induce adequate serological responses by vaccination. Here, to activate B cells that express precursors of broadly neutralizing antibodies within polyclonal repertoires, we developed an immunogen, RC1, that facilitates the recognition of the variable loop 3 (V3)-glycan patch on the envelope protein of HIV-1. RC1 conceals non-conserved immunodominant regions by the addition of glycans and/or multimerization on virus-like particles. Immunization of mice, rabbits and rhesus macaques with RC1 elicited serological responses that targeted the V3-glycan patch. Antibody cloning and cryo-electron microscopy structures of antibody-envelope complexes confirmed that immunization with RC1 expands clones of B cells that carry the anti-V3-glycan patch antibodies, which resemble precursors of human broadly neutralizing antibodies. Thus, RC1 may be a suitable priming immunogen for sequential vaccination strategies in the context of polyclonal repertoires.

Autoregressive sequence models achieve state-of-the-art performance in domains like machine translation. However, due to the autoregressive factorization nature, these models suffer from heavy latency during inference. Recently, non-autoregressive sequence models were proposed to reduce the inference time. However, these models assume that the decoding process of each token is conditionally independent of others. Such a generation process sometimes makes the output sentence inconsistent, and thus the learned non-autoregressive models could only achieve inferior accuracy compared to their autoregressive counterparts. To improve then decoding consistency and reduce the inference cost at the same time, we propose to incorporate a structured inference module into the non-autoregressive models. Specifically, we design an efficient approximation for Conditional Random Fields (CRF) for non-autoregressive sequence models, and further propose a dynamic transition technique to model positional contexts in the CRF. Experiments in machine translation show that while increasing little latency (8~14ms), our model could achieve significantly better translation performance than previous non-autoregressive models on different translation datasets. In particular, for the WMT14 En-De dataset, our model obtains a BLEU score of 26.80, which largely outperforms the previous non-autoregressive baselines and is only 0.61 lower in BLEU than purely autoregressive models., Comment: Accepted to NeurIPS 2019 (Previous title: Structured Decoding for Non-Autoregressive Machine Translation)

The HIV-1 envelope (Env) glycoprotein, a trimer of gp120-gp41 heterodimers, relies on conformational flexibility to function in fusing the viral and host membranes. Fusion is achieved after gp120 binds to CD4, the HIV-1 receptor, and a coreceptor, capturing an open conformational state in which the fusion machinery on gp41 gains access to the target cell membrane. In the well-characterized closed Env conformation, the gp120 V1V2 loops interact at the apex of the Env trimer. Less is known about the structure of the open CD4-bound state, in which the V1V2 loops must rearrange and separate to allow access to the coreceptor binding site. We identified two anti-HIV-1 antibodies, the coreceptor mimicking antibody 17b and the gp120-gp41 interface-spanning antibody 8ANC195, that can be added as Fabs to a soluble native-like Env trimer to stabilize it in a CD4-bound conformation. Here, we present an 8.9-Å cryo-electron microscopy structure of a BG505 Env-sCD4-17b-8ANC195 complex, which reveals large structural rearrangements in gp120, but small changes in gp41, compared with closed Env structures. The gp120 protomers are rotated and separated in the CD4-bound structure, and the three V1V2 loops are displaced by ∼40 Å from their positions at the trimer apex in closed Env to the sides of the trimer in positions adjacent to, and interacting with, the three bound CD4s. These results are relevant to understanding CD4-induced conformational changes leading to coreceptor binding and fusion, and HIV-1 Env conformational dynamics, and describe a target structure relevant to drug design and vaccine efforts.

The HIV-1 envelope (Env) glycoprotein, a trimer of gp120-gp41 heterodimers, relies on conformational flexibility to function in fusing the viral and host membranes. Fusion is achieved after gp120 binds to CD4, the HIV-1 receptor, and a coreceptor, capturing an open conformational state in which the fusion machinery on gp41 gains access to the target cell membrane. In the well-characterized closed Env conformation, the gp120 V1V2 loops interact at the apex of the Env trimer. Less is known about the structure of the open CD4-bound state, in which the V1V2 loops must rearrange and separate to allow access to the coreceptor binding site. We identified two anti-HIV-1 antibodies, the coreceptor mimicking antibody 17b and the gp120-gp41 interface-spanning antibody 8ANC195, that can be added as Fabs to a soluble native-like Env trimer to stabilize it in a CD4-bound conformation. Here, we present an 8.9-Å cryo-electron microscopy structure of a BG505 Env-sCD4-17b-8ANC195 complex, which reveals large structural rearrangements in gp120, but small changes in gp41, compared with closed Env structures. The gp120 protomers are rotated and separated in the CD4-bound structure, and the three V1V2 loops are displaced by ∼40 Å from their positions at the trimer apex in closed Env to the sides of the trimer in positions adjacent to, and interacting with, the three bound CD4s. These results are relevant to understanding CD4-induced conformational changes leading to coreceptor binding and fusion, and HIV-1 Env conformational dynamics, and describe a target structure relevant to drug design and vaccine efforts.

We design and implement two Java optimizations based on JSR308 annota- tions: pure method memoization and dynamic taint analysis optimization. In pure method memoization, we introduce JavaMem, an automatic method memo- ization tool for Java. JavaMem memoizes on Java pure or partially pure method at runtime according to offline execution time profiling results and static analysis- based correctness proof. We achieve 3% speedup on average on selected Dacapo benchmarks. For dynamic taint analysis optimization, we design and implement METEOR, a portable dynamic taint checking tool that instruments taint propaga- tion instructions and optimizes the instrumented code by leveraging annotations. We compete our METEOR against Phosphor. With a slightly slower baseline performance on non-annotated code, METEOR still beats Phosphor on fully an- notated benchmarks.

