3DDF-VAE: Dual-frequency variational autoencoder with pose-consistency validation for rare cryo-EM conformation discovery.

Regions of rare conformation were located in 300 protein crystal structures representing seven major protein folds. A distance matrix algorithm was used to search rapidly for 9-residue fragments of rare backbone conformation using a comparison to a relational database of encoded fragments derived from the database of nonredundant structures. Rare fragments were found in 61% of the analyzed protein structures. Detailed analysis was performed for 78 proteins of different folds. The rare fragments were located near functional sites in 72% of the protein structures. The rare fragments often formed parts of ligand-binding sites (59%), protein-protein interfaces (8%), and domain-domain contacts (5%). Of the remaining structures, 5% had a high average B-factor or high local B-factors. Statistical analysis suggests that the association between ligands and rare regions does not occur by chance alone. The present study is likely to underestimate the number of functional sites, because not all analyzed protein structures contained a ligand. The results suggest that rapid searches for regions with rare local backbone conformations can assist in prediction of functional sites in novel proteins.

Rare and short-lived DNA conformations are proposed to be key drivers of mutagenesis, yet assessing their contribution to mutational signatures found in human cancers remains challenging. Here, we developed an approach that quantifies the sequence-dependent propensity to form a rare DNA conformation and compares the resulting fingerprint against cancer mutational signatures. Using 19F NMR, we measured the propensity for the anionic Watson-Crick-like G•T− conformation across all sixteen triplet sequence contexts and discovered a striking 50-fold variation driven by suboptimal interactions between anionic thymine and its 3’ neighbor. Comparing this fingerprint, and those of other rare DNA states, against the COSMIC database uncovered plausible links to mutational processes associated with exposure to damaging agents and therapies. Thus, integrating molecular biophysics with genomic epidemiology provides a powerful framework to explore how DNA’s dynamic properties shape genome stability and influence human disease.

Rare and short-lived DNA conformations are proposed to be key drivers of mutagenesis, yet assessing their contribution to mutational signatures found in human cancers remains challenging. Here, we develop an approach that quantifies the sequence-dependent propensity to form a rare DNA conformation and compare the resulting fingerprint against cancer mutational signatures. Using 19F NMR, we measure the propensity for the anionic Watson-Crick-like G•T- conformation across all sixteen triplet sequence contexts and discover a striking 50-fold variation driven by suboptimal interactions between anionic thymine and its 3' neighbor. Comparing this fingerprint, and those of other rare DNA states against the Catalogue of Somatic Mutations in Cancer (COSMIC) database uncovers plausible links to mutational processes associated with exposure to damaging agents and therapies. Thus, integrating molecular biophysics with genomic epidemiology provides a powerful framework to explore how DNA's dynamic properties shape genome stability and influence human disease.

N-ethyl maleimide sensitive factor (NSF) belongs to the AAA family of ATPases and is involved in a number of cellular functions, including vesicle fusion and trafficking of membrane proteins. We present the three-dimensional structure of the hydrolysis mutant E329Q of NSF complexed with an ATP-ADP mixture at 11 A resolution by electron cryomicroscopy and single-particle averaging of NSF.alpha-SNAP.SNARE complexes. The NSF domains D1 and D2 form hexameric rings that are arranged in a double-layered barrel. Our structure is more consistent with an antiparallel orientation of the two rings rather than a parallel one. The crystal structure of the D2 domain of NSF was docked into the EM density map and shows good agreement, including details at the secondary structural level. Six protrusions corresponding to the N domain of NSF (NSF-N) emerge from the sides of the D1 domain ring. The density corresponding to alpha-SNAP and SNAREs is located on the 6-fold axis of the structure, near the NSF-N domains. The density of the N domain is weak, suggesting conformational variability in this part of NSF.

Cryogenic electron microscopy (cryo-EM) has revolutionized structural biology, enabling efficient determination of structures at near-atomic resolutions. However, a common challenge arises from the severe imbalance among various conformations of vitrified particles, leading to low-resolution reconstructions in rare conformations due to a lack of particle images in these quasi-stable states. We introduce CryoTRANS, a method that predicts high-resolution maps of rare conformations by constructing a self-supervised pseudo-trajectory between density maps of varying resolutions. This trajectory is represented by an ordinary differential equation parameterized by a deep neural network, ensuring retention of detailed structures from high-resolution density maps. By leveraging a single high-resolution density map, CryoTRANS significantly improves the reconstruction of rare conformations and has been validated on four real-world datasets: alpha-2-macroglobulin, actin-binding protein complexes, SARS-CoV-2 spike glycoprotein, and the 70S ribosome. CryoTRANS can also predict high-resolution structures in cryogenic electron tomography maps using a high-resolution cryo-EM map.

