Michael Unger

Inferring causal molecular networks: empirical assessment through a community-based effort

It remains unclear whether causal, rather than merely correlational, relationships in molecular networks can be inferred in complex biological settings. Here we describe the HPN-DREAM network inference challenge, which focused on learning causal influences in signaling networks. We used phosphoprotein data from cancer cell lines as well as in silico data from a nonlinear dynamical model. Using the phosphoprotein data, we scored more than 2,000 networks submitted by challenge participants. The networks spanned 32 biological contexts and were scored in terms of causal validity with respect to unseen interventional data. A number of approaches were effective, and incorporating known biology was generally advantageous. Additional sub-challenges considered time-course prediction and visualization. Our results suggest that learning causal relationships may be feasible in complex settings such as disease states. Furthermore, our scoring approach provides a practical way to empirically assess inferred molecular networks in a causal sense.

Scalable inference of heterogeneous reaction kinetics from pooled single-cell recordings

Mathematical methods combined with measurements of single-cell dynamics provide a means to reconstruct intracellular processes that are only partly or indirectly accessible experimentally. To obtain reliable reconstructions, the pooling of measurements from several cells of a clonal population is mandatory. However, cell-to-cell variability originating from diverse sources poses computational challenges for such process reconstruction. We introduce a scalable Bayesian inference framework that properly accounts for population heterogeneity. The method allows inference of inaccessible molecular states and kinetic parameters; computation of Bayes factors for model selection; and dissection of intrinsic, extrinsic and technical noise. We show how additional single-cell readouts such as morphological features can be included in the analysis. We use the method to reconstruct the expression dynamics of a gene under an inducible promoter in yeast from time-lapse microscopy data.

Strengths and limitations of microarray-based phenotype prediction: lessons learned from the IMPROVER Diagnostic Signature Challenge

MOTIVATION After more than a decade since microarrays were used to predict phenotype of biological samples, real-life applications for disease screening and identification of patients who would best benefit from treatment are still emerging. The interest of the scientific community in identifying best approaches to develop such prediction models was reaffirmed in a competition style international collaboration called IMPROVER Diagnostic Signature Challenge whose results we describe herein. RESULTS Fifty-four teams used public data to develop prediction models in four disease areas including multiple sclerosis, lung cancer, psoriasis and chronic obstructive pulmonary disease, and made predictions on blinded new data that we generated. Teams were scored using three metrics that captured various aspects of the quality of predictions, and best performers were awarded. This article presents the challenge results and introduces to the community the approaches of the best overall three performers, as well as an R package that implements the approach of the best overall team. The analyses of model performance data submitted in the challenge as well as additional simulations that we have performed revealed that (i) the quality of predictions depends more on the disease endpoint than on the particular approaches used in the challenge; (ii) the most important modeling factor (e.g. data preprocessing, feature selection and classifier type) is problem dependent; and (iii) for optimal results datasets and methods have to be carefully matched. Biomedical factors such as the disease severity and confidence in diagnostic were found to be associated with the misclassification rates across the different teams. AVAILABILITY The lung cancer dataset is available from Gene Expression Omnibus (accession, GSE43580). The maPredictDSC R package implementing the approach of the best overall team is available at www.bioconductor.org or http://bioinformaticsprb.med.wayne.edu/.

A Cellular System for Spatial Signal Decoding in Chemical Gradients.

Directional cell growth requires that cells read and interpret shallow chemical gradients, but how the gradient directional information is identified remains elusive. We use single-cell analysis and mathematical modeling to define the cellular gradient decoding network in yeast. Our results demonstrate that the spatial information of the gradient signal is read locally within the polarity site complex using double-positive feedback between the GTPase Cdc42 and trafficking of the receptor Ste2. Spatial decoding critically depends on low Cdc42 activity, which is maintained by the MAPK Fus3 through sequestration of the Cdc42 activator Cdc24. Deregulated Cdc42 or Ste2 trafficking prevents gradient decoding and leads to mis-oriented growth. Our work discovers how a conserved set of components assembles a network integrating signal intensity and directionality to decode the spatial information contained in chemical gradients.

