Target-based screening is one of the major approaches in drug discovery. Besides the intended target, unexpected drug off-target interactions often occur, and many of them have not been recognized and characterized. The off-target interactions can be responsible for either therapeutic or side effects. Thus, identifying the genome-wide off-targets of lead compounds or existing drugs will be critical for designing effective and safe drugs, and providing new opportunities for drug repurposing. Although many computational methods have been developed to predict drug-target interactions, they are either less accurate than the one that we are proposing here or computationally too intensive, thereby limiting their capability for large-scale off-target identification. In addition, the performances of most machine learning based algorithms have been mainly evaluated to predict off-target interactions in the same gene family for hundreds of chemicals. It is not clear how these algorithms perform in terms of detecting off-targets across gene families on a proteome scale. Here, we are presenting a fast and accurate off-target prediction method, REMAP, which is based on a dual regularized one-class collaborative filtering algorithm, to explore continuous chemical space, protein space, and their interactome on a large scale. When tested in a reliable, extensive, and cross-gene family benchmark, REMAP outperforms the state-of-the-art methods. Furthermore, REMAP is highly scalable. It can screen a dataset of 200 thousands chemicals against 20 thousands proteins within 2 hours. Using the reconstructed genome-wide target profile as the fingerprint of a chemical compound, we predicted that seven FDA-approved drugs can be repurposed as novel anti-cancer therapies. The anti-cancer activity of six of them is supported by experimental evidences. Thus, REMAP is a valuable addition to the existing in silico toolbox for drug target identification, drug repurposing, phenotypic screening, and side effect prediction. The software and benchmark are available at https://github.com/hansaimlim/REMAP. High-throughput techniques have generated vast amounts of diverse omics and phenotypic data. However, these sets of data have not yet been fully explored to improve the effectiveness and efficiency of drug discovery, a process which has traditionally adopted a one-drug-one-gene paradigm. Consequently, the cost of bringing a drug to market is astounding and the failure rate is daunting. The failure of the target-based drug discovery is in large part due to the fact that a drug rarely interacts only with its intended receptor, but also generally binds to other receptors. To rationally design potent and safe therapeutics, we need to identify all the possible cellular proteins interacting with a drug in an organism. Existing experimental techniques are not sufficient to address this problem, and will benefit from computational modeling. However, it is a daunting task to reliably screen millions of chemicals against hundreds of thousands of proteins. Here, we introduce a fast and accurate method REMAP for large-scale predictions of drug-target interactions. REMAP outperforms state-of-the-art algorithms in terms of both speed and accuracy, and has been successfully applied to drug repurposing. Thus, REMAP may have broad applications in drug discovery.

In this letter, we study user grouping and beamforming for high altitude platform (HAP) massive multiple-input multiple-output (MIMO) systems. We first exploit the fact that the signal power mainly concentrates on statistical-eigenmode (SE), and we then use it to perform user grouping and design the outer beamformer. The user grouping is realized based on the average chordal distance between the SEs instead of the chordal distance between the covariance eigenspaces of the users in the same group. In this case, the complexity of a singular value decomposition is avoided. The outer beamformer is designed using reduced-dimensional SEs of the users, and the computational complexity is lower than in the case where the statistical correlation matrix is used. Numerical results demonstrate the performance enhancement of the proposed beamforming scheme in HAP massive MIMO systems compared to the existing schemes based on channel correlation matrix.

The measurement results in land mobile satellite (LMS) communication scenario have shown that the propagation characteristic of LMS channel is a non-stationary process. In this paper, a non-stationary three-dimensional (3-D) wideband geometry-based stochastic model (GBSM) and the corresponding simulation model are proposed for high-altitude platform (HAP) multiple-input multiple-output (MIMO) communication system. The proposed GBSM uses two-dimensional (2-D) one-ring and 3-D multi-cylinders to model moving scatterers and near and far stationary scatterers, respectively. In the proposed GBSM, we assume that the multipath components (MPCs) scattering from the far stationary scatterers could appear again after their disappearance, and a two-state continuous-time Markov process is used to model the dynamic properties of MPCs such as MPCs appearance and disappearance, and the closed-form expressions of the survival probabilities of MPCs are derived. In the proposed GBSM, the long distance and small-scale time-variant parameters are also used to investigate the non-stationarities of HAP-MIMO channel. Using the survival probabilities of MPCs, the statistical properties of the proposed 3-D GBSM are derived. The numerical results demonstrate that the proposed 3-D GBSM is more practical to characterize the HAP-MIMO channel.

