Since its selection as the method of the year in 2013, single-cell technologies have become mature enough to provide answers to complex research questions. With the growth of single-cell profiling technologies, there has also been a significant increase in data collected from single-cell profilings, resulting in computational challenges to process these massive and complicated datasets. To address these challenges, deep learning (DL) is positioned as a competitive alternative for single-cell analyses besides the traditional machine learning approaches. Here, we survey a total of 25 DL algorithms and their applicability for a specific step in the single cell RNA-seq processing pipeline. Specifically, we establish a unified mathematical representation of variational autoencoder, autoencoder, generative adversarial network and supervised DL models, compare the training strategies and loss functions for these models, and relate the loss functions of these models to specific objectives of the data processing step. Such a presentation will allow readers to choose suitable algorithms for their particular objective at each step in the pipeline. We envision that this survey will serve as an important information portal for learning the application of DL for scRNA-seq analysis and inspire innovative uses of DL to address a broader range of new challenges in emerging multi-omics and spatial single-cell sequencing.

Motivation: Transcription factor (TF) binds to the promoter region of a gene to control gene expression. Identifying precise TF binding sites (TFBSs) is essential for understanding the detailed mechanisms of TF-mediated gene regulation. However, there is a shortage of computational approach that can deliver single base pair resolution prediction of TFBS. Results: In this paper, we propose DeepSNR, a Deep Learning algorithm for predicting TF binding location at Single Nucleotide Resolution de novo from DNA sequence. DeepSNR adopts a novel deconvolutional network (deconvNet) model and is inspired by the similarity to image segmentation by deconvNet. The proposed deconvNet architecture is constructed on top of 'DeepBind' and we trained the entire model using TF-specific data from ChIP-exonuclease (ChIP-exo) experiments. DeepSNR has been shown to outperform motif search-based methods for several evaluation metrics. We have also demonstrated the usefulness of DeepSNR in the regulatory analysis of TFBS as well as in improving the TFBS prediction specificity using ChIP-seq data. Availability and implementation: DeepSNR is available open source in the GitHub repository (https://github.com/sirajulsalekin/DeepSNR). Supplementary information: Supplementary data are available at Bioinformatics online.

We present here deep covariance learning models for predicting drivers' drowsy and alert states from Electroencephalography (EEG). Three types of deep covariance learning models are proposed: SPDNet, CNN, and DNN on covariance matrices. Our test results show that all the deep covariance learning methods reported better performance than shallow learning methods including Riemannian methods and STCNN, a previously proposed CNN model for EEG classification. Among the deep covariance learning methods, the best classification performance is obtained by a CNN model applied on sample spatial EEG covariance matrices and it improved the AUC of the best shallow algorithm (logistic regression + Log-Euclidean Metric) by 12.32% from 70.96% to 86.14%. Our study showed that deep covariance learning is a very promising approach for drivers' fatigue prediction.

BACKGROUND: Identifying disease correlated features early before large number of molecules are impacted by disease progression with significant abundance change is very advantageous to biologists for developing early disease diagnosis biomarkers. Disease correlated features have relatively low level of abundance change at early stages. Finding them using existing bioinformatic tools in high throughput data is a challenging task since the technology suffers from limited dynamic range and significant noise. Most existing biomarker discovery algorithms can only detect molecules with high abundance changes, frequently missing early disease diagnostic markers. RESULTS: We present a new statistic called early response index (ERI) to prioritize disease correlated molecules as potential early biomarkers. Instead of classification accuracy, ERI measures the average classification accuracy improvement attainable by a feature when it is united with other counterparts for classification. ERI is more sensitive to abundance changes than other ranking statistics. We have shown that ERI significantly outperforms SAM and Localfdr in detecting early responding molecules in a proteomics study of a mouse model of multiple sclerosis. Importantly, ERI was able to detect many disease relevant proteins before those algorithms detect them at a later time point. CONCLUSIONS: ERI method is more sensitive for significant feature detection during early stage of disease development. It potentially has a higher specificity for biomarker discovery, and can be used to identify critical time frame for disease intervention.

