Refining interaction search through signed iterative Random Forests

AbstractAdvances in supervised learning have enabled accurate prediction in biological systems governed by complex interactions among biomolecules. However, state-of-the-art predictive algorithms are typically “black-boxes,” learning statistical interactions that are difficult to translate into testable hypotheses. The iterative Random Forest (iRF) algorithm took a step towards bridging this gap by providing a computationally tractable procedure to identify the stable, high-order feature interactions that drive the predictive accuracy of Random Forests (RF). Here we refine the interactions identified by iRF to explicitly map responses as a function of interacting features. Our method, signed iRF (s-iRF), describes “subsets” of rules that frequently occur on RF decision paths. We refer to these “rule subsets” as signed interactions. Signed interactions share not only the same set of interacting features but also exhibit similar thresholding behavior, and thus describe a consistent functional relationship between interacting features and responses. We describe stable and predictive importance metrics (SPIMs) to rank signed interactions in terms of their stability, predictive accuracy, and strength of interaction. For each SPIM, we define null importance metrics that characterize its expected behavior under known structure. We evaluate our proposed approach in biologically inspired simulations and two case studies: predicting enhancer activity and spatial gene expression patterns. In the case of enhancer activity, s-iRF recovers one of the few experimentally validated high-order interactions and suggests novel enhancer elements where this interaction may be active. In the case of spatial gene expression patterns, s-iRF recovers all 11 reported links in the gap gene network. By refining the process of interaction recovery, our approach has the potential to guide mechanistic inquiry into systems whose scale and complexity is beyond human comprehension.

Download Full-text

Multiple enhancers ensure precision of gap gene-expression patterns in the Drosophila embryo

Proceedings of the National Academy of Sciences ◽

10.1073/pnas.1109873108 ◽

2011 ◽

Vol 108 (33) ◽

pp. 13570-13575 ◽

Cited By ~ 144

Author(s):

M. W. Perry ◽

A. N. Boettiger ◽

M. Levine

Keyword(s):

Gene Expression ◽

Expression Patterns ◽

Drosophila Embryo ◽

Gene Expression Patterns ◽

Gap Gene

Download Full-text

Sequence-based model of gap gene regulatory network

10.1101/015776 ◽

2015 ◽

Author(s):

Konstantin N Kozlov ◽

Vitaly V Gursky ◽

Ivan V Kulakovskiy ◽

Maria G Samsonova

Keyword(s):

Gene Expression ◽

Transcription Factor ◽

Binding Sites ◽

Gene Network ◽

Expression Patterns ◽

Gene Expression Patterns ◽

Model Output ◽

Wild Type ◽

Gap Gene ◽

Gene Regulatory

Background: The detailed analysis of transcriptional regulation is crucially important for understanding biological processes. The gap gene network in Drosophila attracts large interest among researches studying mechanisms of transcriptional regulation. It implements the most upstream regulatory layer of the segmentation gene network. The knowledge of molecular mechanisms involved in gap gene regulation is far less complete than that of genetics of the system. Mathematical modeling goes beyond insights gained by genetics and molecular approaches. It allows us to reconstruct wild-type gene expression patterns in silico, infer underlying regulatory mechanism and prove its sufficiency. Results: We developed a new model that provides a dynamical description of gap gene regulatory systems, using detailed DNA-based information, as well as spatial transcription factor concentration data at varying time points. We showed that this model correctly reproduces gap gene expression patterns in wild type embryos and is able to predict gap expression patterns in Kr mutants and four reporter constructs. We used four-fold cross validation test and fitting to random dataset to validate the model and proof its sufficiency in data description. The identifiability analysis showed that most model parameters are well identifiable. We reconstructed the gap gene network topology and studied the impact of individual transcription factor binding sites on the model output. We measured this impact by calculating the site regulatory weight as a normalized difference between the residual sum of squares error for the set of all annotated sites and the set, from which the site of interest was left out. Conclusions: The reconstructed topology of the gap gene network is in agreement with previous modeling results and data from literature. We showed that 1) the regulatory weights of transcription factor binding sites show very weak correlation with their PWM score; 2) sites with low regulatory weight are important for the model output; 3) functional important sites are not exclusively located in cis-regulatory elements, but are rather dispersed through regulatory region. It is of importance that some of the sites with high functional impact in hb, Kr and kni regulatory regions coincide with strong sites annotated and verified in Dnase I footprint assays. Keywords: transcription; thermodynamics; reaction-diffusion; drosophila

