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).     

MetAMOS: a modular and open source metagenomic assembly and analysis pipeline

We describe MetAMOS, an open source and modular metagenomic assembly and analysis pipeline. MetAMOS represents an important step towards fully automated metagenomic analysis, starting with next-generation sequencing reads and producing genomic scaffolds, open-reading frames and taxonomic or functional annotations. MetAMOS can aid in reducing assembly errors, commonly encountered when assembling metagenomic samples, and improves taxonomic assignment accuracy while also reducing computational cost. MetAMOS can be downloaded from: https://github.com/treangen/MetAMOS.     

An Algorithm to Automatically Generate the Combinatorial Orbit Counting Equations

Graphlets are small subgraphs, usually containing up to five vertices, that can be found in a larger graph. Identification of the graphlets that a vertex in an explored graph touches can provide useful information about the local structure of the graph around that vertex. Actually finding all graphlets in a large graph can be time-consuming, however. As the graphlets grow in size, more different graphlets emerge and the time needed to find each graphlet also scales up. If it is not needed to find each instance of each graphlet, but knowing the number of graphlets touching each node of the graph suffices, the problem is less hard. Previous research shows a way to simplify counting the graphlets: instead of looking for the graphlets needed, smaller graphlets are searched, as well as the number of common neighbors of vertices. Solving a system of equations then gives the number of times a vertex is part of each graphlet of the desired size. However, until now, equations only exist to count graphlets with 4 or 5 nodes. In this paper, two new techniques are presented. The first allows to generate the equations needed in an automatic way. This eliminates the tedious work needed to do so manually each time an extra node is added to the graphlets. The technique is independent on the number of nodes in the graphlets and can thus be used to count larger graphlets than previously possible. The second technique gives all graphlets a unique ordering which is easily extended to name graphlets of any size. Both techniques were used to generate equations to count graphlets with 4, 5 and 6 vertices, which extends all previous results. Code can be found at https://github.com/IneMelckenbeeck/equation-generator and https://github.com/IneMelckenbeeck/graphlet-naming.     

Enhancing breakpoint resolution with deep segmentation model: A general refinement method for read-depth based structural variant callers

Read-depths (RDs) are frequently used in identifying structural variants (SVs) from sequencing data. For existing RD-based SV callers, it is difficult for them to determine breakpoints in single-nucleotide resolution due to the noisiness of RD data and the bin-based calculation. In this paper, we propose to use the deep segmentation model UNet to learn base-wise RD patterns surrounding breakpoints of known SVs. We integrate model predictions with an RD-based SV caller to enhance breakpoints in single-nucleotide resolution. We show that UNet can be trained with a small amount of data and can be applied both in-sample and cross-sample. An enhancement pipeline named RDBKE significantly increases the number of SVs with more precise breakpoints on simulated and real data. The source code of RDBKE is freely available at https://github.com/yaozhong/deepIntraSV.     

Patient contrastive learning: A performant,expressive, and practical approach to electrocardiogram modeling

Supervised machine learning applications in health care are often limited due to a scarcity of labeled training data. To mitigate the effect of small sample size, we introduce a pre-training approach, Patient Contrastive Learning of Representations (PCLR), which creates latent representations of electrocardiograms (ECGs) from a large number of unlabeled examples using contrastive learning. The resulting representations are expressive, performant, and practical across a wide spectrum of clinical tasks. We develop PCLR using a large health care system with over 3.2 million 12-lead ECGs and demonstrate that training linear models on PCLR representations achieves a 51% performance increase, on average, over six training set sizes and four tasks (sex classification, age regression, and the detection of left ventricular hypertrophy and atrial fibrillation), relative to training neural network models from scratch. We also compared PCLR to three other ECG pre-training approaches (supervised pre-training, unsupervised pre-training with an autoencoder, and pre-training using a contrastive multi ECG-segment approach), and show significant performance benefits in three out of four tasks. We found an average performance benefit of 47% over the other models and an average of a 9% performance benefit compared to best model for each task. We release PCLR to enable others to extract ECG representations at https://github.com/broadinstitute/ml4h/tree/master/model_zoo/PCLR.     

lra: A long read aligner for sequences and contigs

It is computationally challenging to detect variation by aligning single-molecule sequencing (SMS) reads, or contigs from SMS assemblies. One approach to efficiently align SMS reads is sparse dynamic programming (SDP), where optimal chains of exact matches are found between the sequence and the genome. While straightforward implementations of SDP penalize gaps with a cost that is a linear function of gap length, biological variation is more accurately represented when gap cost is a concave function of gap length. We have developed a method, lra, that uses SDP with a concave-cost gap penalty, and used lra to align long-read sequences from PacBio and Oxford Nanopore (ONT) instruments as well as de novo assembly contigs. This alignment approach increases sensitivity and specificity for SV discovery, particularly for variants above 1kb and when discovering variation from ONT reads, while having runtime that are comparable (1.05-3.76×) to current methods. When applied to calling variation from de novo assembly contigs, there is a 3.2% increase in Truvari F1 score compared to minimap2+htsbox. lra is available in bioconda (https://anaconda.org/bioconda/lra) and github (https://github.com/ChaissonLab/LRA).     