Learning diagnostic signatures from microarray data using L1-regularized logistic regression

Making reliable diagnoses and predictions based on high-throughput transcriptional data has attracted immense attention in the past few years. While experimental gene profiling techniques—such as microarray platforms—are advancing rapidly, there is an increasing demand of computational methods being able to efficiently handle such data. In this work we propose a computational workflow for extracting diagnostic gene signatures from high-throughput transcriptional profiling data. In particular, our research was performed within the scope of the first IMPROVER challenge. The goal of that challenge was to extract and verify diagnostic signatures based on microarray gene expression data in four different disease areas: psoriasis, multiple sclerosis, chronic obstructive pulmonary disease and lung cancer. Each of the different disease areas is handled using the same three-stage algorithm. First, the data are normalized based on a multi-array average (RMA) normalization procedure to account for variability among different samples and data sets. Due to the vast dimensionality of the profiling data, we subsequently perform a feature pre-selection using a Wilcoxon’s rank sum statistic. The remaining features are then used to train an L1-regularized logistic regression model which acts as our primary classifier. Using the four different data sets, we analyze the proposed method and demonstrate its use in extracting diagnostic signatures from microarray gene expression data.

Optimal variational perturbations for the inference of stochastic reaction dynamics

Although single-cell techniques are advancing rapidly, quantitative assessment of kinetic parameters is still characterized by ill-posedness and a large degree of uncertainty. In many standard experiments, where transcriptional activation is recorded upon application of a step-like external perturbation, cells almost instantaneously adapt such that only a few informative measurements can be obtained. Consequently, the information gain between subsequent experiments or time points is comparably low, which is reflected in a hardly decreasing parameter uncertainty. However, novel microfluidic techniques can be applied to synthesize more sophisticated perturbations to increase the informativeness of such time-course experiments. Here we introduce a mathematical framework to design optimal perturbations for the inference of stochastic reaction dynamics. Based on Bayesian statistics, we formulate a variational problem to find optimal temporal perturbations and solve it using a stochastic approximation algorithm. Simulations are provided for the realistic scenario of noisy and discrete-time measurements using two simple reaction networks.

From microscopy data to in silico environments for in vivo-oriented simulations.

In our previous study, we introduced a combination methodology of Fluorescence Correlation Spectroscopy (FCS) and Transmission Electron Microscopy (TEM), which is powerful to investigate the effect of intracellular environment to biochemical reaction processes. Now, we developed a reconstruction method of realistic simulation spaces based on our TEM images. Interactive raytracing visualization of this space allows the perception of the overall 3D structure, which is not directly accessible from 2D TEM images. Simulation results show that the diffusion in such generated structures strongly depends on image post-processing. Frayed structures corresponding to noisy images hinder the diffusion much stronger than smooth surfaces from denoised images. This means that the correct identification of noise or structure is significant to reconstruct appropriate reaction environment in silico in order to estimate realistic behaviors of reactants in vivo. Static structures lead to anomalous diffusion due to the partial confinement. In contrast, mobile crowding agents do not lead to anomalous diffusion at moderate crowding levels. By varying the mobility of these non-reactive obstacles (NRO), we estimated the relationship between NRO diffusion coefficient (Dnro) and the anomaly in the tracer diffusion (α). For Dnro=21.96 to 44.49 μm2/s, the simulation results match the anomaly obtained from FCS measurements. This range of the diffusion coefficient from simulations is compatible with the range of the diffusion coefficient of structural proteins in the cytoplasm. In addition, we investigated the relationship between the radius of NRO and anomalous diffusion coefficient of tracers by the comparison between different simulations. The radius of NRO has to be 58 nm when the polymer moves with the same diffusion speed as a reactant, which is close to the radius of functional protein complexes in a cell.

Optimal Perturbations for the Identification of Stochastic Reaction Dynamics

Identification of stochastic reaction dynamics inside the cell is hampered by the low-dimensional readouts available with todays measurement technologies. Moreover, such processes are poorly excited by standard experimental protocols, making identification even more ill-posed. Recent technological advances provide means to design and apply complex extra-cellular stimuli. Based on an information-theoretic setting we present novel Monte Carlo sampling techniques to determine optimal temporal excitation profiles for such stochastic processes. We give a new result for the controlled birth-death process and provide a proof of principle by considering a simple model of regulated gene expression.

Reconstructing dynamic molecular states from single-cell time series.

The notion of state for a system is prevalent in the quantitative sciences and refers to the minimal system summary sufficient to describe the time evolution of the system in a self-consistent manner. This is a prerequisite for a principled understanding of the inner workings of a system. Owing to the complexity of intracellular processes, experimental techniques that can retrieve a sufficient summary are beyond our reach. For the case of stochastic biomolecular reaction networks, we show how to convert the partial state information accessible by experimental techniques into a full system state using mathematical analysis together with a computational model. This is intimately related to the notion of conditional Markov processes and we introduce the posterior master equation and derive novel approximations to the corresponding infinite-dimensional posterior moment dynamics. We exemplify this state reconstruction approach using both in silico data and single-cell data from two gene expression systems in Saccharomyces cerevisiae, where we reconstruct the dynamic promoter and mRNA states from noisy protein abundance measurements.