-
Improving Ab-Initio Cryo-EM Reconstruction with Semi-Amortized Pose Inference
Authors:
Shayan Shekarforoush,
David B. Lindell,
Marcus A. Brubaker,
David J. Fleet
Abstract:
Cryo-Electron Microscopy (cryo-EM) is an increasingly popular experimental technique for estimating the 3D structure of macromolecular complexes such as proteins based on 2D images. These images are notoriously noisy, and the pose of the structure in each image is unknown \textit{a priori}. Ab-initio 3D reconstruction from 2D images entails estimating the pose in addition to the structure. In this…
▽ More
Cryo-Electron Microscopy (cryo-EM) is an increasingly popular experimental technique for estimating the 3D structure of macromolecular complexes such as proteins based on 2D images. These images are notoriously noisy, and the pose of the structure in each image is unknown \textit{a priori}. Ab-initio 3D reconstruction from 2D images entails estimating the pose in addition to the structure. In this work, we propose a new approach to this problem. We first adopt a multi-head architecture as a pose encoder to infer multiple plausible poses per-image in an amortized fashion. This approach mitigates the high uncertainty in pose estimation by encouraging exploration of pose space early in reconstruction. Once uncertainty is reduced, we refine poses in an auto-decoding fashion. In particular, we initialize with the most likely pose and iteratively update it for individual images using stochastic gradient descent (SGD). Through evaluation on synthetic datasets, we demonstrate that our method is able to handle multi-modal pose distributions during the amortized inference stage, while the later, more flexible stage of direct pose optimization yields faster and more accurate convergence of poses compared to baselines. Finally, on experimental data, we show that our approach is faster than state-of-the-art cryoAI and achieves higher-resolution reconstruction.
△ Less
Submitted 14 June, 2024;
originally announced June 2024.
-
Dual-Camera Joint Deblurring-Denoising
Authors:
Shayan Shekarforoush,
Amanpreet Walia,
Marcus A. Brubaker,
Konstantinos G. Derpanis,
Alex Levinshtein
Abstract:
Recent image enhancement methods have shown the advantages of using a pair of long and short-exposure images for low-light photography. These image modalities offer complementary strengths and weaknesses. The former yields an image that is clean but blurry due to camera or object motion, whereas the latter is sharp but noisy due to low photon count. Motivated by the fact that modern smartphones co…
▽ More
Recent image enhancement methods have shown the advantages of using a pair of long and short-exposure images for low-light photography. These image modalities offer complementary strengths and weaknesses. The former yields an image that is clean but blurry due to camera or object motion, whereas the latter is sharp but noisy due to low photon count. Motivated by the fact that modern smartphones come equipped with multiple rear-facing camera sensors, we propose a novel dual-camera method for obtaining a high-quality image. Our method uses a synchronized burst of short exposure images captured by one camera and a long exposure image simultaneously captured by another. Having a synchronized short exposure burst alongside the long exposure image enables us to (i) obtain better denoising by using a burst instead of a single image, (ii) recover motion from the burst and use it for motion-aware deblurring of the long exposure image, and (iii) fuse the two results to further enhance quality. Our method is able to achieve state-of-the-art results on synthetic dual-camera images from the GoPro dataset with five times fewer training parameters compared to the next best method. We also show that our method qualitatively outperforms competing approaches on real synchronized dual-camera captures.
△ Less
Submitted 15 September, 2023;
originally announced September 2023.
-
Residual Multiplicative Filter Networks for Multiscale Reconstruction
Authors:
Shayan Shekarforoush,
David B. Lindell,
David J. Fleet,
Marcus A. Brubaker
Abstract:
Coordinate networks like Multiplicative Filter Networks (MFNs) and BACON offer some control over the frequency spectrum used to represent continuous signals such as images or 3D volumes. Yet, they are not readily applicable to problems for which coarse-to-fine estimation is required, including various inverse problems in which coarse-to-fine optimization plays a key role in avoiding poor local min…
▽ More
Coordinate networks like Multiplicative Filter Networks (MFNs) and BACON offer some control over the frequency spectrum used to represent continuous signals such as images or 3D volumes. Yet, they are not readily applicable to problems for which coarse-to-fine estimation is required, including various inverse problems in which coarse-to-fine optimization plays a key role in avoiding poor local minima. We introduce a new coordinate network architecture and training scheme that enables coarse-to-fine optimization with fine-grained control over the frequency support of learned reconstructions. This is achieved with two key innovations. First, we incorporate skip connections so that structure at one scale is preserved when fitting finer-scale structure. Second, we propose a novel initialization scheme to provide control over the model frequency spectrum at each stage of optimization. We demonstrate how these modifications enable multiscale optimization for coarse-to-fine fitting to natural images. We then evaluate our model on synthetically generated datasets for the the problem of single-particle cryo-EM reconstruction. We learn high resolution multiscale structures, on par with the state-of-the art.
