Learning a confidence score and the latent space of a new supervised autoencoder for diagnosis and prognosis in clinical metabolomic studies.

BACKGROUND: Presently, there is a wide variety of classification methods and deep neural network approaches in bioinformatics. Deep neural networks have proven their effectiveness for classification tasks, and have outperformed classical methods, but they suffer from a lack of interpretability. Therefore, these innovative methods are not appropriate for decision support systems in healthcare. Indeed, to allow clinicians to make informed and well thought out decisions, the algorithm should provide the main pieces of information used to compute the predicted diagnosis and/or prognosis, as well as a confidence score for this prediction. METHODS: Herein, we used a new supervised autoencoder (SAE) approach for classification of clinical metabolomic data. This new method has the advantage of providing a confidence score for each prediction thanks to a softmax classifier and a meaningful latent space visualization and to include a new efficient feature selection method, with a structured constraint, which allows for biologically interpretable results. RESULTS: Experimental results on three metabolomics datasets of clinical samples illustrate the effectiveness of our SAE and its confidence score. The supervised autoencoder provides an accurate localization of the patients in the latent space, and an efficient confidence score. Experiments show that the SAE outperforms classical methods (PLS-DA, Random Forests, SVM, and neural networks (NN)). Furthermore, the metabolites selected by the SAE were found to be biologically relevant. CONCLUSION: In this paper, we describe a new efficient SAE method to support diagnostic or prognostic evaluation based on metabolomics analyses.

Primal-dual for classification with rejection (PD-CR): a novel method for classification and feature selection-an application in metabolomics studies.

BACKGROUND: Supervised classification methods have been used for many years for feature selection in metabolomics and other omics studies. We developed a novel primal-dual based classification method (PD-CR) that can perform classification with rejection and feature selection on high dimensional datasets. PD-CR projects data onto a low dimension space and performs classification by minimizing an appropriate quadratic cost. It simultaneously optimizes the selected features and the prediction accuracy with a new tailored, constrained primal-dual method. The primal-dual framework is general enough to encompass various robust losses and to allow for convergence analysis. Here, we compare PD-CR to three commonly used methods: partial least squares discriminant analysis (PLS-DA), random forests and support vector machines (SVM). We analyzed two metabolomics datasets: one urinary metabolomics dataset concerning lung cancer patients and healthy controls; and a metabolomics dataset obtained from frozen glial tumor samples with mutated isocitrate dehydrogenase (IDH) or wild-type IDH. RESULTS: PD-CR was more accurate than PLS-DA, Random Forests and SVM for classification using the 2 metabolomics datasets. It also selected biologically relevant metabolites. PD-CR has the advantage of providing a confidence score for each prediction, which can be used to perform classification with rejection. This substantially reduces the False Discovery Rate. CONCLUSION: PD-CR is an accurate method for classification of metabolomics datasets which can outperform PLS-DA, Random Forests and SVM while selecting biologically relevant features. Furthermore the confidence score provided with PD-CR can be used to perform classification with rejection and reduce the false discovery rate.

Detecting subtle transcriptomic perturbations induced by lncRNAs knock-down in single-cell CRISPRi screening using a new sparse supervised autoencoder neural network.

Single-cell CRISPR-based transcriptome screens are potent genetic tools for concomitantly assessing the expression profiles of cells targeted by a set of guides RNA (gRNA), and inferring target gene functions from the observed perturbations. However, due to various limitations, this approach lacks sensitivity in detecting weak perturbations and is essentially reliable when studying master regulators such as transcription factors. To overcome the challenge of detecting subtle gRNA induced transcriptomic perturbations and classifying the most responsive cells, we developed a new supervised autoencoder neural network method. Our Sparse supervised autoencoder (SSAE) neural network provides selection of both relevant features (genes) and actual perturbed cells. We applied this method on an in-house single-cell CRISPR-interference-based (CRISPRi) transcriptome screening (CROP-Seq) focusing on a subset of long non-coding RNAs (lncRNAs) regulated by hypoxia, a condition that promote tumor aggressiveness and drug resistance, in the context of lung adenocarcinoma (LUAD). The CROP-seq library of validated gRNA against a subset of lncRNAs and, as positive controls, HIF1A and HIF2A, the 2 main transcription factors of the hypoxic response, was transduced in A549 LUAD cells cultured in normoxia or exposed to hypoxic conditions during 3, 6 or 24 h. We first validated the SSAE approach on HIF1A and HIF2 by confirming the specific effect of their knock-down during the temporal switch of the hypoxic response. Next, the SSAE method was able to detect stable short hypoxia-dependent transcriptomic signatures induced by the knock-down of some lncRNAs candidates, outperforming previously published machine learning approaches. This proof of concept demonstrates the relevance of the SSAE approach for deciphering weak perturbations in single-cell transcriptomic data readout as part of CRISPR-based screening.

Mitotic Index Determination on Live Cells From Label-Free Acquired Quantitative Phase Images Using a Supervised Autoencoder.

This interdisciplinary work focuses on the interest of a new auto-encoder for supervised classification of live cell populations growing in a thermostated imaging station and acquired by a Quantitative Phase Imaging (QPI) camera. This type of camera produces interferograms that have to be processed to extract features derived from quantitative linear retardance and birefringence measurements. QPI is performed on living populations without any manipulation or treatment of the cells. We use the efficient new autoencoder classification method instead of the classical Douglas-Rachford method. Using this new supervised autoencoder, we show that the accuracy of the classification of the cells present in the mitotic phase of the cell cycle is very high using QPI features. This is a very important finding since we demonstrate that it is now possible to very precisely follow cell growth in a non-invasive manner, without any bias. No dye or any kind of markers are necessary for this live monitoring. Any studies requiring analysis of cell growth or cellular response to any treatment could benefit from this new approach by simply monitoring the proportion of cells entering mitosis in the studied cell population.

