BindVAE: Dirichlet variational autoencoders for de novo motif discovery from accessible chromatin.
Ontology highlight
ABSTRACT:
SUBMITTER: Kshirsagar M
PROVIDER: S-EPMC9380350 | biostudies-literature | 2022 Aug
REPOSITORIES: biostudies-literature
Publications
BindVAE: Dirichlet variational autoencoders for de novo motif discovery from accessible chromatin.
Kshirsagar Meghana M Yuan Han H Ferres Juan Lavista JL Leslie Christina C
Genome biology 20220815 1
We present a novel unsupervised deep learning approach called BindVAE, based on Dirichlet variational autoencoders, for jointly decoding multiple TF binding signals from open chromatin regions. BindVAE can disentangle an input DNA sequence into distinct latent factors that encode cell-type specific in vivo binding signals for individual TFs, composite patterns for TFs involved in cooperative binding, and genomic context surrounding the binding sites. On the task of retrieving the motifs of expre ...[more]
Similar Datasets
Project description:BackgroundBacterial surfaces are complex systems, constructed from membranes, peptidoglycan and, importantly, proteins. The proteins play crucial roles as critical regulators of how the bacterium interacts with and survive in its environment. A full catalog of the motifs in protein families and their relative conservation grade is a prerequisite to target the protein-protein interaction that bacterial surface protein makes to host proteins.ResultsIn this paper, we propose a greedy approach to identify conserved motifs in large sequence families iteratively. Each iteration discovers a motif de novo and masks all occurrences of that motif. Remaining unmasked sequences are subjected to the next round of motif detection until no more significant motifs can be found. We demonstrate the utility of the method through the construction of a proteome-wide motif repository for Group A Streptococcus (GAS), a significant human pathogen. GAS produce numerous surface proteins that interact with over 100 human plasma proteins, helping the bacteria to evade the host immune response. We used the repository to find that proteins part of the bacterial surface has motif architectures that differ from intracellular proteins.ConclusionsWe elucidate that the M protein, a coiled-coil homodimer that extends over 500 A from the cell wall, has a motif architecture that differs between various GAS strains. As the M protein is known to bind a variety of different plasma proteins, the results indicate that the different motif architectures are responsible for the quantitative differences of plasma proteins that various strains bind. The speed and applicability of the method enable its application to all major human pathogens.
| S-EPMC6471678 | biostudies-literature
Project description:SummaryHigh-throughput sequencing can enhance the analysis of aptamer libraries generated by the Systematic Evolution of Ligands by EXponential enrichment. Robust analysis of the resulting sequenced rounds is best implemented by determining a ranked consensus of reads following the processing by multiple aptamer detection algorithms. While several such approaches have been developed to this end, their installation and implementation is problematic. We developed AptCompare, a cross-platform program that combines six of the most widely used analytical approaches for the identification of RNA aptamer motifs and uses a simple weighted ranking to order the candidate aptamers, all driven within the same GUI-enabled environment. We demonstrate AptCompare's performance by identifying the top-ranked candidate aptamers from a previously published selection experiment in our laboratory, with follow-up bench assays demonstrating good correspondence between the sequences' rankings and their binding affinities.Availability and implementationThe source code and pre-built virtual machine images are freely available at https://bitbucket.org/shiehk/aptcompare.Supplementary informationSupplementary data are available at Bioinformatics online.
| S-EPMC8524629 | biostudies-literature
Project description:The BaMM web server offers four tools: (i) de-novo discovery of enriched motifs in a set of nucleotide sequences, (ii) scanning a set of nucleotide sequences with motifs to find motif occurrences, (iii) searching with an input motif for similar motifs in our BaMM database with motifs for >1000 transcription factors, trained from the GTRD ChIP-seq database and (iv) browsing and keyword searching the motif database. In contrast to most other servers, we represent sequence motifs not by position weight matrices (PWMs) but by Bayesian Markov Models (BaMMs) of order 4, which we showed previously to perform substantially better in ROC analyses than PWMs or first order models. To address the inadequacy of P- and E-values as measures of motif quality, we introduce the AvRec score, the average recall over the TP-to-FP ratio between 1 and 100. The BaMM server is freely accessible without registration at https://bammmotif.mpibpc.mpg.de.