Crowd counting is the basic task of crowd analysis and it is of great significance in the field of public safety. Therefore, it receives more and more attention recently. The common idea is to combine the crowd counting task with convolutional neural networks to predict the corresponding density map, which is generated by filtering the dot labels with specific Gaussian kernels. Although the counting performance is promoted by the newly proposed networks, they all suffer one conjunct problem, which is due to the perspective effect, there is significant scale contrast among targets in different positions within one scene, but the existing density maps can not represent this scale change well. To address the prediction difficulties caused by target scale variation, we propose a scale-sensitive crowd density map estimation framework, which focuses on dealing with target scale change from density map generation, network design, and model training stage. It consists of the Adaptive Density Map (ADM), Deformable Density Map Decoder (DDMD), and Auxiliary Branch. To be specific, the Gaussian kernel size variates adaptively based on target size to generate ADM that contains scale information for each specific target. DDMD introduces the deformable convolution to fit the Gaussian kernel variation and boosts the model's scale sensitivity. The Auxiliary Branch guides the learning of deformable convolution offsets during the training phase. Finally, we construct experiments on different large-scale datasets. The results show the effectiveness of the proposed ADM and DDMD. Furthermore, the visualization demonstrates that deformable convolution learns the target scale variation.

We articulate a novel approach to geometric model completion via interactive sketches in this paper. First, the initial incomplete model (with holes) is decomposed into a base model and a high-frequency component, which represents global rough shape and geometric details, respectively. We then repair the base model via smooth hole-filling, and compute the geometry detail image using high frequency information. One novel element of our approach is that we allow users to interactively sketch a few structural curves that span across hole regions, with a goal to repair both local geometric details and global structure. With the help of local parameterization, we convert detailed geometry into gradient-domain images which can propagate along user-specified sketches. By integrating recovered gradient-domain images and base shape, we can generate a complete model that faithfully recovers both global structure and local details. The salient contribution of this paper is the unified approach for user interaction, global structure, and geometry details towards high-fidelity model completion. We demonstrate our new approach using a number of examples that exhibit salient global structure as well as local geometry details.

The fatigue analysis of civil structures in single-scale is difficult to capture the nonlinear process of structural response and accumulative fatigue damage on structural details up to failure since the process is trans-scale process of damage evolution, so that it is necessary to analyzing fatigue damage with concurrent multi-scale computation. This paper is aimed to develop a concurrent multi-scale computational method for carrying out fatigue damage analyses of civil structures with local details that are vulnerable to damage and failure. The multi-scale model is developed on the basis of the different features in macro- and meso- scales of structural response and damage behavior, and the concurrent trans-scale computation is implemented by information transferred between scales through the multi-points constraints method. The linear response of global structure and the nonlinear low-cycle fatigue damage behavior at local details are analyzed concurrently in order to meet the needs of evaluation of damage status as well as structural deteriorating. Numerical results on the damage evolution process of a steel truss structure show that, the proposed concurrent multi-scale computational method on structural fatigue damage could simultaneously carry out the analyses both on the linear response of global structure and on the nonlinear low-cycle fatigue damage evolution at local vulnerable details. Besides the accurate description on the structural multi-scale response, the local damage evolution as well as its influence on the structural response could also be obtained.

The extracellular matrix (ECM) molecules play important roles in many biological and pathological processes. During tissue remodeling, the ECM molecules that are glycosylated are different from those of normal tissue owing to changes in the expression of many proteins that are responsible for glycan synthesis. Vitronectin (VN) is a major ECM molecule that recognizes integrin on hepatic stellate cells (HSCs). The present study attempted to elucidate how changes in VN glycans modulate the survival of HSCs, which play a critical role in liver regeneration. Plasma VN was purified from partially hepatectomized (PH) and sham-operated (SH) rats at 24 h after operation and non-operated (NO) rats. Adhesion of rat HSCs (rHSCs), together with phosphorylation of focal adhesion kinase, in PH-VN was decreased to one-half of that in NO- or SH-VN. Spreading of rHSCs on desialylated NO-VN was decreased to one-half of that of control VN, indicating the importance of sialylation of VN for activation of HSCs. Liquid chromatography/multiple-stage mass spectrometry analysis of Glu-C glycopeptides of each VN determined the site-specific glycosylation. In addition to the major biantennary complex-type N-glycans, hybrid-type N-glycans were site-specifically present at Asn(167). Highly sialylated O-glycans were found to be present in the Thr(110)-Thr(124) region. In PH-VN, the disialyl O-glycans and complex-type N-glycans were decreased while core-fucosylated N-glycans were increased. In addition, immunodetection after two-dimensional PAGE indicated the presence of hyper- and hyposialylated molecules in each VN and showed that hypersialylation was markedly attenuated in PH-VN. This study proposes that the alteration of VN glycosylation modulates the substrate adhesion to rat HSCs, which is responsible for matrix restructuring.