MOTIVATION: In liquid chromatography-mass spectrometry/tandem mass spectrometry (LC-MS/MS), it is necessary to link tandem MS-identified peptide peaks so that protein expression changes between the two runs can be tracked. However, only a small number of peptides can be identified and linked by tandem MS in two runs, and it becomes necessary to link peptide peaks with tandem identification in one run to their corresponding ones in another run without identification. In the past, peptide peaks are linked based on similarities in retention time (rt), mass or peak shape after rt alignment, which corrects mean rt shifts between runs. However, the accuracy in linking is still limited especially for complex samples collected from different conditions. Consequently, large-scale proteomics studies that require comparison of protein expression profiles of hundreds of patients can not be carried out effectively. METHOD: In this article, we consider the problem of linking peptides from a pair of LC-MS/MS runs and propose a new method, PeakLink (PL), which uses information in both the time and frequency domain as inputs to a non-linear support vector machine (SVM) classifier. The PL algorithm first uses a threshold on an rt likelihood ratio score to remove candidate corresponding peaks with excessively large elution time shifts, then PL calculates the correlation between a pair of candidate peaks after reducing noise through wavelet transformation. After converting rt and peak shape correlation to statistical scores, an SVM classifier is trained and applied for differentiating corresponding and non-corresponding peptide peaks. RESULTS: PL is tested in multiple challenging cases, in which LC-MS/MS samples are collected from different disease states, different instruments and different laboratories. Testing results show significant improvement in linking accuracy compared with other algorithms. AVAILABILITY AND IMPLEMENTATION: M files for the PL alignment method are available at http://compgenomics.utsa.edu/zgroup/PeakLink. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Peptide detection is a crucial step in mass spectrometry (MS) based proteomics. Most existing algorithms are based upon greedy isotope template matching and thus may be prone to error propagation and ineffective to detect overlapping peptides. In addition, existing algorithms usually work at different charge states separately, isolating useful information that can be drawn from other charge states, which may lead to poor detection of low abundance peptides. RESULTS: BPDA2d models spectra as a mixture of candidate peptide signals and systematically evaluates all possible combinations of possible peptide candidates to interpret the given spectra. For each candidate, BPDA2d takes into account its elution profile, charge state distribution and isotope pattern, and it combines all evidence to infer the candidate's signal and existence probability. By piecing all evidence together--especially by deriving information across charge states--low abundance peptides can be better identified and peptide detection rates can be improved. Instead of local template matching, BPDA2d performs global optimization for all candidates and systematically optimizes their signals. Since BPDA2d looks for the optimal among all possible interpretations of the given spectra, it has the capability in handling complex spectra where features overlap. BPDA2d estimates the posterior existence probability of detected peptides, which can be directly used for probability-based evaluation in subsequent processing steps. Our experiments indicate that BPDA2d outperforms state-of-the-art detection methods on both simulated data and real liquid chromatography-mass spectrometry data, according to sensitivity and detection accuracy. AVAILABILITY: The BPDA2d software package is available at http://gsp.tamu.edu/Publications/supplementary/sun11a/.

A novel quantification method is proposed for labeled LC-MS data. Unlike the traditional extracted ion chromatogram based quantification methods, the new method considers the reliability or the quality of signals with different intensities and applies more weights to the more reliable signals in the relative quantification procedure. Testing shows that the new method has improved accuracy than current quantification software.

Transcriptional regulation by transcription factors (TFs) and microRNAs controls when and how much RNA is created. Due to technical limitations, the protein level expressions of TFs are usually unknown, making computational reconstruction of transcriptional network a difficult task. We proposed here a novel Bayesian non negative hybrid factor model for transcriptional network modeling, which is capable to estimate both the non-negative abundances of the transcription factors, the regulatory effects of TFs and microRNAs, and the sample clustering information by integrating microarray data and existing knowledge regarding TFs and microRNAs regulated target genes. The results demonstrated its validity and effectiveness to reconstructing transcriptional networks through simulated systems and real data.

BACKGROUND: Identifying corresponding features (LC peaks registered by identical peptides) in multiple Liquid Chromatography/Mass Spectrometry (LC-MS) datasets plays a crucial role in the analysis of complex peptide or protein mixtures. Warping functions are commonly used to correct the mean of elution time shifts among LC-MS datasets, which cannot resolve the ambiguity of corresponding feature identification since elution time shifts are random. We propose a Statistical Corresponding Feature Identification Algorithm(SCFIA) based on both elution time shifts and peak shape correlations between corresponding features. SCFIA first trains a set of statistical models, and then, all candidate corresponding features are scored by the statistical models to find the maximum likelihood solution. RESULTS: We test SCFIA on publicly available datasets. We first compare its performance with that of warping function based methods, and the results show significant improvements. The performance of SCFIA on replicates datasets and fractionated datasets is also evaluated. In both cases, the accuracy is above 90%, which is near optimal. Finally the coverage of SCFIA is evaluated, and it is shown that SCFIA can find corresponding features in multiple datasets for over 90% peptides identified by Tandem MS. CONCLUSIONS: SCFIA can be used for accurate corresponding feature identification in LC-MS. We have shown that peak shape correlation can be used effectively for improving the accuracy. SCFIA provides high coverage in corresponding feature identification in multiple datasets, which serves the basis for integrating multiple LC-MS measurements for accurate peptide quantification.