| S-EPMC6030882 | biostudies-literature
Project description:De novo motif discovery has been an important challenge of bioinformatics for the past two decades. Since the emergence of high-throughput techniques like ChIP-seq, ChIP-exo and protein-binding microarrays (PBMs), the focus of de novo motif discovery has shifted to runtime and accuracy on large data sets. For this purpose, specialized algorithms have been designed for discovering motifs in ChIP-seq or PBM data. However, none of the existing approaches work perfectly for all three high-throughput techniques. In this article, we propose Dimont, a general approach for fast and accurate de novo motif discovery from high-throughput data. We demonstrate that Dimont yields a higher number of correct motifs from ChIP-seq data than any of the specialized approaches and achieves a higher accuracy for predicting PBM intensities from probe sequence than any of the approaches specifically designed for that purpose. Dimont also reports the expected motifs for several ChIP-exo data sets. Investigating differences between in vitro and in vivo binding, we find that for most transcription factors, the motifs discovered by Dimont are in good accordance between techniques, but we also find notable exceptions. We also observe that modeling intra-motif dependencies may increase accuracy, which indicates that more complex motif models are a worthwhile field of research.
| S-EPMC3834837 | biostudies-literature
Project description:BackgroundA strong focus of the post-genomic era is mining of the non-coding regulatory genome in order to unravel the function of regulatory elements that coordinate gene expression (Nat 489:57-74, 2012; Nat 507:462-70, 2014; Nat 507:455-61, 2014; Nat 518:317-30, 2015). Whole-genome approaches based on next-generation sequencing (NGS) have provided insight into the genomic location of regulatory elements throughout different cell types, organs and organisms. These technologies are now widespread and commonly used in laboratories from various fields of research. This highlights the need for fast and user-friendly software tools dedicated to extracting cis-regulatory information contained in these regulatory regions; for instance transcription factor binding site (TFBS) composition. Ideally, such tools should not require prior programming knowledge to ensure they are accessible for all users.ResultsWe present TrawlerWeb, a web-based version of the Trawler_standalone tool (Nat Methods 4:563-5, 2007; Nat Protoc 5:323-34, 2010), to allow for the identification of enriched motifs in DNA sequences obtained from next-generation sequencing experiments in order to predict their TFBS composition. TrawlerWeb is designed for online queries with standard options common to web-based motif discovery tools. In addition, TrawlerWeb provides three unique new features: 1) TrawlerWeb allows the input of BED files directly generated from NGS experiments, 2) it automatically generates an input-matched biologically relevant background, and 3) it displays resulting conservation scores for each instance of the motif found in the input sequences, which assists the researcher in prioritising the motifs to validate experimentally. Finally, to date, this web-based version of Trawler_standalone remains the fastest online de novo motif discovery tool compared to other popular web-based software, while generating predictions with high accuracy.ConclusionsTrawlerWeb provides users with a fast, simple and easy-to-use web interface for de novo motif discovery. This will assist in rapidly analysing NGS datasets that are now being routinely generated. TrawlerWeb is freely available and accessible at: http://trawler.erc.monash.edu.au .
| S-EPMC5887194 | biostudies-literature
Project description:Conformational sampling of biomolecules using molecular dynamics simulations often produces a large amount of high dimensional data that makes it difficult to interpret using conventional analysis techniques. Dimensionality reduction methods are thus required to extract useful and relevant information. Here, we devise a machine learning method, Gaussian mixture variational autoencoder (GMVAE), that can simultaneously perform dimensionality reduction and clustering of biomolecular conformations in an unsupervised way. We show that GMVAE can learn a reduced representation of the free energy landscape of protein folding with highly separated clusters that correspond to the metastable states during folding. Since GMVAE uses a mixture of Gaussians as its prior, it can directly acknowledge the multi-basin nature of the protein folding free energy landscape. To make the model end-to-end differentiable, we use a Gumbel-softmax distribution. We test the model on three long-timescale protein folding trajectories and show that GMVAE embedding resembles the folding funnel with folded states down the funnel and unfolded states outside the funnel path. Additionally, we show that the latent space of GMVAE can be used for kinetic analysis and Markov state models built on this embedding produce folding and unfolding timescales that are in close agreement with other rigorous dynamical embeddings such as time independent component analysis.
| S-EPMC8605902 | biostudies-literature
Project description:Assembly of reads from metagenomic samples is a hard problem, often resulting in highly fragmented genome assemblies. Metagenomic binning allows us to reconstruct genomes by re-grouping the sequences by their organism of origin, thus representing a crucial processing step when exploring the biological diversity of metagenomic samples. Here we present Adversarial Autoencoders for Metagenomics Binning (AAMB), an ensemble deep learning approach that integrates sequence co-abundances and tetranucleotide frequencies into a common denoised space that enables precise clustering of sequences into microbial genomes. When benchmarked, AAMB presented similar or better results compared with the state-of-the-art reference-free binner VAMB, reconstructing ~7% more near-complete (NC) genomes across simulated and real data. In addition, genomes reconstructed using AAMB had higher completeness and greater taxonomic diversity compared with VAMB. Finally, we implemented a pipeline Integrating VAMB and AAMB that enabled improved binning, recovering 20% and 29% more simulated and real NC genomes, respectively, compared to VAMB, with moderate additional runtime.
