Design of a biochemical circuit motif for learning linear functions.

Learning and adaptive behaviour are fundamental biological processes. A key goal in the field of bioengineering is to develop biochemical circuit architectures with the ability to adapt to dynamic chemical environments. Here, we present a novel design for a biomolecular circuit capable of supervised learning of linear functions, using a model based on chemical reactions catalysed by DNAzymes. To achieve this, we propose a novel mechanism of maintaining and modifying internal state in biochemical systems, thereby advancing the state of the art in biomolecular circuit architecture. We use simulations to demonstrate that the circuit is capable of learning behaviour and assess its asymptotic learning performance, scalability and robustness to noise. Such circuits show great potential for building autonomous in vivo nanomedical devices. While such a biochemical system can tell us a great deal about the fundamentals of learning in living systems and may have broad applications in biomedicine (e.g. autonomous and adaptive drugs), it also offers some intriguing challenges and surprising behaviours from a machine learning perspective.

Reduced fMRI activity predicts relapse in patients recovering from stimulant dependence.

Relapse presents a significant problem for patients recovering from stimulant dependence. Here we examined the hypothesis that patterns of brain function obtained at an early stage of abstinence differentiates patients who later relapse versus those who remain abstinent. Forty-five recently abstinent stimulant-dependent patients were tested using a randomized event-related functional MRI (ER-fMRI) design that was developed in order to replicate a previous ERP study of relapse using a selective attention task, and were then monitored until 6 months of verified abstinence or stimulant use occurred. SPM revealed smaller absolute blood oxygen level-dependent (BOLD) response amplitude in bilateral ventral posterior cingulate and right insular cortex in 23 patients positive for relapse to stimulant use compared with 22 who remained abstinent. ER-fMRI, psychiatric, neuropsychological, demographic, personal and family history of drug use were compared in order to form predictive models. ER-fMRI was found to predict abstinence with higher accuracy than any other single measure obtained in this study. Logistic regression using fMRI amplitude in right posterior cingulate and insular cortex predicted abstinence with 77.8% accuracy, which increased to 89.9% accuracy when history of mania was included. Using 10-fold cross-validation, Bayesian logistic regression and multilayer perceptron algorithms provided the highest accuracy of 84.4%. These results, combined with previous studies, suggest that the functional organization of paralimbic brain regions including ventral anterior and posterior cingulate and right insula are related to patients' ability to maintain abstinence. Novel therapies designed to target these paralimbic regions identified using ER-fMRI may improve treatment outcome.

High-order interactions observed in multi-task intrinsic networks are dominant indicators of aberrant brain function in schizophrenia.

Identifying the complex activity relationships present in rich, modern neuroimaging data sets remains a key challenge for neuroscience. The problem is hard because (a) the underlying spatial and temporal networks may be nonlinear and multivariate and (b) the observed data may be driven by numerous latent factors. Further, modern experiments often produce data sets containing multiple stimulus contexts or tasks processed by the same subjects. Fusing such multi-session data sets may reveal additional structure, but raises further statistical challenges. We present a novel analysis method for extracting complex activity networks from such multifaceted imaging data sets. Compared to previous methods, we choose a new point in the trade-off space, sacrificing detailed generative probability models and explicit latent variable inference in order to achieve robust estimation of multivariate, nonlinear group factors ("network clusters"). We apply our method to identify relationships of task-specific intrinsic networks in schizophrenia patients and control subjects from a large fMRI study. After identifying network-clusters characterized by within- and between-task interactions, we find significant differences between patient and control groups in interaction strength among networks. Our results are consistent with known findings of brain regions exhibiting deviations in schizophrenic patients. However, we also find high-order, nonlinear interactions that discriminate groups but that are not detected by linear, pairwise methods. We additionally identify high-order relationships that provide new insights into schizophrenia but that have not been found by traditional univariate or second-order methods. Overall, our approach can identify key relationships that are missed by existing analysis methods, without losing the ability to find relationships that are known to be important.

TDCS guided using fMRI significantly accelerates learning to identify concealed objects.