Neural Networks (NNs) are increasingly used across scientific domains to extract knowledge from experimental or computational data. An NN is composed of natural or artificial neurons that serve as simple processing units and are interconnected into a model architecture; it acquires knowledge from the environment through a learning process and stores this knowledge in its connections. The learning process is conducted by training. During NN training, the learning process can be tracked by periodically validating the NN and calculating its fitness. The resulting sequence of fitness values (i.e., validation accuracy or validation loss) is called the NN learning curve. The development of tools for NN design requires knowledge of diverse NNs and their complete learning curves.Generally, only final fully-trained fitness values for highly accurate NNs are made available to the community, hampering efforts to develop tools for NN design and leaving unaddressed aspects such as explaining the generation of an NN and reproducing its learning process. Our dataset fills this gap by fully recording the structure, metadata, and complete learning curves for a wide variety of random NNs throughout their training. Our dataset captures the lifespan of 6000 NNs throughout generation, training, and validation stages. It consists of a suite of 6000 tables, each table representing the lifespan of one NN. We generate each NN with randomized parameter values and train it for 40 epochs on one of three diverse image datasets (i.e., CIFAR-100, FashionMNIST, SVHN). We calculate and record each NN’s fitness with high frequency—every half epoch—to capture the evolution of the training and validation process. As a result, for each NN, we record the generated parameter values describing the structure of that NN, the image dataset on which the NN trained, and all loss and accuracy values for the NN every half epoch.We put our dataset to the service of researchers studying NN performance and its evolution throughout training and validation. Statistical methods can be applied to our dataset to analyze the shape of learning curves in diverse NNs, and the relationship between an NN’s structure and its fitness. Additionally, the structural data and metadata that we record enable the reconstruction and reproducibility of the associated NN.

For crowd counting, the existing methods usually use an end-to-end approach to directly output the final estimated density map and perform the counts. However, as an intermediate representation, the quality of the estimated density map may significantly affect the counting performance. Therefore, some studies have attempted to optimize the estimated density map with additional attention mechanism. But these methods only focus on the high-density crowd areas and ignore the optimization of local detail areas. Consequently, we propose a more intuitive and understandable Density Map Dynamic Refinement Network (DDRNet) consisting of Counter and Refiner to further refine the local detail information of the estimated density map. Our training contains two stages. Specifically, for the first stage, Counter generates the initial density map through the feature extraction module and the backend, while Refiner, which consists of convolutional layers with different dilated rates, further refines the output of the former to obtain the final estimated density map in the second stage. Also, due to the different views of Counter and Refiner during training, we design a dynamic joint training strategy to improve counting performance. Extensive experiments on three crowd counting datasets (ShanghaiTech, UCF_CC_50, UCF-QNRF) demonstrate the effectiveness of the proposed model and achieve superior counting results.

