Biology subjects

Kostka, D.

Publications and source records attributed to Kostka, D..

6 recordsLinked to original sources

bioRxiv2018

The perils of interaction prediction

The availability of genome-wide maps of enhancer-promoter interactions (EPIs) has made it possible to use machine learning approaches to extract and interpret features that determine these interactions in different biological contexts. Multiple methods have claimed to accomplish the task of predicting enhancer-promoter interactions based on corresponding genomic features, but this problem is actually still far from being solved. In our analysis, we show that individual enhancer and promoter regions have widely different marginal interaction probabilities, e.g. propensities, which can lead to overfitting and memorization when random cross-validation is employed. Further even when a proper cross-validation scheme is adopted, a simple propensity-based model can still achieve a competitive performance without capturing any information about the EPI mechanism.

genomics↗

bioRxiv2018

Small non-coding RNA expression in developing mouse nephron progenitor cells

MicroRNAs (miRNAs) are small non-coding RNAs that are essential for the regulation of gene expression and play critical roles in human health and disease. Here we present comprehensive miRNA profiling data for mouse nephron progenitors, which give rise to most of the cell-types of the nephron, the functional units of the kidney. We describe a miRNA expression profile for nephron progenitors, with 162 miRNAs differentially expressed in nephron progenitors when compared to whole kidney. We also annotated 52, and experimentally validated 4 novel miRNAs, which are expressed in developing kidney. Our data is available as a public resource, so that it can be integrated into future studies and analyzed in the context of other functional and epigenomic data in kidney development. Specifically, it will be useful in the effort to shed light on molecular mechanisms underlying processes essential for normal kidney development, such as nephron progenitor specification, self-renewal and differentiation.

developmental biology↗

bioRxiv2018

Random forest based similarity learning for single cell RNA sequencing data

MotivationGenome-wide transcriptome sequencing applied to single cells (scRNA-seq) is rapidly becoming an assay of choice across many fields of biological and biomedical research. Scientific objectives often revolve around discovery or characterization of types or sub-types of cells, and therefore obtaining accurate cell-cell similarities from scRNA-seq data is critical step in many studies. While rapid advances are being made in the development of tools for scRNA-seq data analysis, few approaches exist that explicitly address this task. Furthermore, abundance and type of noise present in scRNA-seq datasets suggest that application of generic methods, or of methods developed for bulk RNA-seq data, is likely suboptimal.\n\nResultsHere we present RAFSIL, a random forest based approach to learn cell-cell similarities from scRNA-seq data. RAFSIL implements a two-step procedure, where feature construction geared towards scRNA-seq data is followed by similarity learning. It is designed to be adaptable and expandable, and RAFSIL similarities can be used for typical exploratory data analysis tasks like dimension reduction, visualization, and clustering. We show that our approach compares favorably with current methods across a diverse collection of datasets, and that it can be used to detect and highlight unwanted technical variation in scRNA-seq datasets in situations where other methods fail. Overall, RAFSIL implements a flexible approach yielding a useful tool that improves the analysis of scRNA-seq data.\n\nAvailability and ImplementationThe RAFSIL R package is available online at www.kostkalab.net/software.html

bioinformatics↗

bioRxiv2017

Developmental loci harbor clusters of accelerated regions that evolved independently in ape lineages

Some of the fastest evolving regions of the human genome are conserved non-coding elements with many human-specific DNA substitutions. These Human Accelerated Regions (HARs) are enriched nearby regulatory genes, and several HARs function as developmental enhancers. To investigate if this evolutionary signature is unique to humans, we quantified evidence of accelerated substitutions in conserved genomic elements across multiple lineages and applied this approach simultaneously to the genomes of five apes: human, chimpanzee, gorilla, orangutan, and gibbon. We find roughly similar numbers and genomic distributions of lineage-specific accelerated regions (linARs) in all five apes. In particular, apes share an enrichment of linARs in regulatory DNA nearby genes involved in development, especially transcription factors and other regulators. Many developmental loci harbor clusters of nonoverlapping linARs from multiple apes, suggesting that accelerated evolution in each species affected distinct regulatory elements that control a shared set of developmental pathways. Our statistical tests distinguish between GC-biased and unbiased accelerated substitution rates, allowing us to quantify the roles of different evolutionary forces in creating linARs. We find evidence of GC-biased gene conversion in each ape, but unbiased acceleration consistent with positive selection or loss of constraint is more common in all five lineages. It therefore appears that similar evolutionary processes created independent accelerated regions in the genomes of different apes, and that these lineage-specific changes to conserved non-coding sequences may have differentially altered expression of a core set of developmental genes across ape evolution.

evolutionary biology↗

bioRxiv2017

Modeling Enhancer-Promoter Interactions with Attention-Based Neural Networks

BackgroundGene regulatory sequences play critical roles in ensuring tightly controlled RNA expression patterns that are essential in a large variety of biological processes. Specifically, enhancer sequences drive expression of their target genes, and the availability of genome-wide maps of enhancer-promoter interactions has opened up the possibility to use machine learning approaches to extract and interpret features that define these interactions in different biological contexts.\n\nMethodsInspired by machine translation models we develop an attention-based neural network model, EPIANN, to predict enhancer-promoter interactions based on DNA sequences. Codes and data are available at https://github.com/wgmao/EPIANN.\n\nResultsOur approach accurately predicts enhancer-promoter interactions across six cell lines. In addition, our method generates pairwise attention scores at the sequence level, which specify how short regions in the enhancer and promoter pair-up to drive the interaction prediction. This allows us to identify over-represented transcription factors (TF) binding sites and TF-pair interactions in the context of enhancer function.

bioinformatics↗

bioRxiv2017

Genome-wide Enhancer Maps Differ Significantly in Genomic Distribution, Evolution, and Function

Non-coding gene regulatory enhancers are essential to transcription in mammalian cells. As a result, a large variety of experimental and computational strategies have been developed to identify cis-regulatory enhancer sequences. In practice, most studies consider enhancers identified by only a single method, and the concordance of enhancers identified by different methods has not been comprehensively evaluated. Here, we assess the similarities of enhancer sets identified by ten representative strategies in four biological contexts and evaluate the robustness of downstream conclusions to the choice of identification strategy. All pairs of enhancer sets we evaluated overlap significantly more than expected by chance; however, we also found significant dissimilarity between enhancer sets in their genomic characteristics, evolutionary conservation, and association with functional loci within each context. We find most regions identified as enhancers are supported by only one method. The disagreement is sufficient to influence interpretation of GWAS SNPs and eQTL, and to lead to disparate conclusions about enhancer biology and disease mechanisms. We also find only limited evidence that regions identified by multiple enhancer identification methods are better candidates than those identified by a single method. Our results highlight the inherent complexity of enhancer biology and argue that current approaches have yet to adequately account for enhancer diversity. As a result, we cannot recommend the use of any single enhancer identification strategy in isolation. To facilitate assessment of enhancer diversity on studies conclusions, we developed creDB, a database of enhancer annotations designed to integrate into bioinformatics workflows. While our findings highlight a major challenge to mapping the genetic architecture of complex disease and interpreting regulatory variants found in patient genomes, a systematic understanding of similarities and differences in enhancer identification methodology will ultimately enable robust inferences about gene regulatory sequences.

genetics↗