△ Less
Submitted 26 October, 2022; v1 submitted 1 June, 2022;
originally announced June 2022.
-
Tunable electronic properties and electric-field-induced phase transition in phosphorene/graphene heterostructures
Authors:
Maryam Mahdavifar,
Sima Shekarforoush,
Farhad Khoeini
Abstract:
The shortcomings of mono-component systems, e.g., the gapless nature of graphene, the lack of air-stability in phosphorene, etc. have drawn great attention toward stacked materials expected to show interesting electronic and optical properties. Using the tight-binding approach and the Green's function method, we investigate the electronic properties of armchair-edged lateral phosphorene/graphene h…
▽ More
The shortcomings of mono-component systems, e.g., the gapless nature of graphene, the lack of air-stability in phosphorene, etc. have drawn great attention toward stacked materials expected to show interesting electronic and optical properties. Using the tight-binding approach and the Green's function method, we investigate the electronic properties of armchair-edged lateral phosphorene/graphene heterostructures, which are either semiconductor/semiconductor or semiconductor/metal heterostructures, depending on the width of graphene ribbon. It is found that the system is narrow-gapped, and the bandgap can be modulated by tuning the size of the domains. Besides, the analysis of the bandgap variation against the width of the component phosphorene ribbon indicates that, in semiconductor/metal heterostructure, phosphorene ribbon does not induce any electronic state near the Fermi level, suggesting that the suppressed electron transport should be attributed to the hole transfer across the interface. Furthermore, we show that the transverse electric field can significantly diversify the electronic behavior of the heterostructure, i.e., the heterostructure undergoes the semiconductor-metal phase transition. Moreover, tuning the transverse electric field yields an intriguing possibility that the system can undergo a topological phase transition from a band insulator to a topological insulator.
△ Less
Submitted 19 November, 2020;
originally announced November 2020.
-
InceptionGCN: Receptive Field Aware Graph Convolutional Network for Disease Prediction
Authors:
Anees Kazi,
Shayan shekarforoush,
S. Arvind krishna,
Hendrik Burwinkel,
Gerome Vivar,
Karsten Kortuem,
Seyed-Ahmad Ahmadi,
Shadi Albarqouni,
Nassir Navab
Abstract:
Geometric deep learning provides a principled and versatile manner for the integration of imaging and non-imaging modalities in the medical domain. Graph Convolutional Networks (GCNs) in particular have been explored on a wide variety of problems such as disease prediction, segmentation, and matrix completion by leveraging large, multimodal datasets. In this paper, we introduce a new spectral doma…
▽ More
Geometric deep learning provides a principled and versatile manner for the integration of imaging and non-imaging modalities in the medical domain. Graph Convolutional Networks (GCNs) in particular have been explored on a wide variety of problems such as disease prediction, segmentation, and matrix completion by leveraging large, multimodal datasets. In this paper, we introduce a new spectral domain architecture for deep learning on graphs for disease prediction. The novelty lies in defining geometric 'inception modules' which are capable of capturing intra- and inter-graph structural heterogeneity during convolutions. We design filters with different kernel sizes to build our architecture. We show our disease prediction results on two publicly available datasets. Further, we provide insights on the behaviour of regular GCNs and our proposed model under varying input scenarios on simulated data.
△ Less
Submitted 11 March, 2019;
originally announced March 2019.