High-dimensional statistical measure for region-of-interest tracking.

This paper deals with region-of-interest (ROI) tracking in video sequences. The goal is to determine in successive frames the region which best matches, in terms of a similarity measure, a ROI defined in a reference frame. Some tracking methods define similarity measures which efficiently combine several visual features into a probability density function (PDF) representation, thus building a discriminative model of the ROI. This approach implies dealing with PDFs with domains of definition of high dimension. To overcome this obstacle, a standard solution is to assume independence between the different features in order to bring out low-dimension marginal laws and/or to make some parametric assumptions on the PDFs at the cost of generality. We discard these assumptions by proposing to compute the Kullback-Leibler divergence between high-dimensional PDFs using the k th nearest neighbor framework. In consequence, the divergence is expressed directly from the samples, i.e., without explicit estimation of the underlying PDFs. As an application, we defined 5, 7, and 13-dimensional feature vectors containing color information (including pixel-based, gradient-based and patch-based) and spatial layout. The proposed procedure performs tracking allowing for translation and scaling of the ROI. Experiments show its efficiency on a movie excerpt and standard test sequences selected for the specific conditions they exhibit: partial occlusions, variations of luminance, noise, and complex motion.

The OncoAge Consortium: Linking Aging and Oncology from Bench to Bedside and Back Again.

It is generally accepted that carcinogenesis and aging are two biological processes, which are known to be associated. Notably, the frequency of certain cancers (including lung cancer), increases significantly with the age of patients and there is now a wealth of data showing that multiple mechanisms leading to malignant transformation and to aging are interconnected, defining the so-called common biology of aging and cancer. OncoAge, a consortium launched in 2015, brings together the multidisciplinary expertise of leading public hospital services and academic laboratories to foster the transfer of scientific knowledge rapidly acquired in the fields of cancer biology and aging into innovative medical practice and silver economy development. This is achieved through the development of shared technical platforms (for research on genome stability, (epi)genetics, biobanking, immunology, metabolism, and artificial intelligence), clinical research projects, clinical trials, and education. OncoAge focuses mainly on two pilot pathologies, which benefit from the expertise of several members, namely lung and head and neck cancers. This review outlines the broad strategic directions and key advances of OncoAge and summarizes some of the issues faced by this consortium, as well as the short- and long-term perspectives.

Robust real-time segmentation of images and videos using a smooth-spline snake-based algorithm.

This paper deals with fast image and video segmentation using active contours. Region-based active contours using level sets are powerful techniques for video segmentation, but they suffer from large computational cost. A parametric active contour method based on B-Spline interpolation has been proposed in to highly reduce the computational cost, but this method is sensitive to noise. Here, we choose to relax the rigid interpolation constraint in order to robustify our method in the presence of noise: by using smoothing splines, we trade a tunable amount of interpolation error for a smoother spline curve. We show by experiments on natural sequences that this new flexibility yields segmentation results of higher quality at no additional computational cost. Hence, real-time processing for moving objects segmentation is preserved.

Gentle Nearest Neighbors Boosting over Proper Scoring Rules.

Tailoring nearest neighbors algorithms to boosting is an important problem. Recent papers study an approach, UNN, which provably minimizes particular convex surrogates under weak assumptions. However, numerical issues make it necessary to experimentally tweak parts of the UNN algorithm, at the possible expense of the algorithm's convergence and performance. In this paper, we propose a lightweight Newton-Raphson alternative optimizing proper scoring rules from a very broad set, and establish formal convergence rates under the boosting framework that compete with those known for UNN. To the best of our knowledge, no such boosting-compliant convergence rates were previously known in the popular Gentle Adaboost's lineage. We provide experiments on a dozen domains, including Caltech and SUN computer vision databases, comparing our approach to major families including support vector machines, (Ada)boosting and stochastic gradient descent. They support three major conclusions: (i) GNNB significantly outperforms UNN, in terms of convergence rate and quality of the outputs, (ii) GNNB performs on par with or better than computationally intensive large margin approaches, (iii) on large domains that rule out those latter approaches for computational reasons, GNNB provides a simple and competitive contender to stochastic gradient descent. Experiments include a divide-and-conquer improvement of GNNB exploiting the link with proper scoring rules optimization.

Design of signal-adapted multidimensional lifting scheme for lossy coding.

This paper proposes a new method for the design of lifting filters to compute a multidimensional nonseparable wavelet transform. Our approach is stated in the general case, and is illustrated for the 2-D separable and for the quincunx images. Results are shown for the JPEG2000 database and for satellite images acquired on a quincunx sampling grid. The design of efficient quincunx filters is a difficult challenge which has already been addressed for specific cases. Our approach enables the design of less expensive filters adapted to the signal statistics to enhance the compression efficiency in a more general case. It is based on a two-step lifting scheme and joins the lifting theory with Wiener's optimization. The prediction step is designed in order to minimize the variance of the signal, and the update step is designed in order to minimize a reconstruction error. Application for lossy compression shows the performances of the method.

From snakes to region-based active contours defined by region-dependent parameters.

Image and sequence segmentation of a the segmentation task are discussed from the point of view of optimizing the segmentation criterion. Such a segmentation criterion involves so-called (boundary and region) descriptors, which, in general, may depend on their respective boundaries or regions. This dependency must be taken into account when one is computing the criterion derivative with respect to the unknown object domain (defined by its boundary). If this dependency not considered, some correctional terms may be omitted. Computing the derivative of the segmentation criterion with a dynamic scheme is described. The scheme is general enough to provide a framework for a wide variety of applications in segmentation. It also provides a theoretical meaning to the philosophy of active contours.