BACKGROUND: Relative isotope abundance quantification, which can be used for peptide identification and differential peptide quantification, plays an important role in liquid chromatography-mass spectrometry (LC-MS)-based proteomics. However, several major issues exist in the relative isotopic quantification of peptides on time-of-flight (TOF) instruments: LC peak boundary detection, thermal noise suppression, interference removal and mass drift correction. We propose to use the Maximum Ratio Combining (MRC) method to extract MS signal templates for interference detection/removal and LC peak boundary detection. In our method, MRCQuant, MS templates are extracted directly from experimental values, and the mass drift in each LC-MS run is automatically captured and compensated. We compared the quantification accuracy of MRCQuant to that of another representative LC-MS quantification algorithm (msInspect) using datasets downloaded from a public data repository. RESULTS: MRCQuant showed significant improvement in the number of accurately quantified peptides. CONCLUSIONS: MRCQuant effectively addresses major issues in the relative quantification of LC-MS-based proteomics data, and it provides improved performance in the quantification of low abundance peptides.

Analysis of peptide profiles from a Liquid Chromatography Fourier Transform Mass Spectrometry (LC/FTMS) measurement reveals a non-linear distortion in intensity. Investigation of the measured CVi /C12 ratios comparing with theoretical ones shows that the non-linearity can be attributed to low intensity signal suppression of low abundance peptide peaks. We find that the suppression is homogenous for different isotopes of identical peptides but non-homogenous for different peptides. We developed an iterative correction algorithm that corrects the intensity distortions for peptides with relatively high abundance. This algorithm can be applied in a wide range of applications using FTMS.

The problem of uncovering transcriptional regulation by transcription factors (TFs) based on microarray data is considered. A novel Bayesian sparse correlated rectified factor model (BSCRFM) coupled with its ICM solution is proposed. BSCRFM models the unknown TF protein level activity, the correlated regulations between TFs, and the sparse nature of TF regulated genes and it admits prior knowledge from existing database regarding TF regulated target genes. An efficient Iterated Conditional Modes (ICM) algorithm is developed, and a maximum a posterior (MAP) solution is calculated from multiple ICM results to avoid the local maximum problem, a context-specific transcriptional regulatory network specific to the experimental condition of the microarray data can then be obtained. The proposed model's ICM algorithm and MAP solution are evaluated on the simulated systems and results demonstrated the validity and effectiveness of the proposed approach. The proposed model is also applied to the breast cancer microarray data and a TF regulated network is obtained.

BACKGROUND: Mass spectrometry (MS) is an essential analytical tool in proteomics. Many existing algorithms for peptide detection are based on isotope template matching and usually work at different charge states separately, making them ineffective to detect overlapping peptides and low abundance peptides. RESULTS: We present BPDA, a Bayesian approach for peptide detection in data produced by MS instruments with high enough resolution to baseline-resolve isotopic peaks, such as MALDI-TOF and LC-MS. We model the spectra as a mixture of candidate peptide signals, and the model is parameterized by MS physical properties. BPDA is based on a rigorous statistical framework and avoids problems, such as voting and ad-hoc thresholding, generally encountered in algorithms based on template matching. It systematically evaluates all possible combinations of possible peptide candidates to interpret a given spectrum, and iteratively finds the best fitting peptide signal in order to minimize the mean squared error of the inferred spectrum to the observed spectrum. In contrast to previous detection methods, BPDA performs deisotoping and deconvolution of mass spectra simultaneously, which enables better identification of weak peptide signals and produces higher sensitivities and more robust results. Unlike template-matching algorithms, BPDA can handle complex data where features overlap. Our experimental results indicate that BPDA performs well on simulated data and real MS data sets, for various resolutions and signal to noise ratios, and compares very favorably with commonly used commercial and open-source software, such as flexAnalysis, OpenMS, and Decon2LS, according to sensitivity and detection accuracy. CONCLUSION: Unlike previous detection methods, which only employ isotopic distributions and work at each single charge state alone, BPDA takes into account the charge state distribution as well, thus lending information to better identify weak peptide signals and produce more robust results. The proposed approach is based on a rigorous statistical framework, which avoids problems generally encountered in algorithms based on template matching. Our experiments indicate that BPDA performs well on both simulated data and real data, and compares very favorably with commonly used commercial and open-source software. The BPDA software can be downloaded from http://gsp.tamu.edu/Publications/supplementary/sun10a/bpda.