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.     

CoCoNet—boosting RNA contact prediction by convolutional neural networks

Co-evolutionary models such as direct coupling analysis (DCA) in combination with machine learning (ML) techniques based on deep neural networks are able to predict accurate protein contact or distance maps. Such information can be used as constraints in structure prediction and massively increase prediction accuracy. Unfortunately, the same ML methods cannot readily be applied to RNA as they rely on large structural datasets only available for proteins. Here, we demonstrate how the available smaller data for RNA can be used to improve prediction of RNA contact maps. We introduce an algorithm called CoCoNet that is based on a combination of a Coevolutionary model and a shallow Convolutional Neural Network. Despite its simplicity and the small number of trained parameters, the method boosts the positive predictive value (PPV) of predicted contacts by about 70% with respect to DCA as tested by cross-validation of about eighty RNA structures. However, the direct inclusion of the CoCoNet contacts in 3D modeling tools does not result in a proportional increase of the 3D RNA structure prediction accuracy. Therefore, we suggest that the field develops, in addition to contact PPV, metrics which estimate the expected impact for 3D structure modeling tools better. CoCoNet is freely available and can be found at https://github.com/KIT-MBS/coconet.     

mirMark: a site-level and UTR-level classifier for miRNA target prediction

MiRNAs play important roles in many diseases including cancers. However computational prediction of miRNA target genes is challenging and the accuracies of existing methods remain poor. We report mirMark, a new machine learning-based method of miRNA target prediction at the site and UTR levels. This method uses experimentally verified miRNA targets from miRecords and mirTarBase as training sets and considers over 700 features. By combining Correlation-based Feature Selection with a variety of statistical or machine learning methods for the site- and UTR-level classifiers, mirMark significantly improves the overall predictive performance compared to existing publicly available methods. MirMark is available from https://github.com/lanagarmire/MirMark.

Electronic supplementary material

The online version of this article (doi:10.1186/s13059-014-0500-5) contains supplementary material, which is available to authorized users.     

cLoops2: a full-stack comprehensive analytical tool for chromatin interactions

Investigating chromatin interactions between regulatory regions such as enhancer and promoter elements is vital for understanding the regulation of gene expression. Compared to Hi-C and its variants, the emerging 3D mapping technologies focusing on enriched signals, such as TrAC-looping, reduce the sequencing cost and provide higher interaction resolution for cis-regulatory elements. A robust pipeline is needed for the comprehensive interpretation of these data, especially for loop-centric analysis. Therefore, we have developed a new versatile tool named cLoops2 for the full-stack analysis of these 3D chromatin interaction data. cLoops2 consists of core modules for peak-calling, loop-calling, differentially enriched loops calling and loops annotation. It also contains multiple modules for interaction resolution estimation, data similarity estimation, features quantification, feature aggregation analysis, and visualization. cLoops2 with documentation and example data are open source and freely available at GitHub: https://github.com/KejiZhaoLab/cLoops2.     

A Real-Time All-Atom Structural Search Engine for Proteins

Protein designers use a wide variety of software tools for de novo design, yet their repertoire still lacks a fast and interactive all-atom search engine. To solve this, we have built the Suns program: a real-time, atomic search engine integrated into the PyMOL molecular visualization system. Users build atomic-level structural search queries within PyMOL and receive a stream of search results aligned to their query within a few seconds. This instant feedback cycle enables a new “designability”-inspired approach to protein design where the designer searches for and interactively incorporates native-like fragments from proven protein structures. We demonstrate the use of Suns to interactively build protein motifs, tertiary interactions, and to identify scaffolds compatible with hot-spot residues. The official web site and installer are located at http://www.degradolab.org/suns/ and the source code is hosted at https://github.com/godotgildor/Suns (PyMOL plugin, BSD license), https://github.com/Gabriel439/suns-cmd (command line client, BSD license), and https://github.com/Gabriel439/suns-search (search engine server, GPLv2 license).

This is a PLOS Computational Biology Software Article

     

Ultra-large alignments using phylogeny-aware profiles

Many biological questions, including the estimation of deep evolutionary histories and the detection of remote homology between protein sequences, rely upon multiple sequence alignments and phylogenetic trees of large datasets. However, accurate large-scale multiple sequence alignment is very difficult, especially when the dataset contains fragmentary sequences. We present UPP, a multiple sequence alignment method that uses a new machine learning technique, the ensemble of hidden Markov models, which we propose here. UPP produces highly accurate alignments for both nucleotide and amino acid sequences, even on ultra-large datasets or datasets containing fragmentary sequences. UPP is available at https://github.com/smirarab/sepp.