Article Figures and data Abstract eLife digest Introduction Results Discussion Materials and methods Appendix 1 Appendix 2 Data availability References Decision letter Author response Article and author information Metrics Abstract Transcription factor IIH (TFIIH) is a heterodecameric protein complex critical for transcription initiation by RNA polymerase II and nucleotide excision DNA repair. The TFIIH core complex is sufficient for its repair functions and harbors the XPB and XPD DNA-dependent ATPase/helicase subunits, which are affected by human disease mutations. Transcription initiation additionally requires the CdK activating kinase subcomplex. Previous structural work has provided only partial insight into the architecture of TFIIH and its interactions within transcription pre-initiation complexes. Here, we present the complete structure of the human TFIIH core complex, determined by phase-plate cryo-electron microscopy at 3.7 Å resolution. The structure uncovers the molecular basis of TFIIH assembly, revealing how the recruitment of XPB by p52 depends on a pseudo-symmetric dimer of homologous domains in these two proteins. The structure also suggests a function for p62 in the regulation of XPD, and allows the mapping of previously unresolved human disease mutations. https://doi.org/10.7554/eLife.44771.001 eLife digest The DNA inside a cell carries the instructions it needs to survive. Living cells use many different proteins to read and maintain this store of information. For example, a group of ten proteins collectively called TFIIH is often involved in both reading and repairing the DNA. Proteins in the TFIIH complex include p52, p62, XPB and XPD. Understanding the structure of the proteins in TFIIH could reveal much about how it works and how changes to its structure contribute to various medical conditions. Yet TFIIH is a dynamic assembly of molecules and includes many proteins, which makes examining its structure challenging. An ideal protein structure should provide an accurate map of the positions of all the atoms in a protein. Previously, it has not been possible to get this level of detail for TFIIH. Greber et al. used an approach called cryo-electron microscopy (also called cryo-EM) to reveal the structure of TFIIH collected from human cells. The structure revealed several new details, including how p52 helps XPB attach to the rest of TFIIH, and that p62 helps to control the activity of XPD. With such a detailed structure, Greber et al. could link changes in TFIIH that are seen in different human diseases to specific parts of the complex. Examining the atomic details of proteins can reveal a lot about how they work and the changes that occur during different diseases. These structures can also help to reveal aspects of how DNA is read and repaired, and may help to design new approaches to treat diseases in the future. https://doi.org/10.7554/eLife.44771.002 Introduction Transcription factor IIH (TFIIH) is a 10-subunit protein complex with a total molecular weight of 0.5 MDa that serves a dual role as a general transcription factor for transcription initiation by eukaryotic RNA polymerase II (Pol II), and as a DNA helicase complex in nucleotide excision DNA repair (NER) (Compe and Egly, 2016; Sainsbury et al., 2015). Mutations in TFIIH subunits that cause the inherited autosomal recessive disorders xeroderma pigmentosum (XP), trichothiodystrophy (TTD), and Cockayne syndrome (CS) are characterized by high incidence of cancer or premature ageing (Cleaver et al., 1999; Rapin, 2013). Furthermore, TFIIH is a possible target for anti-cancer compounds (Berico and Coin, 2018) and therefore of great importance for human health and disease. The TFIIH core complex is composed of the seven subunits XPB, XPD, p62, p52, p44, p34, and p8, and is the form of TFIIH active in DNA repair (Svejstrup et al., 1995), where TFIIH serves as a DNA damage verification factor (Li et al., 2015; Mathieu et al., 2013) and is responsible for opening a repair bubble around damaged nucleotides. This activity depends on both the SF2-family DNA-dependent ATPase XPB, and the DNA helicase activity of XPD (Coin et al., 2007; Evans et al., 1997; Kuper et al., 2014). TFIIH function in transcription initiation requires the double-stranded DNA translocase activity of XPB to regulate opening of the transcription bubble (Alekseev et al., 2017; Fishburn et al., 2015; Grünberg et al., 2012), and additionally the CdK activating kinase (CAK) complex, which harbors the kinase activity of CDK7 as well as the Cyclin H and MAT1 subunits (Devault et al., 1995; Fisher et al., 1995; Fisher and Morgan, 1994; Shiekhattar et al., 1995; Svejstrup et al., 1995). Targets of human CDK7 include the C-terminal heptapeptide repeat domain of the largest subunit of Pol II, as well as cell-cycle regulating CDKs (Fisher and Morgan, 1994; Shiekhattar et al., 1995). MAT1 serves as a bridging subunit that promotes CAK subcomplex formation by interacting with Cyclin H and CDK7 (Devault et al., 1995; Fisher et al., 1995), recruits the CAK to the core complex by interactions with XPD and XPB (Abdulrahman et al., 2013; Busso et al., 2000; Greber et al., 2017; Rossignol et al., 1997), and also aids in Pol II-PIC formation by establishing interactions with the core PIC (He et al., 2013; He et al., 2016; Schilbach et al., 2017). The presence of MAT1 inhibits the helicase activity of XPD (Abdulrahman et al., 2013; Sandrock and Egly, 2001), but the mechanism of this inhibition is not fully understood. While the enzymatic activity of XPD is not required for transcription initiation, it is critical for the DNA repair function of TFIIH (Dubaele et al., 2003; Evans et al., 1997; Kuper et al., 2014). Therefore, NER requires the release of the CAK subcomplex from the core complex (Coin et al., 2008). The activities of both XPB and XPD are regulated by interactions with additional TFIIH components, including that of p44 with XPD (Coin et al., 1998; Dubaele et al., 2003; Kim et al., 2015), and those of the p52-p8 module with XPB (Coin et al., 2007; Coin et al., 2006; Jawhari et al., 2002; Kainov et al., 2008). These interactions are likely to be crucial for TFIIH function, as some are affected by disease mutations (Cleaver et al., 1999), but they have been only partially characterized mechanistically. Our previous