-
Self-Attention Equipped Graph Convolutions for Disease Prediction
Authors:
Anees Kazi,
S. Arvind krishna,
Shayan Shekarforoush,
Karsten Kortuem,
Shadi Albarqouni,
Nassir Navab
Abstract:
Multi-modal data comprising imaging (MRI, fMRI, PET, etc.) and non-imaging (clinical test, demographics, etc.) data can be collected together and used for disease prediction. Such diverse data gives complementary information about the patientś condition to make an informed diagnosis. A model capable of leveraging the individuality of each multi-modal data is required for better disease prediction.…
▽ More
Multi-modal data comprising imaging (MRI, fMRI, PET, etc.) and non-imaging (clinical test, demographics, etc.) data can be collected together and used for disease prediction. Such diverse data gives complementary information about the patientś condition to make an informed diagnosis. A model capable of leveraging the individuality of each multi-modal data is required for better disease prediction. We propose a graph convolution based deep model which takes into account the distinctiveness of each element of the multi-modal data. We incorporate a novel self-attention layer, which weights every element of the demographic data by exploring its relation to the underlying disease. We demonstrate the superiority of our developed technique in terms of computational speed and performance when compared to state-of-the-art methods. Our method outperforms other methods with a significant margin.
△ Less
Submitted 24 December, 2018;
originally announced December 2018.
-
Optical Transitions and Localized Edge States in Skewed Phosphorene Nanoribbons
Authors:
Sima Shekarforoush,
Farhad Khoeini,
Daryoush Shiri
Abstract:
Using the Tight Binding (TB) parameters extracted from Density Functional Theory (DFT) and Recursive Green's Function method, it is shown that skewed-zigzag black phosphorous (phosphorene) nanoribbons obtain large and tuneable bandgap in response to vertical and transverse electric fields. Depending on the direction of the applied field the mid-gap states could possess the localized or metallic na…
▽ More
Using the Tight Binding (TB) parameters extracted from Density Functional Theory (DFT) and Recursive Green's Function method, it is shown that skewed-zigzag black phosphorous (phosphorene) nanoribbons obtain large and tuneable bandgap in response to vertical and transverse electric fields. Depending on the direction of the applied field the mid-gap states could possess the localized or metallic nature i.e. non-zero mid-gap density of states. Adjustability of the bandgap and optical dipole transition matrix elements are explained based on the symmetry of involved band edge states. This promises new electronic and optical devices based on phosphorene nanoribbons.
△ Less
Submitted 29 June, 2018; v1 submitted 6 February, 2018;
originally announced February 2018.
-
Nonlinear Optical Susceptibilities and Linear Absorption in Phosphorene Nanoribbons: Ab initio study
Authors:
Sima Shekarforoush,
Daryoush Shiri,
Farhad Khoeini
Abstract:
Using Density Functional Theory (DFT) method we compute linear optical absorption spectra and nonlinear optical susceptibilities of hydrogen passivated armchair and zigzag Phosphorous Nanoribbons (aPNR and zPNR) as well as α-phase phosphorous monolayer. We observe that: (a) Crystallographic direction has a strong effect on the band edge absorption which causes optical anisotropy as well as a red s…
▽ More
Using Density Functional Theory (DFT) method we compute linear optical absorption spectra and nonlinear optical susceptibilities of hydrogen passivated armchair and zigzag Phosphorous Nanoribbons (aPNR and zPNR) as well as α-phase phosphorous monolayer. We observe that: (a) Crystallographic direction has a strong effect on the band edge absorption which causes optical anisotropy as well as a red shift of absorption spectra by increasing the nanoribbon width. (b) The absorption values are in the order of $10^{5} cm^{-1}$ which are similar to the experimentally measured values. (c) There is two orders of magnitude enhancement of the 2nd order nonlinear optical susceptibility, $χ^{(2)}$, in nanoribbons which emanates from breaking the centro-symmetric structure of a monolayer phosphorene by hydrogen surface terminations. (d) Chief among our results is that the 3rd order susceptibility, $χ^{(3)}$, for phosphorene monolayer and nanoribbons are about $~10^{-13}$ esu ($~10^{-21} \frac{m^{2}}{V^{2}}$) which are in close agreement with experimentally reported values as well as a recently calculated value based on semi-analytic method. This strongly supports reliability of our method in calculating nonlinear optical susceptibilities of phosphorene and in general other nanostructures. Enhanced 2nd order optical nonlinearity in phosphorene promises better second harmonic and frequency difference (THz) generation for photonics applications.
△ Less
Submitted 16 October, 2017; v1 submitted 6 September, 2017;
originally announced September 2017.