Electronic supplementary material

The online version of this article (doi:10.1186/s13059-015-0688-z) contains supplementary material, which is available to authorized users.     

PhyloFisher: A phylogenomic package for resolving eukaryotic relationships

Phylogenomic analyses of hundreds of protein-coding genes aimed at resolving phylogenetic relationships is now a common practice. However, no software currently exists that includes tools for dataset construction and subsequent analysis with diverse validation strategies to assess robustness. Furthermore, there are no publicly available high-quality curated databases designed to assess deep (>100 million years) relationships in the tree of eukaryotes. To address these issues, we developed an easy-to-use software package, PhyloFisher (https://github.com/TheBrownLab/PhyloFisher), written in Python 3. PhyloFisher includes a manually curated database of 240 protein-coding genes from 304 eukaryotic taxa covering known eukaryotic diversity, a novel tool for ortholog selection, and utilities that will perform diverse analyses required by state-of-the-art phylogenomic investigations. Through phylogenetic reconstructions of the tree of eukaryotes and of the Saccharomycetaceae clade of budding yeasts, we demonstrate the utility of the PhyloFisher workflow and the provided starting database to address phylogenetic questions across a large range of evolutionary time points for diverse groups of organisms. We also demonstrate that undetected paralogy can remain in phylogenomic “single-copy orthogroup” datasets constructed using widely accepted methods such as all vs. all BLAST searches followed by Markov Cluster Algorithm (MCL) clustering and application of automated tree pruning algorithms. Finally, we show how the PhyloFisher workflow helps detect inadvertent paralog inclusions, allowing the user to make more informed decisions regarding orthology assignments, leading to a more accurate final dataset.

Phylogenomic analyses of hundreds of protein-coding genes aimed at resolving phylogenetic relationships is now a common practice. This article presents PhyloFisher, a community-driven tool for phylogenomic dataset construction to infer deep and shallow phylogenetic relationships among eukaryotes.     

AIME: Autoencoder-based integrative multi-omics data embedding that allows for confounder adjustments

In the integrative analyses of omics data, it is often of interest to extract data representation from one data type that best reflect its relations with another data type. This task is traditionally fulfilled by linear methods such as canonical correlation analysis (CCA) and partial least squares (PLS). However, information contained in one data type pertaining to the other data type may be complex and in nonlinear form. Deep learning provides a convenient alternative to extract low-dimensional nonlinear data embedding. In addition, the deep learning setup can naturally incorporate the effects of clinical confounding factors into the integrative analysis. Here we report a deep learning setup, named Autoencoder-based Integrative Multi-omics data Embedding (AIME), to extract data representation for omics data integrative analysis. The method can adjust for confounder variables, achieve informative data embedding, rank features in terms of their contributions, and find pairs of features from the two data types that are related to each other through the data embedding. In simulation studies, the method was highly effective in the extraction of major contributing features between data types. Using two real microRNA-gene expression datasets, one with confounder variables and one without, we show that AIME excluded the influence of confounders, and extracted biologically plausible novel information. The R package based on Keras and the TensorFlow backend is available at https://github.com/tianwei-yu/AIME.     

MicrobeTrace: Retooling molecular epidemiology for rapid public health response

Outbreak investigations use data from interviews, healthcare providers, laboratories and surveillance systems. However, integrated use of data from multiple sources requires a patchwork of software that present challenges in usability, interoperability, confidentiality, and cost. Rapid integration, visualization and analysis of data from multiple sources can guide effective public health interventions. We developed MicrobeTrace to facilitate rapid public health responses by overcoming barriers to data integration and exploration in molecular epidemiology. MicrobeTrace is a web-based, client-side, JavaScript application (https://microbetrace.cdc.gov) that runs in Chromium-based browsers and remains fully operational without an internet connection. Using publicly available data, we demonstrate the analysis of viral genetic distance networks and introduce a novel approach to minimum spanning trees that simplifies results. We also illustrate the potential utility of MicrobeTrace in support of contact tracing by analyzing and displaying data from an outbreak of SARS-CoV-2 in South Korea in early 2020. MicrobeTrace is developed and actively maintained by the Centers for Disease Control and Prevention. Users can email vog.cdc@ecarteborcim for support. The source code is available at https://github.com/cdcgov/microbetrace.     

CHANCE: comprehensive software for quality control and validation of ChIP-seq data

ChIP-seq is a powerful method for obtaining genome-wide maps of protein-DNA interactions and epigenetic modifications. CHANCE (CHip-seq ANalytics and Confidence Estimation) is a standalone package for ChIP-seq quality control and protocol optimization. Our user-friendly graphical software quickly estimates the strength and quality of immunoprecipitations, identifies biases, compares the user''s data with ENCODE''s large collection of published datasets, performs multi-sample normalization, checks against quantitative PCR-validated control regions, and produces informative graphical reports. CHANCE is available at https://github.com/songlab/chance.