Download Full-text

Deep learning reveals evolutionary conservation and divergence of sequence properties underlying gene regulatory enhancers across mammals

10.1101/110676 ◽

2017 ◽

Cited By ~ 1

Author(s):

Ling Chen ◽

Alexandra E. Fish ◽

John A. Capra

Keyword(s):

Gene Expression ◽

Expression Patterns ◽

Support Vector ◽

Gene Expression Patterns ◽

Mammalian Evolution ◽

Short Sequence ◽

Enhancer Activity ◽

Enhancer Prediction ◽

Sequence Patterns ◽

Gene Regulatory

AbstractIn mammals, genomic regions with enhancer activity turnover rapidly; in contrast, gene expression patterns and transcription factor binding preferences are largely conserved. Based on this conservation, we hypothesized that enhancers active in different mammals would exhibit conserved sequence patterns in spite of their different genomic locations. We tested this hypothesis by quantifying the conservation of sequence patterns underlying histone-mark defined enhancers across six diverse mammals in two machine learning frameworks. We first trained support vector machine (SVM) classifiers based on the frequency spectrum of short DNA sequence patterns. These classifiers accurately identified many adult liver, developing limb, and developing brain enhancers in each species. Then, we applied these classifiers across species and found that classifiers trained in one species and tested in another performed nearly as well as classifiers trained and tested on the same species. This indicates that the short sequence patterns predictive of enhancers are largely conserved. We also observed similar cross-species conservation when applying the models to human and mouse enhancers validated in transgenic assays. The sequence patterns most predictive of enhancers in each species matched the binding motifs for a common set of TFs enriched for expression in relevant tissues, supporting the biological relevance of the learned features. To test the conservation of more complex sequences patterns, we trained convolutional neural networks (CNNs) on enhancer sequences in each species. The CNNs demonstrated better performance overall, but worse cross-species generalization than SVMs, suggesting the importance of combinatorial interactions between motifs, but less conservation of these more complex sequence patterns. Thus, despite the rapid change of active enhancer locations between mammals, cross-species enhancer prediction is often possible. Furthermore, short sequence patterns encoding enhancer activity have been maintained across more than 180 million years of mammalian evolution, with evolutionary change in more complex sequence patterns.Author summaryAlterations in gene expression levels are a driving force of both speciation and complex disease; therefore, it is of great importance to understand the mechanisms underlying the evolution and function gene regulatory DNA sequences. Recent studies have revealed that while gene expression patterns and transcription factor binding preferences are broadly conserved across diverse animals, there is extensive turnover in distal gene regulatory regions, called enhancers, between closely related species. We investigate this seeming incongruence by analyzing genome-wide enhancer datasets from six diverse mammalian species. We trained two machine-learning classifiers—a k-mer spectrum support vector machine (SVM) and convolutional neural network (CNN)—to distinguish enhancers from the genomic background. The k-mer spectrum SVM models the occurrences of short sequence patterns while the CNN models both the short sequences patterns and their combinatorial patterns. Both the SVM and CNN enhancer prediction models trained in one species are able to predict enhancers in the same cellular context in other species. However, CNNs performed better at predicting enhancers in each species, but they generalize less well across species than the SVMs. This argues that the short sequence properties encoding regulatory activity are remarkably conserved across more than 180 million years of mammalian evolution with more evolutionary turnover in the more complex combinations of the conserved short sequence motifs.

Download Full-text