Learning,visualizing and exploring 16S rRNA structure using an attention-based deep neural network

Recurrent neural networks with memory and attention mechanisms are widely used in natural language processing because they can capture short and long term sequential information for diverse tasks. We propose an integrated deep learning model for microbial DNA sequence data, which exploits convolutional neural networks, recurrent neural networks, and attention mechanisms to predict taxonomic classifications and sample-associated attributes, such as the relationship between the microbiome and host phenotype, on the read/sequence level. In this paper, we develop this novel deep learning approach and evaluate its application to amplicon sequences. We apply our approach to short DNA reads and full sequences of 16S ribosomal RNA (rRNA) marker genes, which identify the heterogeneity of a microbial community sample. We demonstrate that our implementation of a novel attention-based deep network architecture, Read2Pheno, achieves read-level phenotypic prediction. Training Read2Pheno models will encode sequences (reads) into dense, meaningful representations: learned embedded vectors output from the intermediate layer of the network model, which can provide biological insight when visualized. The attention layer of Read2Pheno models can also automatically identify nucleotide regions in reads/sequences which are particularly informative for classification. As such, this novel approach can avoid pre/post-processing and manual interpretation required with conventional approaches to microbiome sequence classification. We further show, as proof-of-concept, that aggregating read-level information can robustly predict microbial community properties, host phenotype, and taxonomic classification, with performance at least comparable to conventional approaches. An implementation of the attention-based deep learning network is available at https://github.com/EESI/sequence_attention (a python package) and https://github.com/EESI/seq2att (a command line tool).     

Identifying ribosome heterogeneity using ribosome profiling

Recent studies have revealed multiple mechanisms that can lead to heterogeneity in ribosomal composition. This heterogeneity can lead to preferential translation of specific panels of mRNAs, and is defined in large part by the ribosomal protein (RP) content, amongst other things. However, it is currently unknown to what extent ribosomal composition is heterogeneous across tissues, which is compounded by a lack of tools available to study it. Here we present dripARF, a method for detecting differential RP incorporation into the ribosome using Ribosome Profiling (Ribo-seq) data. We combine the ‘waste’ rRNA fragment data generated in Ribo-seq with the known 3D structure of the human ribosome to predict differences in the composition of ribosomes in the material being studied. We have validated this approach using publicly available data, and have revealed a potential role for eS25/RPS25 in development. Our results indicate that ribosome heterogeneity can be detected in Ribo-seq data, providing a new method to study this phenomenon. Furthermore, with dripARF, previously published Ribo-seq data provides a wealth of new information, allowing the identification of RPs of interest in many disease and normal contexts. dripARF is available as part of the ARF R package and can be accessed through https://github.com/fallerlab/ARF.     

RNA Thermodynamic Structural Entropy

Conformational entropy for atomic-level, three dimensional biomolecules is known experimentally to play an important role in protein-ligand discrimination, yet reliable computation of entropy remains a difficult problem. Here we describe the first two accurate and efficient algorithms to compute the conformational entropy for RNA secondary structures, with respect to the Turner energy model, where free energy parameters are determined from UV absorption experiments. An algorithm to compute the derivational entropy for RNA secondary structures had previously been introduced, using stochastic context free grammars (SCFGs). However, the numerical value of derivational entropy depends heavily on the chosen context free grammar and on the training set used to estimate rule probabilities. Using data from the Rfam database, we determine that both of our thermodynamic methods, which agree in numerical value, are substantially faster than the SCFG method. Thermodynamic structural entropy is much smaller than derivational entropy, and the correlation between length-normalized thermodynamic entropy and derivational entropy is moderately weak to poor. In applications, we plot the structural entropy as a function of temperature for known thermoswitches, such as the repression of heat shock gene expression (ROSE) element, we determine that the correlation between hammerhead ribozyme cleavage activity and total free energy is improved by including an additional free energy term arising from conformational entropy, and we plot the structural entropy of windows of the HIV-1 genome. Our software RNAentropy can compute structural entropy for any user-specified temperature, and supports both the Turner’99 and Turner’04 energy parameters. It follows that RNAentropy is state-of-the-art software to compute RNA secondary structure conformational entropy. Source code is available at https://github.com/clotelab/RNAentropy/; a full web server is available at http://bioinformatics.bc.edu/clotelab/RNAentropy, including source code and ancillary programs.     

