Motivation Reconstruction of genome-scale networks from gene expression data is an actively studied problem. A wide range of methods that differ between the types of interactions they uncover with varying trade-offs between sensitivity and specificity have been proposed. To leverage benefits of multiple such methods, ensemble network methods that combine predictions from resulting networks have been developed, promising results better than or as good as the individual networks. Perhaps owing to the difficulty in obtaining accurate training examples, these ensemble methods hitherto are unsupervised. Results In this paper, we introduce EnGRaiN, the first supervised ensemble learning method to construct gene networks. The supervision for training is provided by small training datasets of true edge connections (positives) and edges known to be absent (negatives) among gene pairs. We demonstrate the effectiveness of EnGRaiN using simulated datasets as well as a curated collection of A. thaliana datasets we created from microarray datasets available from public repositories. EnGRaiN shows better results not only in terms of ROC and PR characteristics for both real and simulated datasets compared to unsupervised methods for ensemble network construction, but also generates networks that can be mined for elucidating complex biological interactions. Availability EnGRaiN software and the datasets used in the study are publicly available at the github repository: https://github.com/AluruLab/EnGRaiN. Supplementary information Supplementary data are available at Bioinformatics online.

No abstract

Determining mortality risk is important for critical decisions in Intensive Care Units (ICU). The need for machine learning models that provide accurate patient-specific prediction of mortality is well recognized. We present a new algorithm for ICU mortality prediction that is designed to address the problem of imbalance, which occurs, in the context of binary classification, when one of the two classes is significantly under--represented in the data. We take a fundamentally new approach in exploiting the class imbalance through a feature transformation such that the transformed features are easier to classify. Hypothesis testing is used for classification with a test statistic that follows the distribution of the difference of two chi-squared random variables, for which there are no analytic expressions and we derive an accurate approximation. Experiments on a benchmark dataset of 4000 ICU patients show that our algorithm surpasses the best competing methods for mortality prediction.

Motivation: Gene regulatory networks (GRNs) are graphs that specify the interactions between transcription factors (TFs) and their target genes. Understanding these interactions is crucial for studying the mechanisms in cell differentiation, growth and development. Computational methods are needed to infer these networks from measured data. Although the availability of single cell RNA-Sequencing (scRNA-Seq) data provides unprecedented scale and resolution of gene-expression data, the inference of GRNs remains a challenge, mainly due to the complexity of the regulatory relationships and the noise in the data. Results: We propose GRNUlar, a novel deep learning architecture based on the unrolled algorithms idea for GRN inference from scRNA-Seq data. Like some existing methods which use prior information of which genes are TFs, GRNUlar also incorporates this TF information using a sparse multi-task deep learning architecture. We also demonstrate the application of a recently developed unrolled architecture GLAD to recover undirected GRNs in the absence of TF information. These unrolled architectures require supervision to train, for which we leverage the existing synthetic data simulators which generate scRNA-Seq data guided by a GRN. We show that unrolled algorithms outperform the state-of-the-art methods on synthetic data as well as real datasets in both the settings of TF information being absent or available.

Recovering sparse conditional independence graphs from data is a fundamental problem in machine learning with wide applications. A popular formulation of the problem is an 1 regularized maximum likelihood estimation. Many convex optimization algorithms have been designed to solve this formulation to recover the graph structure. Recently, there is a surge of interest to learn algorithms directly based on data, and in this case, learn to map empirical covariance to the sparse precision matrix. However, it is a challenging task in this case, since the symmetric positive definiteness (SPD) and sparsity of the matrix are not easy to enforce in learned algorithms, and a direct mapping from data to precision matrix may contain many parameters. We propose a deep learning architecture, GLAD, which uses an Alternating Minimization (AM) algorithm as our model inductive bias, and learns the model parameters via supervised learning. We show that GLAD learns a very compact and effective model for recovering sparse graph from data.Recently, there has been a surge of interest in a new paradigm of algorithm design, where algorithms are augmented with learning modules trained directly with data, rather than prescribing every step of the algorithms. This is meaningful because very often a family of optimization problems needs to be solved again and again, similar in structures but different in data. A data-driven algorithm may be able to leverage this distribution of problem instances, and learn an algorithm which performs better than traditional convex formulation. In our case, the sparse graph recovery problem may also Preprint. Under review.