In this paper, a novel Bayesian peak detection algorithm is proposed for peptide peak detection in high resolution prOTOFtrade MALDI Mass Spectrometry(MS) data. A nonlinear parametric model is proposed for modeling the peptide signals, chemical noise, and thermal noise. A metropolized Gibbs sampling algorithm is derived for Bayesian peak detection. The proposed algorithm is compared with a popular wavelet-based algorithm and the results show a significant improvement in performance on simulated data. The algorithm is finally tested on real MS MALDI data and the results agree with visual inspection very well.

We investigate in this paper reverse engineering of gene regulatory networks from time series microarray data. We propose a dynamic Bayesian networks (DBNs) modeling and a full Bayesian learning scheme. The proposed DBN models directly the continuous expression levels and also is associated with parameters that indicate the degree as well as the types of regulations. To learn the network from data, we proposed a reversible jump Markov chain Monte Carlo (RJMCMC) algorithm. The RJMCMC algorithm can provide not only more accurate inference results than the deterministic alternative algorithms but also an estimate on the a posteriori probabilities (APPs) of the network topology. The estimated APPs provide useful information on the confidence of the inferred results and can also be used for efficient Bayesian data integration. The proposed approach was tested on yeast cell cycle microarray data and the results were compared with the KEGG pathway map

This paper considers the capacity of non-coherent multiple-input multiple-output (MIMO) fading channels with memory. A general decomposition for the MIMO mutual information is found first, then asymptotic lower and upper bounds are derived based on the decomposition. It is found that the upper and the lower bound will converge and a MIMO fading number is derived when the number of receive antennas approaches to infinity. It is shown that inputs with independent Euclidean norm and phase-amplitude-ratio vector can achieve this fading number when the number of receive antennas is large. The derived MIMO fading number reduces to the fading number of single-input multiple-output (SIMO) systems. A MIMO non-asymptotic lower bound also is presented

The capacity of non-coherent single-input multiple-output (SIMO) fading channels with memory is considered in this paper. A non-asymptotic SIMO capacity lower bound is provided, and it is shown that this lower bound is asymptotically tight. An explicit form of the SIMO fading number is also derived, which is equivalent to a previous form of the SIMO fading number

We propose in this paper a novel soft-input-soft-output (SISO) multiuser detector for synchronous CDMA systems. The detector is called the belief-directed sequential probabilistic data association detector (BD-SPDAD). The BD-SPDAD is developed based on a general framework of the probabilistic data association detector (PDAD) proposed in (Y Huang et al, IEEE Int. Symp. on Inf. Theo., 2004). However, novel extensions to the general framework are proposed for the BD-SPDAD, which result in a low complexity sequential implementation. Specifically, the complexity of the BD-SPDAD is reduced from O(K/sup 3/) of the original PDAD to O(K/sup 2/), where K is the number of users in the system. Moreover, we show through simulation that the performance of the BD-SPDAD is comparable and even better than the original PDAD especially in high signal-to-noise regions.

In this paper, we discuss the upper and lower bounds of the flat fading Gauss-Markov channel capacity. It is shown that due to phase variations on a correlated channel, the amount of information carried by phase will not grow with SNR asymptotically. As a result, optimal coherent designs or designs based on quasi-static assumptions can not be optimal for correlated fading channels. We also show that channel correlation may provide significant capacity gain over the capacity of independent fading channels. It is also shown that if the input is i.i.d., the asymptotic growth rate of the capacity is the same as that of independent Rayleigh flat fading channels. This indicates that the asymptotic region is very power inefficient.

In this paper, we propose a method for blind multiuser detection (MUD) in synchronous systems over flat and fast Rayleigh fading channels. We adopt an autoregressive-moving-average (ARMA) process to model the temporal correlation of the channels. Based on the ARMA process, we propose a novel time-observation state space model (TOSSM) that describes the dynamics of the addressed multiuser system. The TOSSM allows an MUD with natural blending of low complexity particle filtering (PF) and mixture Kalman filtering (for channel estimation). We further propose to use a more efficient PF algorithm known as the stochastic M-algorithm (SMA), which, although having lower complexity than the generic PF implementation, maintains comparable performance.