We design and implement two Java optimizations based on JSR308 annota- tions: pure method memoization and dynamic taint analysis optimization. In pure method memoization, we introduce JavaMem, an automatic method memo- ization tool for Java. JavaMem memoizes on Java pure or partially pure method at runtime according to offline execution time profiling results and static analysis- based correctness proof. We achieve 3% speedup on average on selected Dacapo benchmarks. For dynamic taint analysis optimization, we design and implement METEOR, a portable dynamic taint checking tool that instruments taint propaga- tion instructions and optimizes the instrumented code by leveraging annotations. We compete our METEOR against Phosphor. With a slightly slower baseline performance on non-annotated code, METEOR still beats Phosphor on fully an- notated benchmarks.

Some HIV-1-infected patients develop broad and potent HIV-1 neutralizing antibodies (bNAbs) that when passively transferred to mice or macaques can treat or prevent infection. However, bNAbs typically fail to neutralize coexisting autologous viruses due to antibody-mediated selection against sensitive viral strains. We describe an HIV-1 controller expressing HLA-B57*01 and HLA-B27*05 who maintained low viral loads for 30 years after infection and developed broad and potent serologic activity against HIV-1. Neutralization was attributed to three different bNAbs targeting nonoverlapping sites on the HIV-1 envelope trimer (Env). One of the three, BG18, an antibody directed against the glycan-V3 portion of Env, is the most potent member of this class reported to date and, as revealed by crystallography and electron microscopy, recognizes HIV-1 Env in a manner that is distinct from other bNAbs in this class. Single-genome sequencing of HIV-1 from serum samples obtained over a period of 9 years showed a diverse group of circulating viruses, 88.5% (31 of 35) of which remained sensitive to at least one of the temporally coincident autologous bNAbs and the individual's serum. Thus, bNAb-sensitive strains of HIV-1 coexist with potent neutralizing antibodies that target the virus and may contribute to control in this individual. When administered as a mix, the three bNAbs controlled viremia in HIV-1YU2-infected humanized mice. Our finding suggests that combinations of bNAbs may contribute to control of HIV-1 infection.

Some HIV-1-infected patients develop broad and potent HIV-1 neutralizing antibodies (bNAbs) that when passively transferred to mice or macaques can treat or prevent infection. However, bNAbs typically fail to neutralize coexisting autologous viruses due to antibody-mediated selection against sensitive viral strains. We describe an HIV-1 controller expressing HLA-B57*01 and HLA-B27*05 who maintained low viral loads for 30 years after infection and developed broad and potent serologic activity against HIV-1. Neutralization was attributed to three different bNAbs targeting nonoverlapping sites on the HIV-1 envelope trimer (Env). One of the three, BG18, an antibody directed against the glycan-V3 portion of Env, is the most potent member of this class reported to date and, as revealed by crystallography and electron microscopy, recognizes HIV-1 Env in a manner that is distinct from other bNAbs in this class. Single-genome sequencing of HIV-1 from serum samples obtained over a period of 9 years showed a diverse group of circulating viruses, 88.5% (31 of 35) of which remained sensitive to at least one of the temporally coincident autologous bNAbs and the individual's serum. Thus, bNAb-sensitive strains of HIV-1 coexist with potent neutralizing antibodies that target the virus and may contribute to control in this individual. When administered as a mix, the three bNAbs controlled viremia in HIV-1YU2-infected humanized mice. Our finding suggests that combinations of bNAbs may contribute to control of HIV-1 infection.

The HIV-1 envelope (Env) spike contains limited epitopes for broadly neutralizing antibodies (bNAbs); thus, most neutralizing antibodies are strain specific. The 8ANC195 epitope, defined by crystal and electron microscopy (EM) structures of bNAb 8ANC195 complexed with monomeric gp120 and trimeric Env, respectively, spans the gp120 and gp41 Env subunits. To investigate 8ANC195's gp41 epitope at higher resolution, we solved a 3.58 Å crystal structure of 8ANC195 complexed with fully glycosylated Env trimer, revealing 8ANC195 insertion into a glycan shield gap to contact gp120 and gp41 glycans and protein residues. To determine whether 8ANC195 recognizes the CD4-bound open Env conformation that leads to co-receptor binding and fusion, one of several known conformations of virion-associated Env, we solved EM structures of an Env/CD4/CD4-induced antibody/8ANC195 complex. 8ANC195 binding partially closed the CD4-bound trimer, confirming structural plasticity of Env by revealing a previously unseen conformation. 8ANC195's ability to bind different Env conformations suggests advantages for potential therapeutic applications.

The HIV-1 envelope (Env) spike contains limited epitopes for broadly neutralizing antibodies (bNAbs); thus, most neutralizing antibodies are strain specific. The 8ANC195 epitope, defined by crystal and electron microscopy (EM) structures of bNAb 8ANC195 complexed with monomeric gp120 and trimeric Env, respectively, spans the gp120 and gp41 Env subunits. To investigate 8ANC195's gp41 epitope at higher resolution, we solved a 3.58 Å crystal structure of 8ANC195 complexed with fully glycosylated Env trimer, revealing 8ANC195 insertion into a glycan shield gap to contact gp120 and gp41 glycans and protein residues. To determine whether 8ANC195 recognizes the CD4-bound open Env conformation that leads to co-receptor binding and fusion, one of several known conformations of virion-associated Env, we solved EM structures of an Env/CD4/CD4-induced antibody/8ANC195 complex. 8ANC195 binding partially closed the CD4-bound trimer, confirming structural plasticity of Env by revealing a previously unseen conformation. 8ANC195's ability to bind different Env conformations suggests advantages for potential therapeutic applications.