It is known that a global navigation satellite system (GNSS) receiver is vulnerable to interference signals, and the interference type recognition, i.e., classification, is helpful for the receiver to sense the spectrum, select an efficient mitigation solution, or identify an interferer. However, conventional interference recognition methods are subject to feature-based expert classification methodologies that are very sensitive to the choice of features and often lack flexibility to different tasks. Recently, some researchers rely on convolutional networks and transform the interference recognition to image classification. Although straightforward, this type of method suffers problems when classifying more complex interference patterns. In this article, a new GNSS interference classifier is proposed, which is composed of a devised interference fingerprint spectrum (FPS) and an especially designed deep convolutional neural network, named as FPS deep convolutional neural network (DNN). The design of the FPS makes different interference signals more discernible. The proposed deep convolutional neural network greatly improves recognition accuracies, especially at low interference powers. The tests show that the average accuracy of the proposed FPS-DNN can reach more than 95% for nine types of interferences at −110 dBm power. This result is significantly superior to the feature-based expert classifier and the conventional convolutional network classifier. It is also demonstrated that the proposed FPS-DNN has a better generalization performance even if the interference is out of the parameter space of the training dataset.

In the indoor environment, where global positioning system (GPS) fails, the presence of multipath, non-line-of-sight (NLOS), and dense scatterer yield unreliable position estimation. In the massive multiple-input multiple-output (massive MIMO) system, the sparse channel property is exploited here for user localization as an integral part of the network to reduce the power of user equipment (UE). In this study, a novel spatial spectrum fusion estimation and localization (SSFEAL) algorithm using the uniform circular array (UCA) structure is proposed, which is based on the two-dimensional (2D) direction of arrival (DOA) estimation. The spatial spectrum of the incoherent narrowband uplink pilot signal on each frequency bin received at the distributed massive MIMO base stations is used for spectrum fusion. The main contributions of this research work are as follows: (1) The channel’s sparse characteristics at individual base stations are used to synthesize the covariance matrix of the received signal at the central fusion center. (2) The proposed method has 2D spatial information measurement fusion to generate informative spatial spectrum followed by grid refinement processing to reduce the localization error to sub-meter level. It is demonstrated that the proposed method using UCA gives higher-resolution estimation performance compared with the rectangular array in the localization accuracy. Comprehensive Monte Carlo simulation and real application test results show the validity and sub-meter precision in the indoor environment.

The clustered property among sparse weight vector is prevalent in applications such as direction of arrival (DOA) estimation. To address this, block structure has been utilized to capture the clustered information expressed by the continuity of nonzero elements in a sparse weight vector. However, many compressive sensing (CS)-based methods exploit inter-block sparsity while ignoring intra-block correlations. Sparse Bayesian learning (SBL)-based methods with parameterized distribution provide a flexible approach for handling block structures, but most of them assume known block partitioning. In addition, off-grid problem and multi-measurement vector (MMV) scenario are also common and challenging issues in CS-based methods. In this work, we fully utilize the potential clustered structure in the sparse solution of DOA estimation and derive a structure-aware SBL framework (SASBL) in MMV scenarios. The structure-aware matrix characterizes the correlation within each block, and the weight vector can describe the inter-block sparsity. A new learning rule for the weight vector is derived based on the Majorization-Minimization algorithm. Furthermore, the off-grid parameter is incorporated into the SASBL framework to obtain higher DOA estimation resolution. Simulation results demonstrate that the proposed SASBL outperforms subspace-based, typical SBL-based, off-grid SBL-based and gridless methods.