structure of the TFIIH core-MAT1 complex at 4.4 Å resolution (Greber et al., 2017) allowed modeling of TFIIH in the best-resolved parts of the density map, but several functionally important regions remained unassigned or only partially interpreted because reliable de novo tracing of entire domains in the absence of existing structural models was not possible. Here, we present the complete structure of the human TFIIH core complex in association with the CAK subunit MAT1, determined by phase plate cryo-electron microscopy (cryo-EM) at 3.7 Å resolution. Our structure reveals the complete architecture of the TFIIH core complex and provides detailed insight into the interactions that govern its assembly. Additionally, our cryo-EM maps define the molecular contacts that control the regulation of the XPB and XPD subunits of TFIIH, including the critical p52-XBP interaction, and an extensive regulatory network around XPD, formed by XPB, p62, p44, and MAT1. Results Structure determination of TFIIH To determine the complete structure of the human TFIIH core complex, we collected several large cryo-EM datasets (Supplementary file 1) of TFIIH immuno-purified from HeLa cells using an electron microscope equipped with a Volta phase plate (VPP) (Danev and Baumeister, 2017) and a direct electron detector camera mounted behind an energy filter. From a homogeneous subset of approximately 140,000 TFIIH particle images identified by 3D classification (Scheres, 2010), we reconstructed a 3D cryo-EM density map at 3.7 Å resolution (Figure 1—figure supplements 1 and 2A–C). This VPP-based cryo-EM map was substantially improved compared to our previous maps obtained without phase plate, both in resolution and interpretability (Figure 1—figure supplement 2D–G), and enabled building, refinement, and full validation of an atomic model of the TFIIH core complex and the MAT1 subunit of the CAK subcomplex (Figure 1A–C, Figure 1—figure supplement 2B,C, Supplementary file 2, 3), while the remainder of the CAK subcomplex is invisible in our map because it is flexibly tethered to the TFIIH core complex. Tracing and sequence register assignment of protein components modeled de novo was facilitated by density maps obtained from focused classification and multibody refinement (Figure 1—figure supplements 3–5) (Bai et al., 2015; Nakane et al., 2018), which resulted in density maps of improved interpretability for all three sub-volumes and a slightly improved resolution of 3.6 Å for the XPD-MAT1 region. Both the overall and multibody-refined maps showed clear side chain information (Figure 1—figure supplement 6A–D). Furthermore, our model was corroborated by existing chemical crosslinking-mass spectrometry (CX-MS) data of human TFIIH (Luo et al., 2015) and site-specific crosslinks from yeast TFIIH (Warfield et al., 2016) (Figure 1—figure supplement 6E–I, Supplementary file 4). Figure 1 with 6 supplements see all Download asset Open asset Structure of the TFIIH core complex. (A, B, C) Three views of the structure of the TFIIH core complex and MAT1. Subunits are color-coded and labeled (in color); individual domains are labeled (in black) and circled if needed for clarity. (D) Domain-level protein-protein interaction network between the components of the TFIIH core complex and MAT1 derived from the interactions observed in our structure. Proteins are shown with the same colors as in A and major unmodeled regions are shown in grey. Abbreviations: CTD: C-terminal domain; DRD: DNA damage recognition domain; FeS: iron sulfur cluster domain; NTD: N-terminal domain; vWFA: von Willebrand Factor A. https://doi.org/10.7554/eLife.44771.003 Detailed architecture of TFIIH and structure of p62 Our structure of the TFIIH core complex shows its horseshoe-like overall shape (Figure 1A–C, Video 1), as observed in previous lower-resolution reconstructions of free and PIC-bound TFIIH (Gibbons et al., 2012; Greber et al., 2017; He et al., 2016; Murakami et al., 2015; Schilbach et al., 2017), and allows us to define the complete set of inter-subunit interactions that lead to the formation of the TFIIH core complex directly from our structure (Figure 1D). Video 1 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Architecture of the TFIIH core complex. Rotating structure of the TFIIH core complex, followed by views that highlight the interactions of p62 near the nucleotide binding pocket of XPD and near the substrate binding cleft of XPD (binding sites are indicated by a flashing ADP molecule and DNA strand, respectively). Bound substrates, which are not present in our structure, were superposed from PDB ID 6FWS (Cheng and Wigley, 2018). https://doi.org/10.7554/eLife.44771.010 The largest subunits of the complex, the SF2-family DNA-dependent ATPases XPB and XPD, both containing two RecA-like domains (RecA1 and RecA2), interact directly (Greber et al., 2017), are on one side of the complex, and are additionally bridged by MAT1 (Figure 1B), which has been shown to interact with either ATPase in isolation (Busso et al., 2000). On the side facing away from MAT1, XPD interacts with the von Willebrand Factor A (vWFA) domain of p44 (Coin et al., 1998; Dubaele et al., 2003; He et al., 2016; Kim et al., 2015), which in turn forms a tight interaction with p34 via interlocking eZnF domains (Schilbach et al., 2017) and a p44 RING domain interaction (Radu et al., 2017) (Figure 1B–D, Figure 1—figure supplement 6J,K), consistent with the formation of a multivalent interaction network between p34 and p44 (Radu et al., 2017). The vWFA domain of p34 recruits p52 by a three-way interaction that involves the most N-terminal winged helix domain in p52 and a helical segment of p62 (Schilbach et al., 2017) (Figure 1—figure supplement 6L). The p52 C-terminal region comprises two domains; first, the ‘clutch’ that interacts with XPB (Jawhari et al., 2002) and second, a dimerization module that binds p8 (Kainov et al., 2008), thereby recruiting XPB to TFIIH and cradling XPB RecA2 (see below). In addition to this structural framework that is formed by folded domains, our cryo-EM map reveals several interactions involving extended protein segments, including several interactions formed by p62 (Figure 2), and an interaction between the p44 N-terminal extension (NTE) and the N-terminal domain (NTD) of XPB (Figure 1B). To form this interaction, approx. 