The accurate identification of obscured and concealed objects in complex environments was an important skill required for survival during human evolution, and is required today for many forms of expertise. Here we used transcranial direct current stimulation (tDCS) guided using neuroimaging to increase learning rate in a novel, minimally guided discovery-learning paradigm. Ninety-six subjects identified threat-related objects concealed in naturalistic virtual surroundings used in real-world training. A variety of brain networks were found using functional magnetic resonance imaging (fMRI) data collected at different stages of learning, with two of these networks focused in right inferior frontal and right parietal cortex. Anodal 2.0 mA tDCS performed for 30 min over these regions in a series of single-blind, randomized studies resulted in significant improvements in learning and performance compared with 0.1 mA tDCS. This difference in performance increased to a factor of two after a one-hour delay. A dose-response effect of current strength on learning was also found. Taken together, these brain imaging and stimulation studies suggest that right frontal and parietal cortex are involved in learning to identify concealed objects in naturalistic surroundings. Furthermore, they suggest that the application of anodal tDCS over these regions can greatly increase learning, resulting in one of the largest effects on learning yet reported. The methods developed here may be useful to decrease the time required to attain expertise in a variety of settings.

Smoothness and Structure Learning by Proxy.

As data sets grow in size, the ability of learning methods to find structure in them is increasingly hampered by the time needed to search the large spaces of possibilities and generate a score for each that takes all of the observed data into account. For instance, Bayesian networks, the model chosen in this paper, have a super-exponentially large search space for a fixed number of variables. One possible method to alleviate this problem is to use a proxy, such as a Gaussian Process regressor, in place of the true scoring function, training it on a selection of sampled networks. We prove here that the use of such a proxy is well-founded, as we can bound the smoothness of a commonly-used scoring function for Bayesian network structure learning. We show here that, compared to an identical search strategy using the network's exact scores, our proxy-based search is able to get equivalent or better scores on a number of data sets in a fraction of the time.

A multiple network learning approach to capture system-wide condition-specific responses.

MOTIVATION: Condition-specific networks capture system-wide behavior under varying conditions such as environmental stresses, cell types or tissues. These networks frequently comprise parts that are unique to each condition, and parts that are shared among related conditions. Existing approaches for learning condition-specific networks typically identify either only differences or only similarities across conditions. Most of these approaches first learn networks per condition independently, and then identify similarities and differences in a post-learning step. Such approaches do not exploit the shared information across conditions during network learning. RESULTS: We describe an approach for learning condition-specific networks that identifies the shared and unique subgraphs during network learning simultaneously, rather than as a post-processing step. Our approach learns networks across condition sets, shares data from different conditions and produces high-quality networks that capture biologically meaningful information. On simulated data, our approach outperformed an existing approach that learns networks independently for each condition, especially for small training datasets. On microarray data of hundreds of deletion mutants in two, yeast stationary-phase cell populations, the inferred network structure identified several common and population-specific effects of these deletion mutants and several high-confidence cases of double-deletion pairs, which can be experimentally tested. Our results are consistent with and extend the existing knowledge base of differentiated cell populations in yeast stationary phase. AVAILABILITY AND IMPLEMENTATION: C++ code can be accessed from http://www.broadinstitute.org/~sroy/condspec/ .

Effective connectivity analysis of fMRI and MEG data collected under identical paradigms.

Estimation of effective connectivity, a measure of the influence among brain regions, can potentially reveal valuable information about organization of brain networks. Effective connectivity is usually evaluated from the functional data of a single modality. In this paper we show why that may lead to incorrect conclusions about effective connectivity. In this paper we use Bayesian networks to estimate connectivity on two different modalities. We analyze structures of estimated effective connectivity networks using aggregate statistics from the field of complex networks. Our study is conducted on functional MRI and magnetoencephalography data collected from the same subjects under identical paradigms. Results showed some similarities but also revealed some striking differences in the conclusions one would make on the fMRI data compared with the MEG data and are strongly supportive of the use of multiple modalities in order to gain a more complete picture of how the brain is organized given the limited information one modality is able to provide.

Correlated Noise: How it Breaks NMF, and What to Do About It.