| S-EPMC10590447 | biostudies-literature
Project description:Deep generative models, such as variational autoencoders (VAE), have gained increasing attention in computational biology due to their ability to capture complex data manifolds which subsequently can be used to achieve better performance in downstream tasks, such as cancer type prediction or subtyping of cancer. However, these models are difficult to train due to the large number of hyperparameters that need to be tuned. To get a better understanding of the importance of the different hyperparameters, we examined six different VAE models when trained on TCGA transcriptomics data and evaluated on the downstream tasks of cluster agreement with cancer subtypes and survival analysis. We studied the effect of the latent space dimensionality, learning rate, optimizer, initialization and activation function on the quality of subsequent downstream tasks on the TCGA samples. We found β-TCVAE and DIP-VAE to have a good performance, on average, despite being more sensitive to hyperparameters selection. Based on these experiments, we derived recommendations for selecting the different hyperparameters settings. To ensure generalization, we tested all hyperparameter configurations on the GTEx dataset. We found a significant correlation (ρ = 0.7) between the hyperparameter effects on clustering performance in the TCGA and GTEx datasets. This highlights the robustness and generalizability of our recommendations. In addition, we examined whether the learned latent spaces capture biologically relevant information. Hereto, we measured the correlation and mutual information of the different representations with various data characteristics such as gender, age, days to metastasis, immune infiltration, and mutation signatures. We found that for all models the latent factors, in general, do not uniquely correlate with one of the data characteristics nor capture separable information in the latent factors even for models specifically designed for disentanglement.
| S-EPMC10553230 | biostudies-literature
Project description:The vast expansion of protein sequence databases provides an opportunity for new protein design approaches which seek to learn the sequence-function relationship directly from natural sequence variation. Deep generative models trained on protein sequence data have been shown to learn biologically meaningful representations helpful for a variety of downstream tasks, but their potential for direct use in the design of novel proteins remains largely unexplored. Here we show that variational autoencoders trained on a dataset of almost 70000 luciferase-like oxidoreductases can be used to generate novel, functional variants of the luxA bacterial luciferase. We propose separate VAE models to work with aligned sequence input (MSA VAE) and raw sequence input (AR-VAE), and offer evidence that while both are able to reproduce patterns of amino acid usage characteristic of the family, the MSA VAE is better able to capture long-distance dependencies reflecting the influence of 3D structure. To confirm the practical utility of the models, we used them to generate variants of luxA whose luminescence activity was validated experimentally. We further showed that conditional variants of both models could be used to increase the solubility of luxA without disrupting function. Altogether 6/12 of the variants generated using the unconditional AR-VAE and 9/11 generated using the unconditional MSA VAE retained measurable luminescence, together with all 23 of the less distant variants generated by conditional versions of the models; the most distant functional variant contained 35 differences relative to the nearest training set sequence. These results demonstrate the feasibility of using deep generative models to explore the space of possible protein sequences and generate useful variants, providing a method complementary to rational design and directed evolution approaches.
| S-EPMC7946179 | biostudies-literature
Project description:Gene expression regulation is highly dependent on binding of RNA-binding proteins (RBPs) to their RNA targets. Growing evidence supports the notion that both RNA primary sequence and its local secondary structure play a role in specific Protein-RNA recognition and binding. Despite the great advance in high-throughput experimental methods for identifying sequence targets of RBPs, predicting the specific sequence and structure binding preferences of RBPs remains a major challenge. We present a novel webserver, SMARTIV, designed for discovering and visualizing combined RNA sequence and structure motifs from high-throughput RNA-binding data, generated from in-vivo experiments. The uniqueness of SMARTIV is that it predicts motifs from enriched k-mers that combine information from ranked RNA sequences and their predicted secondary structure, obtained using various folding methods. Consequently, SMARTIV generates Position Weight Matrices (PWMs) in a combined sequence and structure alphabet with assigned P-values. SMARTIV concisely represents the sequence and structure motif content as a single graphical logo, which is informative and easy for visual perception. SMARTIV was examined extensively on a variety of high-throughput binding experiments for RBPs from different families, generated from different technologies, showing consistent and accurate results. Finally, SMARTIV is a user-friendly webserver, highly efficient in run-time and freely accessible via http://smartiv.technion.ac.il/.
| S-EPMC6030986 | biostudies-literature