Real-time resolution of short-read assembly graph using ONT long reads

A streaming assembly pipeline utilising real-time Oxford Nanopore Technology (ONT) sequencing data is important for saving sequencing resources and reducing time-to-result. A previous approach implemented in npScarf provided an efficient streaming algorithm for hybrid assembly but was relatively prone to mis-assemblies compared to other graph-based methods. Here we present npGraph, a streaming hybrid assembly tool using the assembly graph instead of the separated pre-assembly contigs. It is able to produce more complete genome assembly by resolving the path finding problem on the assembly graph using long reads as the traversing guide. Application to synthetic and real data from bacterial isolate genomes show improved accuracy while still maintaining a low computational cost. npGraph also provides a graphical user interface (GUI) which provides a real-time visualisation of the progress of assembly. The tool and source code is available at https://github.com/hsnguyen/assembly.     

Joint Inference of Migration and Reassortment Patterns for Viruses with Segmented Genomes

The structured coalescent allows inferring migration patterns between viral subpopulations from genetic sequence data. However, these analyses typically assume that no genetic recombination process impacted the sequence evolution of pathogens. For segmented viruses, such as influenza, that can undergo reassortment this assumption is broken. Reassortment reshuffles the segments of different parent lineages upon a coinfection event, which means that the shared history of viruses has to be represented by a network instead of a tree. Therefore, full genome analyses of such viruses are complex or even impossible. Although this problem has been addressed for unstructured populations, it is still impossible to account for population structure, such as induced by different host populations, whereas also accounting for reassortment. We address this by extending the structured coalescent to account for reassortment and present a framework for investigating possible ties between reassortment and migration (host jump) events. This method can accurately estimate subpopulation dependent effective populations sizes, reassortment, and migration rates from simulated data. Additionally, we apply the new model to avian influenza A/H5N1 sequences, sampled from two avian host types, Anseriformes and Galliformes. We contrast our results with a structured coalescent without reassortment inference, which assumes independently evolving segments. This reveals that taking into account segment reassortment and using sequencing data from several viral segments for joint phylodynamic inference leads to different estimates for effective population sizes, migration, and clock rates. This new model is implemented as the Structured Coalescent with Reassortment package for BEAST 2.5 and is available at https://github.com/jugne/SCORE.     

Hidden Markov Modeling with HMMTeacher

Is it possible to learn and create a first Hidden Markov Model (HMM) without programming skills or understanding the algorithms in detail? In this concise tutorial, we present the HMM through the 2 general questions it was initially developed to answer and describe its elements. The HMM elements include variables, hidden and observed parameters, the vector of initial probabilities, and the transition and emission probability matrices. Then, we suggest a set of ordered steps, for modeling the variables and illustrate them with a simple exercise of modeling and predicting transmembrane segments in a protein sequence. Finally, we show how to interpret the results of the algorithms for this particular problem. To guide the process of information input and explicit solution of the basic HMM algorithms that answer the HMM questions posed, we developed an educational webserver called HMMTeacher. Additional solved HMM modeling exercises can be found in the user’s manual and answers to frequently asked questions. HMMTeacher is available at https://hmmteacher.mobilomics.org, mirrored at https://hmmteacher1.mobilomics.org. A repository with the code of the tool and the webpage is available at https://gitlab.com/kmilo.f/hmmteacher.     

HAlign 3: Fast Multiple Alignment of Ultra-Large Numbers of Similar DNA/RNA Sequences

HAlign is a cross-platform program that performs multiple sequence alignments based on the center star strategy. Here we present two major updates of HAlign 3, which helped improve the time efficiency and the alignment quality, and made HAlign 3 a specialized program to process ultra-large numbers of similar DNA/RNA sequences, such as closely related viral or prokaryotic genomes. HAlign 3 can be easily installed via the Anaconda and Java release package on macOS, Linux, Windows subsystem for Linux, and Windows systems, and the source code is available on GitHub (https://github.com/malabz/HAlign-3).     