Non-negative matrix factorization (NMF) is a problem of decomposing multivariate data into a set of features and their corresponding activations. When applied to experimental data, NMF has to cope with noise, which is often highly correlated. We show that correlated noise can break the Donoho and Stodden separability conditions of a dataset and a regular NMF algorithm will fail to decompose it, even when given freedom to be able to represent the noise as a separate feature. To cope with this issue, we present an algorithm for NMF with a generalized least squares objective function (glsNMF) and derive multiplicative updates for the method together with proving their convergence. The new algorithm successfully recovers the true representation from the noisy data. Robust performance can make glsNMF a valuable tool for analyzing empirical data.

MEG and fMRI Fusion for Non-Linear Estimation of Neural and BOLD Signal Changes.

The combined analysis of magnetoencephalography (MEG)/electroencephalography and functional magnetic resonance imaging (fMRI) measurements can lead to improvement in the description of the dynamical and spatial properties of brain activity. In this paper we empirically demonstrate this improvement using simulated and recorded task related MEG and fMRI activity. Neural activity estimates were derived using a dynamic Bayesian network with continuous real valued parameters by means of a sequential Monte Carlo technique. In synthetic data, we show that MEG and fMRI fusion improves estimation of the indirectly observed neural activity and smooths tracking of the blood oxygenation level dependent (BOLD) response. In recordings of task related neural activity the combination of MEG and fMRI produces a result with greater signal-to-noise ratio, that confirms the expectation arising from the nature of the experiment. The highly non-linear model of the BOLD response poses a difficult inference problem for neural activity estimation; computational requirements are also high due to the time and space complexity. We show that joint analysis of the data improves the system's behavior by stabilizing the differential equations system and by requiring fewer computational resources.

An Automated Method for Segmenting White Matter Lesions through Multi-Level Morphometric Feature Classification with Application to Lupus.

We demonstrate an automated, multi-level method to segment white matter brain lesions and apply it to lupus. The method makes use of local morphometric features based on multiple MR sequences, including T1-weighted, T2-weighted, and fluid attenuated inversion recovery. After preprocessing, including co-registration, brain extraction, bias correction, and intensity standardization, 49 features are calculated for each brain voxel based on local morphometry. At each level of segmentation a supervised classifier takes advantage of a different subset of the features to conservatively segment lesion voxels, passing on more difficult voxels to the next classifier. This multi-level approach allows for a fast lesion classification method with tunable trade-offs between sensitivity and specificity producing accuracy comparable to a human rater.

Exploiting amino acid composition for predicting protein-protein interactions.

BACKGROUND: Computational prediction of protein interactions typically use protein domains as classifier features because they capture conserved information of interaction surfaces. However, approaches relying on domains as features cannot be applied to proteins without any domain information. In this paper, we explore the contribution of pure amino acid composition (AAC) for protein interaction prediction. This simple feature, which is based on normalized counts of single or pairs of amino acids, is applicable to proteins from any sequenced organism and can be used to compensate for the lack of domain information. RESULTS: AAC performed at par with protein interaction prediction based on domains on three yeast protein interaction datasets. Similar behavior was obtained using different classifiers, indicating that our results are a function of features and not of classifiers. In addition to yeast datasets, AAC performed comparably on worm and fly datasets. Prediction of interactions for the entire yeast proteome identified a large number of novel interactions, the majority of which co-localized or participated in the same processes. Our high confidence interaction network included both well-studied and uncharacterized proteins. Proteins with known function were involved in actin assembly and cell budding. Uncharacterized proteins interacted with proteins involved in reproduction and cell budding, thus providing putative biological roles for the uncharacterized proteins. CONCLUSION: AAC is a simple, yet powerful feature for predicting protein interactions, and can be used alone or in conjunction with protein domains to predict new and validate existing interactions. More importantly, AAC alone performs at par with existing, but more complex, features indicating the presence of sequence-level information that is predictive of interaction, but which is not necessarily restricted to domains.

A framework for multiple kernel support vector regression and its applications to siRNA efficacy prediction.

