Efficient DNA sequence compression with neural networks

Generating 2D Barcode for DNA Barcode Sequences

DNA barcode sequence is a short DNA sequence representing a sample from a particular species. The commonly used DNA barcodes are at least 200 bps long. This large number of characters cannot be encoded in two-dimensional codes for sample recognition and tracking. In the present study, we described a method that can be used to compress the DNA sequences and then generate the corresponding QR code. With the large numbers of software and hardware, the QR code can be used efficiently for printing, labeling, and scanning.

AGC: compact representation of assembled genomes with fast queries and updates

MOTIVATION

High-quality sequence assembly is the ultimate representation of complete genetic information of an individual. Several ongoing pangenome projects are producing collections of high-quality assemblies of various species. Each project has already generated assemblies of hundreds of gigabytes on disk, greatly impeding the distribution of and access to such rich datasets.

RESULTS

Here, we show how to reduce the size of the sequenced genomes by 2-3 orders of magnitude. Our tool compresses the genomes significantly better than the existing programs and is much faster. Moreover, its unique feature is the ability to access any contig (or its part) in a fraction of a second and easily append new samples to the compressed collections. Thanks to this, AGC could be useful not only for backup or transfer purposes but also for routine analysis of pangenome sequences in common pipelines. With the rapidly reduced cost and improved accuracy of sequencing technologies, we anticipate more comprehensive pangenome projects with much larger sample sizes. AGC is likely to become a foundation tool to store, distribute and access pangenome data.

AVAILABILITY AND IMPLEMENTATION

The source code of AGC is available at https://github.com/refresh-bio/agc. The package can be installed via Bioconda at https://anaconda.org/bioconda/agc.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Deep Learning in Population Genetics

Population genetics is transitioning into a data-driven discipline thanks to the availability of large-scale genomic data and the need to study increasingly complex evolutionary scenarios. With likelihood and Bayesian approaches becoming either intractable or computationally unfeasible, machine learning, and in particular deep learning, algorithms are emerging as popular techniques for population genetic inferences. These approaches rely on algorithms that learn non-linear relationships between the input data and the model parameters being estimated through representation learning from training data sets. Deep learning algorithms currently employed in the field comprise discriminative and generative models with fully connected, convolutional, or recurrent layers. Additionally, a wide range of powerful simulators to generate training data under complex scenarios are now available. The application of deep learning to empirical data sets mostly replicates previous findings of demography reconstruction and signals of natural selection in model organisms. To showcase the feasibility of deep learning to tackle new challenges, we designed a branched architecture to detect signals of recent balancing selection from temporal haplotypic data, which exhibited good predictive performance on simulated data. Investigations on the interpretability of neural networks, their robustness to uncertain training data, and creative representation of population genetic data, will provide further opportunities for technological advancements in the field.

DDQR (dynamic DNA QR coding): An efficient algorithm to represent DNA barcode sequences

A DNA barcode is a short piece of standard DNA sequence used for species determination and discrimination. Representation of DNA barcodes is essential for DNA barcodes' applications in the transportation and recognition of biological materials. Previously, we have compared different strategies for representing the DNA barcodes. In the present study, we have developed a compression algorithm based on binary coding or Huffman coding scheme, followed by converting the binary digits into Base64 digits. The combination of this compression algorithm and the QR representation leads to the dynamic DNA QR coding algorithm (DDQR). We tested the DDQR algorithm on simulated data and real DNA barcode sequences from the commonly used plant and animal DNA barcode markers: rbcL, matK, trnH-psbA, ITS2, and COI. We compared the compression efficiency of DDQR and another state-of-the-art DNA compression algorithm GeCo3 for sequences with various base compositions and lengths. We found that DDQR had a higher compression rate than GeCo3 for DNA sequences shorter than 800 bp, which is the typical size range for DNA barcodes. We also upgraded a web server (http://www.1kmpg.cn/ddqr) that provides three functions: retrieval of DNA barcode sequences, encoding DNA barcode sequences to DDQR codes, and decoding DDQR codes to DNA barcode sequences. The DDQR algorithm and the webserver will be invaluable to applying DNA barcode technology in the food and traditional medicine industries.

AlcoR: alignment-free simulation, mapping, and visualization of low-complexity regions in biological data

BACKGROUND

Low-complexity data analysis is the area that addresses the search and quantification of regions in sequences of elements that contain low-complexity or repetitive elements. For example, these can be tandem repeats, inverted repeats, homopolymer tails, GC-biased regions, similar genes, and hairpins, among many others. Identifying these regions is crucial because of their association with regulatory and structural characteristics. Moreover, their identification provides positional and quantity information where standard assembly methodologies face significant difficulties because of substantial higher depth coverage (mountains), ambiguous read mapping, or where sequencing or reconstruction defects may occur. However, the capability to distinguish low-complexity regions (LCRs) in genomic and proteomic sequences is a challenge that depends on the model's ability to find them automatically. Low-complexity patterns can be implicit through specific or combined sources, such as algorithmic or probabilistic, and recurring to different spatial distances-namely, local, medium, or distant associations.