UniqTag: Content-Derived Unique and Stable Identifiers for Gene Annotation

When working on an ongoing genome sequencing and assembly project, it is rather inconvenient when gene identifiers change from one build of the assembly to the next. The gene labelling system described here, UniqTag, addresses this common challenge. UniqTag assigns a unique identifier to each gene that is a representative k-mer, a string of length k, selected from the sequence of that gene. Unlike serial numbers, these identifiers are stable between different assemblies and annotations of the same data without requiring that previous annotations be lifted over by sequence alignment. We assign UniqTag identifiers to ten builds of the Ensembl human genome spanning eight years to demonstrate this stability. The implementation of UniqTag in Ruby and an R package are available at https://github.com/sjackman/uniqtag sjackman/uniqtag. The R package is also available from CRAN: install.packages ("uniqtag"). Supplementary material and code to reproduce it is available at https://github.com/sjackman/uniqtag-paper.     

MacSyFinder: A Program to Mine Genomes for Molecular Systems with an Application to CRISPR-Cas Systems

Motivation

Biologists often wish to use their knowledge on a few experimental models of a given molecular system to identify homologs in genomic data. We developed a generic tool for this purpose.

Results

Macromolecular System Finder (MacSyFinder) provides a flexible framework to model the properties of molecular systems (cellular machinery or pathway) including their components, evolutionary associations with other systems and genetic architecture. Modelled features also include functional analogs, and the multiple uses of a same component by different systems. Models are used to search for molecular systems in complete genomes or in unstructured data like metagenomes. The components of the systems are searched by sequence similarity using Hidden Markov model (HMM) protein profiles. The assignment of hits to a given system is decided based on compliance with the content and organization of the system model. A graphical interface, MacSyView, facilitates the analysis of the results by showing overviews of component content and genomic context. To exemplify the use of MacSyFinder we built models to detect and class CRISPR-Cas systems following a previously established classification. We show that MacSyFinder allows to easily define an accurate “Cas-finder” using publicly available protein profiles.

Availability and Implementation

MacSyFinder is a standalone application implemented in Python. It requires Python 2.7, Hmmer and makeblastdb (version 2.2.28 or higher). It is freely available with its source code under a GPLv3 license at https://github.com/gem-pasteur/macsyfinder. It is compatible with all platforms supporting Python and Hmmer/makeblastdb. The “Cas-finder” (models and HMM profiles) is distributed as a compressed tarball archive as Supporting Information.     

Memes: A motif analysis environment in R using tools from the MEME Suite

Identification of biopolymer motifs represents a key step in the analysis of biological sequences. The MEME Suite is a widely used toolkit for comprehensive analysis of biopolymer motifs; however, these tools are poorly integrated within popular analysis frameworks like the R/Bioconductor project, creating barriers to their use. Here we present memes, an R package that provides a seamless R interface to a selection of popular MEME Suite tools. memes provides a novel “data aware” interface to these tools, enabling rapid and complex discriminative motif analysis workflows. In addition to interfacing with popular MEME Suite tools, memes leverages existing R/Bioconductor data structures to store the multidimensional data returned by MEME Suite tools for rapid data access and manipulation. Finally, memes provides data visualization capabilities to facilitate communication of results. memes is available as a Bioconductor package at https://bioconductor.org/packages/memes, and the source code can be found at github.com/snystrom/memes.     

A zero inflated log-normal model for inference of sparse microbial association networks

The advent of high-throughput metagenomic sequencing has prompted the development of efficient taxonomic profiling methods allowing to measure the presence, abundance and phylogeny of organisms in a wide range of environmental samples. Multivariate sequence-derived abundance data further has the potential to enable inference of ecological associations between microbial populations, but several technical issues need to be accounted for, like the compositional nature of the data, its extreme sparsity and overdispersion, as well as the frequent need to operate in under-determined regimes.The ecological network reconstruction problem is frequently cast into the paradigm of Gaussian Graphical Models (GGMs) for which efficient structure inference algorithms are available, like the graphical lasso and neighborhood selection. Unfortunately, GGMs or variants thereof can not properly account for the extremely sparse patterns occurring in real-world metagenomic taxonomic profiles. In particular, structural zeros (as opposed to sampling zeros) corresponding to true absences of biological signals fail to be properly handled by most statistical methods.We present here a zero-inflated log-normal graphical model (available at https://github.com/vincentprost/Zi-LN) specifically aimed at handling such “biological” zeros, and demonstrate significant performance gains over state-of-the-art statistical methods for the inference of microbial association networks, with most notable gains obtained when analyzing taxonomic profiles displaying sparsity levels on par with real-world metagenomic datasets.     