The cell defense mechanism of RNA interference has applications in gene function analysis and promising potentials in human disease therapy. To effectively silence a target gene, it is desirable to select appropriate initiator siRNA molecules having satisfactory silencing capabilities. Computational prediction for silencing efficacy of siRNAs can assist this screening process before using them in biological experiments. String kernel functions, which operate directly on the string objects representing siRNAs and target mRNAs, have been applied to support vector regression for the prediction and improved accuracy over numerical kernels in multidimensional vector spaces constructed from descriptors of siRNA design rules. To fully utilize information provided by string and numerical data, we propose to unify the two in a kernel feature space by devising a multiple kernel regression framework where a linear combination of the kernels is used. We formulate the multiple kernel learning into a quadratically constrained quadratic programming (QCQP) problem, which although yields global optimal solution, is computationally demanding and requires a commercial solver package. We further propose three heuristics based on the principle of kernel-target alignment and predictive accuracy. Empirical results demonstrate that multiple kernel regression can improve accuracy, decrease model complexity by reducing the number of support vectors, and speed up computational performance dramatically. In addition, multiple kernel regression evaluates the importance of constituent kernels, which for the siRNA efficacy prediction problem, compares the relative significance of the design rules. Finally, we give insights into the multiple kernel regression mechanism and point out possible extensions.

Inference of functional networks of condition-specific response--a case study of quiescence in yeast.

Analysis of condition-specific behavior under stressful environmental conditions can provide insight into mechanisms causing different healthy and diseased cellular states. Functional networks (edges representing statistical dependencies) inferred from condition-specific expression data can provide fine-grained, network level information about conserved and specific behavior across different conditions. In this paper, we examine novel microarray compendia measuring gene expression from two unique stationary phase yeast cell populations, quiescent and non-quiescent. We make the following contributions: (a) develop a new algorithm to infer functional networks modeled as undirected probabilistic graphical models, Markov random fields, (b) infer functional networks for quiescent, non-quiescent cells and exponential cells, and (c) compare the inferred networks to identify processes common and different across these cells. We found that both non-quiescent and exponential cells have more gene ontology enrichment than quiescent cells. The exponential cells share more processes with non-quiescent than with quiescent, highlighting the novel and relatively under-studied characteristics of quiescent cells. Analysis of inferred subgraphs identified processes enriched in both quiescent and non-quiescent cells as well processes specific to each cell type. Finally, SNF1, which is crucial for quiescence, occurs exclusively among quiescent network hubs, while non-quiescent network hubs are enriched in human disease causing homologs.

Discrete dynamic Bayesian network analysis of fMRI data.

We examine the efficacy of using discrete Dynamic Bayesian Networks (dDBNs), a data-driven modeling technique employed in machine learning, to identify functional correlations among neuroanatomical regions of interest. Unlike many neuroimaging analysis techniques, this method is not limited by linear and/or Gaussian noise assumptions. It achieves this by modeling the time series of neuroanatomical regions as discrete, as opposed to continuous, random variables with multinomial distributions. We demonstrated this method using an fMRI dataset collected from healthy and demented elderly subjects (Buckner, et al., 2000: J Cogn Neurosci 12:24-34) and identify correlates based on a diagnosis of dementia. The results are validated in three ways. First, the elicited correlates are shown to be robust over leave-one-out cross-validation and, via a Fourier bootstrapping method, that they were not likely due to random chance. Second, the dDBNs identified correlates that would be expected given the experimental paradigm. Third, the dDBN's ability to predict dementia is competitive with two commonly employed machine-learning classifiers: the support vector machine and the Gaussian naive Bayesian network. We also verify that the dDBN selects correlates based on non-linear criteria. Finally, we provide a brief analysis of the correlates elicited from Buckner et al.'s data that suggests that demented elderly subjects have reduced involvement of entorhinal and occipital cortex and greater involvement of the parietal lobe and amygdala in brain activity compared with healthy elderly (as measured via functional correlations among BOLD measurements). Limitations and extensions to the dDBN method are discussed.

Learning structurally consistent undirected probabilistic graphical models.

In many real-world domains, undirected graphical models such as Markov random fields provide a more natural representation of the statistical dependency structure than directed graphical models. Unfortunately, structure learning of undirected graphs using likelihood-based scores remains difficult because of the intractability of computing the partition function. We describe a new Markov random field structure learning algorithm, motivated by canonical parameterization of Abbeel et al. We provide computational improvements on their parameterization by learning per-variable canonical factors, which makes our algorithm suitable for domains with hundreds of nodes. We compare our algorithm against several algorithms for learning undirected and directed models on simulated and real datasets from biology. Our algorithm frequently outperforms existing algorithms, producing higher-quality structures, suggesting that enforcing consistency during structure learning is beneficial for learning undirected graphs.