15 residues of p44 span the distance between the p44 vWFA domain and the XPB NTD, where a small helical motif in p44 contacts XPB residues 72–75, 95–102, and 139–143, in agreement with CX-MS data (Luo et al., 2015) (Figure 1—figure supplement 6E). Partial deletion of the p44 NTE in yeast causes a slow-growth phenotype, suggesting a functional role for this p44-XPB interaction (Warfield et al., 2016). Figure 2 with 1 supplement see all Download asset Open asset The structure of p62. (A) View of the top surface of the TFIIH core complex; p62 is color-coded by structural region. (B) The BSD2 (blue) and XPD anchor segments (teal) of p62 (surface) interact with the region around the XPD substrate-binding cavity. (C) Residues 346–365 of p62 (yellow) approach the nucleotide-binding pocket of XPD. ADP superposed from the structure of the DinG helicase (Cheng and Wigley, 2018). (D) Superposition of DNA from the structure of the substrate-bound DinG helicase (Cheng and Wigley, 2018) shows that the positions of p62 and RecA1-bound DNA overlap. https://doi.org/10.7554/eLife.44771.011 The p62 subunit is almost completely resolved in our structure and exhibits a complex beads-on-a-string-like topology. It fully encircles the top surface of TFIIH (Figures 1C and 2A, Figure 2—figure supplement 1), interacting with XPD, p52, p44, and p34, in agreement with previous structural findings (Greber et al., 2017; Schilbach et al., 2017). Based on these interactions, p62 can be subdivided into three functional regions: (i) the N-terminal PH-domain, disordered in our structure, is responsible for mediating interactions with components of the core transcriptional machinery (Di Lello et al., 2008; He et al., 2016; Schilbach et al., 2017), transcriptional regulators (Di Lello et al., 2006), and DNA repair pathways (Gervais et al., 2004; Lafrance-Vanasse et al., 2013; Okuda et al., 2017); (ii) residues 108–148 and 454–548 of p62, including the first BSD (BTF2-like, synapse-associated, DOS2-like) domain (BSD1) and the C-terminal 3-helix bundle, play an architectural role by binding to p34 and the extended zinc finger (eZnF) domain of p44 (Figure 2A, Figure 1—figure supplement 6J–L, Figure 2—figure supplement 1B,C); and (iii) residues 160–365, including the BSD2 domain, are responsible for interactions with and regulation of XPD (Figure 2B–D, Figure 2—figure supplement 1D–F). Specifically, p62 residues 160–365 form three structural elements that interact with XPD (Figure 2B, Video 1), in agreement with previous biochemical, structural, and CX-MS data (Figure 1—figure supplement 6G) (Jawhari et al., 2004; Luo et al., 2015; Schilbach et al., 2017). First, an α-helix formed by p62 residues 295–318 binds directly to XPD RecA2 and thereby recruits residues 160–258 of p62, comprising the BSD2 domain and adjacent sequence elements, to this surface of XPD RecA2 (Figure 2A,B, Figure 2—figure supplement 1D). Second, p62 residues 266–287 are inserted into the DNA-binding cavity of XPD (Figure 2B,D), in agreement with previous observations (Schilbach et al., 2017). This inserted p62 segment directly blocks a DNA-binding site on XPD RecA1 (Figure 2D) and localizes near the access path to a pore-like structure between the XPD FeS and ARCH domains. While p62 does not directly contact the DNA-binding surface on XPD RecA2, it may still sterically interfere with DNA binding or access to the helicase elements of XPD in this region (Figure 2—figure supplement 1E). Therefore, this segment of p62 may need to move away when XPD binds and unwinds DNA. Third, p62 residues 350–358 form a short α-helix that binds in a cleft between the two RecA-like domains of XPD (Figure 2C), so that it not only closes the entrance to the nucleotide binding pocket in XPD RecA1 (Figure 2—figure supplement 1F), but also partially overlaps with the predicted location of the nucleotide itself (Figure 2C), strongly suggesting a role for this p62 sequence element in XPD regulation. The density for these structural elements of p62 (residues 260–300 and 346–365) in our cryo-EM map is weaker than for the remainder of the complex, suggesting a dynamic interaction with XPD that enables them to modulate the access to the nucleotide-binding pocket, the DNA-binding cavity, and the DNA-translocating pore of XPD, depending on the functional state of TFIIH. 3D reconstructions of TFIIH classified for these regions of p62 (Figure 1—figure supplement 3) show globally intact TFIIH, both in the presence and absence of the p62 segments at these XPD sites (Figure 2—figure supplement 1G–J), supporting our hypothesis of dynamic regulation, rather than the alternative hypothesis of p62 binding to XPD as a requirement for TFIIH stability (Luo et al., 2015). Molecular basis of XPB recruitment by p52 Our structure of TFIIH resolves the structure and interactions of all four folded domains of human XPB – two RecA-like domains that form the SF2-family type helicase cassette, a DNA damage recognition domain (DRD)-like domain, and an N-terminal domain (NTD) (Figure 3A, Figure 3—figure supplement 1) – and reveals the molecular basis of XPB recruitment by p52. The XPB NTD encompasses residues 1–165, with the first 55 residues forming an N-terminal extension (NTE), and the remainder assuming a mixed α/β-fold with four α-helices and five β-strands (Figure 3A). The side chain densities in the cryo-EM map (Figure 1—figure supplement 6A) and CX-MS data (Luo et al., 2015) (Figure 1—figure supplement 6F) both confirm our assignment of this domain. Existing biochemical data show that the XPB NTD is required for integration of XPB into TFIIH (Jawhari et al., 2002) by forming an interaction with p52 that has been referred to as the ‘clutch’ (Schilbach et al., 2017). In our structure, the p52 contribution to the clutch encompasses p52 residues 306–399, which, strikingly, assume the same overall fold as the XPD NTD (Figure 3B), as hypothesized previously (He et al., 2016; Luo et al., 2015), thereby forming a pseudo-symmetric dimer of structurally homologous domains. The two domains interact through their β-sheets, via both hydrophobic and charged interactions (Figure 3—figure supplement 2A–C), and with the most N-terminal β-strand emanating from the XPB NTD extending the p52 β-sheet by additional lateral interactions (Figure 3A,C). Figure 