FINDINGS

This article addresses the challenge of automatically modeling and distinguishing LCRs, providing a new method and tool (AlcoR) for efficient and accurate segmentation and visualization of these regions in genomic and proteomic sequences. The method enables the use of models with different memories, providing the ability to distinguish local from distant low-complexity patterns. The method is reference and alignment free, providing additional methodologies for testing, including a highly flexible simulation method for generating biological sequences (DNA or protein) with different complexity levels, sequence masking, and a visualization tool for automatic computation of the LCR maps into an ideogram style. We provide illustrative demonstrations using synthetic, nearly synthetic, and natural sequences showing the high efficiency and accuracy of AlcoR. As large-scale results, we use AlcoR to unprecedentedly provide a whole-chromosome low-complexity map of a recent complete human genome and the haplotype-resolved chromosome pairs of a heterozygous diploid African cassava cultivar.

CONCLUSIONS

The AlcoR method provides the ability of fast sequence characterization through data complexity analysis, ideally for scenarios entangling the presence of new or unknown sequences. AlcoR is implemented in C language using multithreading to increase the computational speed, is flexible for multiple applications, and does not contain external dependencies. The tool accepts any sequence in FASTA format. The source code is freely provided at https://github.com/cobilab/alcor.

The complexity landscape of viral genomes

BACKGROUND

Viruses are among the shortest yet highly abundant species that harbor minimal instructions to infect cells, adapt, multiply, and exist. However, with the current substantial availability of viral genome sequences, the scientific repertory lacks a complexity landscape that automatically enlights viral genomes' organization, relation, and fundamental characteristics.

RESULTS

This work provides a comprehensive landscape of the viral genome's complexity (or quantity of information), identifying the most redundant and complex groups regarding their genome sequence while providing their distribution and characteristics at a large and local scale. Moreover, we identify and quantify inverted repeats abundance in viral genomes. For this purpose, we measure the sequence complexity of each available viral genome using data compression, demonstrating that adequate data compressors can efficiently quantify the complexity of viral genome sequences, including subsequences better represented by algorithmic sources (e.g., inverted repeats). Using a state-of-the-art genomic compressor on an extensive viral genomes database, we show that double-stranded DNA viruses are, on average, the most redundant viruses while single-stranded DNA viruses are the least. Contrarily, double-stranded RNA viruses show a lower redundancy relative to single-stranded RNA. Furthermore, we extend the ability of data compressors to quantify local complexity (or information content) in viral genomes using complexity profiles, unprecedently providing a direct complexity analysis of human herpesviruses. We also conceive a features-based classification methodology that can accurately distinguish viral genomes at different taxonomic levels without direct comparisons between sequences. This methodology combines data compression with simple measures such as GC-content percentage and sequence length, followed by machine learning classifiers.

CONCLUSIONS

This article presents methodologies and findings that are highly relevant for understanding the patterns of similarity and singularity between viral groups, opening new frontiers for studying viral genomes' organization while depicting the complexity trends and classification components of these genomes at different taxonomic levels. The whole study is supported by an extensive website (https://asilab.github.io/canvas/) for comprehending the viral genome characterization using dynamic and interactive approaches.

Comparative studies on the high-performance compression of SARS-CoV-2 genome collections

The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is fast mutating worldwide. The mutated strains have been timely sequenced by worldwide labs, accumulating a huge amount of viral genome sequences open to public for biomedicine research such as mRNA vaccine design and drug recommendation. It is inefficient to transmit the millions of genome sequences without compression. In this study, we benchmark the performance of reference-free and reference-based compression algorithms on SARS-CoV-2 genome collections extracted from NCBI. Experimental results show that reference-based two-level compression is the most suitable approach to the compression, achieving the best compression ratio 1019.33-fold for compressing 132 372 genomes and 949.73-fold for compressing 416 238 genomes. This enormous file size reduction and efficient decompression have enabled a 5-min download and decompression of $10^5$ SARS-CoV-2 genomes. As compression on datasets containing such big numbers of genomes has been explored seldom before, our comparative analysis of the state-of-the-art compression algorithms provides practical guidance for the selection of compression tools and their parameters such as reference genomes to compress viral genome databases with similar characteristics. We also suggested a genome clustering approach using multiple references for a better compression. It is anticipated that the increased availability of SARS-CoV-2 genome datasets will make biomedicine research more productive.

MBGC: Multiple Bacteria Genome Compressor

BACKGROUND

Genomes within the same species reveal large similarity, exploited by specialized multiple genome compressors. The existing algorithms and tools are however targeted at large, e.g., mammalian, genomes, and their performance on bacteria strains is rather moderate.