Detecting DNA Modifications from SMRT Sequencing Data by Modeling Sequence Context Dependence of Polymerase Kinetic

DNA modifications such as methylation and DNA damage can play critical regulatory roles in biological systems. Single molecule, real time (SMRT) sequencing technology generates DNA sequences as well as DNA polymerase kinetic information that can be used for the direct detection of DNA modifications. We demonstrate that local sequence context has a strong impact on DNA polymerase kinetics in the neighborhood of the incorporation site during the DNA synthesis reaction, allowing for the possibility of estimating the expected kinetic rate of the enzyme at the incorporation site using kinetic rate information collected from existing SMRT sequencing data (historical data) covering the same local sequence contexts of interest. We develop an Empirical Bayesian hierarchical model for incorporating historical data. Our results show that the model could greatly increase DNA modification detection accuracy, and reduce requirement of control data coverage. For some DNA modifications that have a strong signal, a control sample is not even needed by using historical data as alternative to control. Thus, sequencing costs can be greatly reduced by using the model. We implemented the model in a R package named seqPatch, which is available at https://github.com/zhixingfeng/seqPatch.     

Seeker: alignment-free identification of bacteriophage genomes by deep learning

Recent advances in metagenomic sequencing have enabled discovery of diverse, distinct microbes and viruses. Bacteriophages, the most abundant biological entity on Earth, evolve rapidly, and therefore, detection of unknown bacteriophages in sequence datasets is a challenge. Most of the existing detection methods rely on sequence similarity to known bacteriophage sequences, impeding the identification and characterization of distinct, highly divergent bacteriophage families. Here we present Seeker, a deep-learning tool for alignment-free identification of phage sequences. Seeker allows rapid detection of phages in sequence datasets and differentiation of phage sequences from bacterial ones, even when those phages exhibit little sequence similarity to established phage families. We comprehensively validate Seeker''s ability to identify previously unidentified phages, and employ this method to detect unknown phages, some of which are highly divergent from the known phage families. We provide a web portal (seeker.pythonanywhere.com) and a user-friendly Python package (github.com/gussow/seeker) allowing researchers to easily apply Seeker in metagenomic studies, for the detection of diverse unknown bacteriophages.     

tidy tree: A New Layout for Phylogenetic Trees

Many layouts exist for visualizing phylogenetic trees, allowing to display the same information (evolutionary relationships) in different ways. For large phylogenies, the choice of the layout is a key element, because the printable area is limited, and because interactive on-screen visualizers can lead to unreadable phylogenetic relationships at high zoom levels. A visual inspection of available layouts for rooted trees reveals large empty areas that one may want to fill in order to use less drawing space and eventually gain readability. This can be achieved by using the nonlayered tidy tree layout algorithm that was proposed earlier but was never used in a phylogenetic context so far. Here, we present its implementation, and we demonstrate its advantages on simulated and biological data (the measles virus phylogeny). Our results call for the integration of this new layout in phylogenetic software. We implemented the nonlayered tidy tree layout in R language as a stand-alone function (available at https://github.com/damiendevienne/non-layered-tidy-trees), as an option in the tree plotting function of the R package ape, and in the recent tool for visualizing reconciled phylogenetic trees thirdkind (https://github.com/simonpenel/thirdkind/wiki).     

Recursive MAGUS: Scalable and accurate multiple sequence alignment

Multiple sequence alignment tools struggle to keep pace with rapidly growing sequence data, as few methods can handle large datasets while maintaining alignment accuracy. We recently introduced MAGUS, a new state-of-the-art method for aligning large numbers of sequences. In this paper, we present a comprehensive set of enhancements that allow MAGUS to align vastly larger datasets with greater speed. We compare MAGUS to other leading alignment methods on datasets of up to one million sequences. Our results demonstrate the advantages of MAGUS over other alignment software in both accuracy and speed. MAGUS is freely available in open-source form at https://github.com/vlasmirnov/MAGUS.