3 with 2 supplements see all Download asset Open asset Structure and interactions of XPB. (A) Bottom lobe of TFIIH. XPB RecA1/2 teal, DRD blue, NTD dark blue, p52 yellow, p8 green, p44 NTE red. (B) Superposition of the XPB NTD and the p52 clutch domain. (C) Mapping of mutations on the XPB NTD and the p52 clutch domain; mutated regions are color-coded or shown as spheres (see text for (D) The interactions of the p52 the and the XPB NTD with XPB RecA2 may the of XPB RecA2 to XPB An extension of the DRD (blue) contacts XPD The sequence for which formation of an α-helix is predicted et al., 2015) is The DRD extension overlaps with the substrate-binding site on XPD DNA modeled from PDB ID et al., 2016). Our structural findings biochemical data that show that deletion of XPB residues but not deletion of residues the interaction (Jawhari et al., 2002) (Figure Our structure is also consistent with data that p52 residues are critical for the interaction (Coin et al., 2007; Jawhari et al., but does not show contacts that could that binding of XPB to p52 residues or (Jawhari et al., 2002) (Figure 3—figure supplement The interaction between p52 and XPB not only recruits XPB to TFIIH, but also its ATPase activity in (Coin et al., our structure does not shown elements of p52 the XPB nucleotide-binding pocket, we that this is likely by the interactions of p52 with the XPB NTD and RecA2, which with p8 (Coin et al., 2006), the XPB helicase to and (Figure et al., The XPB NTD is the site of the two human disease mutations and which cause and (Cleaver et al., structure shows that of these residues is in direct contact with p52 or the RecA-like domains of XPB, suggesting that the and mutations their through structural of the XPB NTD (Figure 3—figure supplements and Specifically, is near a turn at the of a β-strand (Figure 3—figure supplement where its side chain the this is in eukaryotic XPB from to in some and XPB (Figure 3—figure supplement 1B), and in the structurally homologous clutch domain in p52 (Figure 3—figure supplement This suggests that a at this is important for the of this domain in and that the may cause its in of active in overall of TFIIH have been shown to be a of et al., 2002; Dubaele et al., 2003; et al., and could the of in TFIIH this some activity in both transcription initiation and NER (Coin et al., A likely the of the in the p52 is that is involved in an interaction with a factor that is critical for function, for in The a that is eukaryotic XPB (Figure 3—figure supplement 1B), is inserted into a hydrophobic pocket, and localizes to an α-helix at the XPB contact site with the p44 N-terminal extension (Figure 3—figure supplement This is likely to the stability and of the XPB this to DNA opening in NER interaction with p52, ATPase activity (Coin et al., and in DNA damage repair et al., 1999), suggesting a on the structure of the XPB and mutations in the of p52 that lead to in and have when into human cells et al., map directly to the their (Figure Figure 3—figure supplement Our structure XPB residues to a domain that the NTD to the RecA-like domain (Figure 3A, Figure 3—figure supplement the deletion of which is in yeast (Warfield et al., 2016). The DRD is a DNA-binding domain in DNA repair and et al., et al., and has been in DNA damage recognition in XPB et al., 2006; and Our 3.7 map of TFIIH reveals that in eukaryotic XPB, one β-strand of the DRD of XPB is by an of approximately residues that exhibits sequence (Figure Figure 3—figure supplement and the domain of the human XPB domain with to previous sequence et al., 2006; et al., The of this resolved in our map of a charged and an element that contacts XPD directly (Figure The surface on XPD involved in this interaction has been in the of DNA substrate binding by XPD et al., 2016; Kuper et al., and structure the presence of several side of XPB near the (Figure where they form contacts those of of DNA (Figure XPB may modulate substrate binding by XPD, the that XPD activity is regulated by several components of TFIIH. of the TFIIH core complex In to the of TFIIH, we the of the in our cryo-EM (Figure Figure supplement see Materials and methods for et al., 2018). This revealed the of the two ATPases and their domains (Figure The major of which involves the of the interaction between XPB and XPD, the of TFIIH when it the Pol II-PIC and binds to DNA (Greber et al., 2017; He et al., 2016; Schilbach et al., 2017) (Figure supplement Video of our TFIIH structure and with that of the complex within the Pol II-PIC maps (He et al., 2016; Schilbach et al., 2017) allowed us to a specific at the between the clutch and adjacent winged helix domain in p52 (Figure as the basis of this A structural composed of XPB, p8 and the clutch domain of p52 a DNA-binding within the Pol II-PIC (Figure Figure supplement This in TFIIH PIC also to the interaction between MAT1 and the XPB domain (Figure which in turn to of the H dimer within the CAK subcomplex at the location for Pol in the Pol II-PIC (Figure et al., 2016; Schilbach et al., 2017). Our structural also reveals that a interaction that has been in XPB regulation (Schilbach et al., 2017) may on the of the of TFIIH, as be in a complex involving the of TFIIH (Figure supplement Figure with 2 supplements see all Download asset Open asset of TFIIH. (A) Results of multibody (also see Figure supplement 1 and Materials and methods for major of indicated by Å are not the enzymatic subunits of the TFIIH core complex or their domains. (B) of the p52 clutch domain and XPB NTD blue, to the remainder of p52 on of free and PIC-bound TFIIH structures and of domains into Pol II-PIC cryo-EM maps (He et al., 2016; Schilbach et al., 2017). (C) XPB from the and the The contact during this (D) model for the in MAT1 and of the CAK kinase module during Pol II-PIC of TFIIH. Video 2 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg of TFIIH during into the Pol of TFIIH from the observed in our structure of free TFIIH to the present in Pol II-PIC TFIIH, followed by a of TFIIH in the of the Pol PDB and maps used for II-PIC and of TFIIH from (He et al., 2016; Schilbach et al., 2017). Structure of XPD The structure of XPD shows the domain of two RecA-like domains (RecA1 and RecA2), with the FeS and ARCH domain in RecA1 et al., 2016; et al., 2008; Kuper et al., The of the map allowed us