We investigate turbo equalization using an algorithm called probabilistic data association (PDA). We first propose a general structure for PDA, which consists of a linear interference cancellation step followed by a probabilistic data association step in every iteration. Based on the general structure, we show that the original PDA belongs to one variation and it is computationally inefficient. We then unveil that the popular soft linear MMSE (SLMMSE) equalizer can be considered as one sweep within a generalized PDA. Such a connection implies that further performance improvement over the SLMMSE equalizer is possible if the PDA is applied instead in turbo equalization. We also provide a way for the PDA equalizer to incorporate the a priori probability, which makes the PDA readily applicable to turbo equalization.

In digital communications, the symbol timing estimation is an very important element for high quality data detection. This paper considers the problem of joint symbol detection and timing estimation, whose optimal solution is analytically intractable. In this paper, a stochastic M-algorithm is proposed for the solution. The stochastic M-algorithm is a novel efficient particle filtering algorithm designed for discrete unknowns. To accommodate, in the stochastic M-algorithm, the continuous unknown symbol timing, the unscented Kalman filter is introduced, which leads to a very efficient implementation. The simulation results illustrated that the stochastic M-algorithm achieves similar performance to particle filtering with less than 1/25 of the complexity.

We present in this paper a general framework for the probabilistic data association multiuser detector (PDAD). This generalization allows for different algorithm variations though the design of interference cancellation (IC) filters. We examine two possible variations. In the first variation, we demonstrate a computational more efficient algorithm than the original PDAD. In the second variation, we draw the connection between the PDAD and some popular soft IC detectors and this connection implies that further performance improvement over the soft IC detectors is possible in a turbo multiuser detection (MUD) by the PDAD. Moreover, we show the equivalence of the two variations, which suggests that the IC step is computationally redundant for both the second variation and the soft IC detectors.

The paper addresses the problem of joint velocity estimation and data detection in a realistic scenario where mobile velocity changes continuously, resulting in non-stationary fast fading channels. A time-varying AR model and a Gauss-Markov model are used to describe the respective fading channel and variation of velocity. A connection is shown between the coefficients of the TVAR model and mobile velocity which makes the joint estimation and detection possible. A hierarchical dynamic state space model is formed for the problem, and a particle filtering algorithm is proposed. In particular, a hybrid importance function and the mixture Kalman filter are used to achieve efficient implementation of particle filtering. Simulation results are provided that show the performance of the particle filtering algorithm.

Bayesian estimators have been applied for both random and deterministic parameters. In the Bayesian estimation of deterministic parameters, the randomness is introduced only through the observations, and the prior distributions are adopted to impose certain constraints. In such cases, neither the well known Cramer-Rao lower bound (CRLB) or the posterior CRLB can be used reasonably as the performance lower bounds. We extend the theory of CRLB under the Bayesian framework to provide the lower bounds for both unbiased and biased Bayesian estimators of deterministic parameters. An example is provided to show the effectiveness of the proposed lower bound over other popular lower bounds.

This work demonstrates the use of particle filtering for detection in BLAST systems. A novel dynamic state-space model (DSSM) is constructed for BLAST systems that are crucial for development of particle filtering algorithms. The proposed DSSM is based on QR decomposition and the output of the feedforward filter, and it evolves in space. The particle filtering solution does not suffer from error propagation, and our simulation show that it greatly outperforms the V-BLAST and provides near optimum performance.

This correspondence analyzes the bit-error rate (BER) performance of coded synchronous code-division multiple-access (CDMA) systems assuming perfect channel state information (CSI) and optimal joint multiuser detection/decoding (OJMUDD). Our analysis is conducted in the same framework as that of uncoded systems. First, we derive the precise probability of an error event, then we provide an upper bound on the BER based on the sum of pairwise error probabilities, and, finally, we tighten the upper bound by considering decomposable error events. Many new concepts unique to coded systems are introduced. We propose to use quasi parity checks for identifying permissible error events, introduce the concept of compatible probability of error matrices, extend the list of conditions for identifying decomposable error events, and introduce the concept of conjugate sets to explore the symmetry among indecomposable error events. Simulation results are given along with theoretical predictions.