RESULTS

In this work, we propose MBGC, a specialized genome compressor making use of specific redundancy of bacterial genomes. Its characteristic features are finding both direct and reverse-complemented LZ-matches, as well as a careful management of a reference buffer in a multi-threaded implementation. Our tool is not only compression efficient but also fast. On a collection of 168,311 bacterial genomes, totalling 587 GB, we achieve a compression ratio of approximately a factor of 1,265 and compression (respectively decompression) speed of ∼1,580 MB/s (respectively 780 MB/s) using 8 hardware threads, on a computer with a 14-core/28-thread CPU and a fast SSD, being almost 3 times more succinct and >6 times faster in the compression than the next best competitor.

The visual story of data storage: From storage properties to user interfaces

About fifty times more data has been created than there are stars in the observable universe. Current trends in data creation and consumption mean that the devices and storage media we use will require more physical space. Novel data storage media such as DNA are considered a viable alternative. Yet, the introduction of new storage technologies should be accompanied by an evaluation of user requirements. To assess such needs, we designed and conducted a survey to rank different storage properties adapted for visualization. That is, accessibility, capacity, usage, mutability, lifespan, addressability, and typology. Withal, we reported different storage devices over time while ranking them by their properties. Our results indicated a timeline of three distinct periods: magnetic, optical and electronic, and alternative media. Moreover, by investigating user interfaces across different operating systems, we observed a predominant presence of bar charts and tree maps for the usage of a medium and its file directory hierarchy, respectively. Taken together with the results of our survey, this allowed us to create a customized user interface that includes data visualizations that can be toggled for both user groups: Experts and Public.

AC2: An Efficient Protein Sequence Compression Tool Using Artificial Neural Networks and Cache-Hash Models

Recently, the scientific community has witnessed a substantial increase in the generation of protein sequence data, triggering emergent challenges of increasing importance, namely efficient storage and improved data analysis. For both applications, data compression is a straightforward solution. However, in the literature, the number of specific protein sequence compressors is relatively low. Moreover, these specialized compressors marginally improve the compression ratio over the best general-purpose compressors. In this paper, we present AC2, a new lossless data compressor for protein (or amino acid) sequences. AC2 uses a neural network to mix experts with a stacked generalization approach and individual cache-hash memory models to the highest-context orders. Compared to the previous compressor (AC), we show gains of 2–9% and 6–7% in reference-free and reference-based modes, respectively. These gains come at the cost of three times slower computations. AC2 also improves memory usage against AC, with requirements about seven times lower, without being affected by the sequences’ input size. As an analysis application, we use AC2 to measure the similarity between each SARS-CoV-2 protein sequence with each viral protein sequence from the whole UniProt database. The results consistently show higher similarity to the pangolin coronavirus, followed by the bat and human coronaviruses, contributing with critical results to a current controversial subject. AC2 is available for free download under GPLv3 license.

Efficient DNA sequence compression with neural networks

BACKGROUND

The increasing production of genomic data has led to an intensified need for models that can cope efficiently with the lossless compression of DNA sequences. Important applications include long-term storage and compression-based data analysis. In the literature, only a few recent articles propose the use of neural networks for DNA sequence compression. However, they fall short when compared with specific DNA compression tools, such as GeCo2. This limitation is due to the absence of models specifically designed for DNA sequences. In this work, we combine the power of neural networks with specific DNA models. For this purpose, we created GeCo3, a new genomic sequence compressor that uses neural networks for mixing multiple context and substitution-tolerant context models.

FINDINGS

We benchmark GeCo3 as a reference-free DNA compressor in 5 datasets, including a balanced and comprehensive dataset of DNA sequences, the Y-chromosome and human mitogenome, 2 compilations of archaeal and virus genomes, 4 whole genomes, and 2 collections of FASTQ data of a human virome and ancient DNA. GeCo3 achieves a solid improvement in compression over the previous version (GeCo2) of $2.4\%$, $7.1\%$, $6.1\%$, $5.8\%$, and $6.0\%$, respectively. To test its performance as a reference-based DNA compressor, we benchmark GeCo3 in 4 datasets constituted by the pairwise compression of the chromosomes of the genomes of several primates. GeCo3 improves the compression in $12.4\%$, $11.7\%$, $10.8\%$, and $10.1\%$ over the state of the art. The cost of this compression improvement is some additional computational time (1.7-3 times slower than GeCo2). The RAM use is constant, and the tool scales efficiently, independently of the sequence size. Overall, these values outperform the state of the art.

CONCLUSIONS

GeCo3 is a genomic sequence compressor with a neural network mixing approach that provides additional gains over top specific genomic compressors. The proposed mixing method is portable, requiring only the probabilities of the models as inputs, providing easy adaptation to other data compressors or compression-based data analysis tools. GeCo3 is released under GPLv3 and is available for free download at https://github.com/cobilab/geco3.