The major challenges for density maps estimation and accurate counting stem from the large-scale variations, serious occlusions, and perspective distortions. Existing methods generally suffer from the blurred density maps, which are caused by average convolution kernel, and the ineffective estimation across different crowd scenes. In this paper, we propose a multi-scale fusion conditional generative adversarial network (MFC-GAN) that can generate high-resolution and high-quality density maps. The fusion module of MFC-GAN is embedded in a multi-scale generator and discriminator architecture with a novel adversarial loss, which is designed to guide high-resolution density maps generation. In order to address the problem of scale variation, we further propose a bidirectional fusion module. It combines deep global semantic features and shallow local information by leveraging feature maps presented in different layers of the generator. Furthermore, in order to increase the effectiveness of the multi-scale fusion, we design a cross-attention fusion module, which weights the multi-scale fused feature and learns context-aware feature maps for generating high quality density maps. The experiments on four challenging datasets show the effectiveness, feasibility and robustness of the proposed MFC-GAN.

In this article, a multiscale generative adversarial network (MS-GAN) is proposed for generating high-quality crowd density maps of arbitrary crowd density scenes. The task of crowd counting has many challenges, such as severe occlusions in extremely dense crowd scenes, perspective distortion, and high visual similarity between the pedestrians and background elements. To address these problems, the proposed MS-GAN combines a multiscale convolutional neural network (generator) and an adversarial network (discriminator) to generate a high-quality density map and accurately estimate the crowd count in complex crowd scenes. The multiscale generator utilizes the fusion features from multiple hierarchical layers to detect people with large-scale variation. The resulting density map produced by the multiscale generator is processed by a discriminator network trained to solve a binary classification task between a poor quality density map and real ground-truth ones. The additional adversarial loss can improve the quality of the density map, which is critical to accurately estimate the crowd counts. The experiments were conducted on multiple datasets with different crowd scenes and densities. The results showed that the proposed method provided better performance compared to current state-of-the-art methods.

The research evidence in orthographically transparent language Kannada indicates that phonological processing significantly influences reading acquisition, and deficits increase the risk of dyslexia. A comprehensive assessment of phonological processing that includes phonological awareness, phonological memory, and phonological naming is crucial for implementing effective intervention strategies, thereby reducing the risk of dyslexia. However, there is a notable absence of phonological processing assessment tools specifically designed and validated for children learning to read alphasyllabary Kannada. The present study addresses this gap by developing and validating a phonological processing assessment tool for children between Grade I and Grade III learning to read alphasyllabary languages such as Kannada. The study was conducted in two distinct phases. The first phase consisted of developing and piloting a phonological processing assessment tool. It included the stages of task selection, item generation, content validation, pilot testing, and reliability analysis. In the second phase, the developed tool was validated by administering it to both typically developing children and children at-risk for dyslexia from Grade I through Grade III. Subsequently, the developmental appropriateness of the tool was tested by comparing the performance of typically developing children between the grades. Additionally, diagnostic validity, including sensitivity, specificity, and area under the curve, was established by comparing the performance of typically developing and at-risk children. The study makes a substantial contribution to research on reading in Akshara orthographies, offering a valuable clinical tool for identifying children at-risk for dyslexia.