A system for generating transcription regulatory networks with combinatorial control of transcription.

UNLABELLED: We have developed a new software system, REgulatory Network generator with COmbinatorial control (RENCO), for automatic generation of differential equations describing pre-transcriptional combinatorics in artificial regulatory networks. RENCO has the following benefits: (a) it explicitly models protein-protein interactions among transcription factors, (b) it captures combinatorial control of transcription factors on target genes and (c) it produces output in Systems Biology Markup Language (SBML) format, which allows these equations to be directly imported into existing simulators. Explicit modeling of the protein interactions allows RENCO to incorporate greater mechanistic detail of the transcription machinery compared to existing models and can provide a better assessment of algorithms for regulatory network inference. AVAILABILITY: RENCO is a C++ command line program, available at http://sourceforge.net/projects/renco/

SiRNA silencing efficacy prediction using the RNA string kernel.

While most existing string kernels are developed for general purpose sequences and have been applied to text and protein classifications, the RNA string kernel is particularly designed to model mismatches, G-U wobbles, and bulges of RNA biology. We adapt the RNA kernel to compute the similarity of the short interfering RNAs (siRNAs), initiators of RNA interference, and use it in support vector regression to predict the siRNA silencing efficacy treated as a continuous variable. Empirical results on biological data sets demonstrate that the RNA string kernel performed favourably. In addition, it is simple to implement and fast to compute.

Implications of phase transitions in knockdown networks of transitive RNAi.

Gene silencing by RNA interference (RNAi) has been observed even in the presence of imperfect complementarity in the siRNA-mRNA hybridization. Since more permissive mismatches gives rise to higher chances of off-target gene silencing, the number of mismatched nucleotides allowed by nature becomes an important quantity in characterizing RNAi specificity and RNAi design. To estimate the allowable flexibility, we use scale-free graphs to model the knockdown interactions among genes by examining transitive RNAi (tRNAi), which amplifies siRNA and cyclically silences targets. We removed inefficient siRNA sequences using the commonly used siRNA efficacy rules, avoided redundant siRNAs using barcoding techniques, and employed both contiguous and scattered mismatches to emulate the siRNA-mRNA binding. Simulations in multiple organisms indicate that the fraction of the transcriptome silenced by tRNAi rises drastically with increased number of allowed mismatches and eventually tRNAi became self-destructive rather than defensive. At the phase transition, the number of mismatches implies a critical value beyond which tRNAi would cause the transcription of an organism to be instable. This critical value suggests an upper limit of no more than 6 nt mismatches in the hybridization in general.

RNA string kernels for RNAi off-target evaluation.

RNA interference (RNAi) is a posttranscriptional gene silencing mechanism used to study gene functions, knock down viral genes, and treat diseases therapeutically. However, an 'off-target effect' deteriorates its specificity and applicability. Complete off-target effects can only be characterised by examining each gene in a genome, which is too expensive to conduct experimentally and motivates a computational study. To simulate the sequence matching between an siRNA and its target mRNA allowing for mismatches, G-U wobbles and bulges, we propose string kernels and develop their efficient implementations for off-target detection. We evaluate RNAi specificities in Schizosaccharomyces pombe, Caenorhabdithis elegans, and human genomes.

A hidden-state Markov model for cell population deconvolution.

Microarrays measure gene expression typically from a mixture of cell populations during different stages of a biological process. However, the specific effects of the distinct or pure populations on measured gene expression are difficult or impossible to determine. The ability to deconvolve measured gene expression into the contributions from pure populations is critical to maximizing the potential of microarray analysis for investigating complex biological processes. In this paper, we describe a novel approach called the multinomial hidden Markov model (MHMM) that produces: (i) a maximum a posteriori estimate of the fraction represented by each pure population and (ii) gene expression values for each pure population. Our method uses an unsupervised, probabilistic approach for handling missing data points and clusters genes based on expression in pure populations. MHMM, used with several yeast datasets, identified statistically significant temporal dynamics. This method, unlike the linear decomposition models used previously for deconvolution, can extract information from different types of data, does not require a priori identification of pure gene expression, exploits the temporal nature of time series data, and is less affected by missing data.