Advances in optical mapping for genomic research

Abstract

Recent advances in optical mapping have allowed the construction of improved genome assemblies with greater contiguity. Optical mapping also enables genome comparison and identification of large-scale structural variations. Association of these large-scale genomic features with biological functions is an important goal in plant and animal breeding and in medical research. Optical mapping has also been used in microbiology and still plays an important role in strain typing and epidemiological studies. Here, we review the development of optical mapping in recent decades to illustrate its importance in genomic research. We detail its applications and algorithms to show its specific advantages. Finally, we discuss the challenges required to facilitate the optimization of optical mapping and improve its future development and application.

1. Introduction

Short-read sequencing (SRS) has made genomic and genetic studies in various species affordable [1], [2], but the limited sequence lengths generated by SRS platforms prevent the reads from spanning most repetitive and complex regions, leading to fragmented and collapsed genome assemblies [2], [3]. The use of assemblies based on SRS can therefore reduce the accuracy of downstream analyses, such as those used for the detection of genomic variations [1], [4]. Although different methods and computational algorithms have been developed to solve this problem [5], [6], [7], [8], [9], it is difficult to fully overcome the deficiencies inherited from the short read length.

Long-read sequencing provided by Pacific Biosciences (PacBio) and Oxford Nanopore technologies now help to cover most repeats and produces more complete genome assemblies [10], [11], [12], [13]. However, the read lengths (~15 kb on average) are still insufficient to cover some large repetitive and complex genomic regions [14], [15], thus hampering biological studies of these regions. In addition, while PacBio or Oxford Nanopore sequencing can produce substantial long-sequence reads for prokaryote genome assembly, for eukaryotic genomes, the reads can generally only be assembled to the scaffold-level, but not to the chromosome-level [16].

Short-read long-insert technologies, linked-read sequencing technologies, high-throughput chromosome conformation capture (Hi-C) technologies and optical mapping are used to solve these assembly problems [10], [17]. Extended from the concept of paired-end reads to contain a fixed length insertion between two sequences at the ends of a DNA fragment, the mate-pair and similar methods construct libraries joining ends of circularized longer fragments (2–5 kb) in single reads to resolve adjacent sequences at longer intervals. Similarly, linked-read sequencing technologies, such as those provided by 10x Genomics, and Hi-C technologies such as those provided by Dovetail Genomics and Phase Genomics, use fluidics and other ligation methods to capture proximity information. The variations in library preparation ultimately rely on short-read sequencing [17]. Although the cost of these technologies is relatively low, sequencing biases, such as those caused by PCR amplification and enzyme selection [18], [19], can make these technologies error-prone [10]. Moreover, Burton et al. [20] and Bickhart et al. [21] found that Hi-C technologies can lead to mis-assembly, such as false inversion and scaffold misplacement, which has to be corrected with data from orthogonal methods such as radiation hybrid maps. Compared to single-molecule technologies, linked reads have coverage drawbacks as they cannot fully cover the source DNA fragments. When using linked reads, they can introduce gaps particularly during eukaryotic genome assembly [22]. Additionally, the current linked-read sequencing methods have a limited ability to process polyploid genomes [23], [24].

In contrast to the above technologies used to capture linkage information or to determine 3D genome architecture, optical mapping uses a light microscope-based technique to physically locate specific enzymes or sequence motifs to produce DNA sequence fingerprints [25]. The resulting optical maps contain only the physical locations of selected enzymes, rather than base-by-base nucleotide information. Specifically, during the generation of optical maps, selected enzymes are used to fluorescently label DNA molecules, and images of the fluorescent signal patterns are then used to produce maps.

The average molecule length of optical maps (~225 kb) is substantially greater than the read length produced by short-read sequencing (typically 150–300 bp) and long-read sequencing (typically ~ 15 kb on average) [26]. Optical maps can easily span genomic regions that are difficult to resolve by DNA sequencing. The physical location and relative separation of the restriction/labelling sites obtained from optical mapping facilitate genome scaffolding, genome-assembly completeness validation, and large-scale structural variation detection [27].

Optical mapping is intensively used in genomic studies of microorganisms, plants, animals, and human diseases [10], [28]. Ongoing efforts to improve optical mapping technology will enable further increases in the accuracy of genomic research. Here we review the development of optical mapping, detail its algorithms and applications, and emphasize future perspectives and applications for optical mapping that could assist genomic research.

2. Development of optical mapping

Since being pioneered by Schwartz et al. [25] in 1993, the technology used in optical mapping has been improved substantially (Fig. 1). Initially, optical mapping was used to assist genome assembly of microorganisms, such as yeast and bacteria [29]. However, due to its low throughput, high error rate, and imprecise DNA-length measurement, optical mapping has not been widely adopted for genomic studies in higher species [14]. Recent advances in technologies and algorithms have led to further improvements in optical mapping, making it an important auxiliary tool in genomic research.

Fig 1.

Optical mapping milestone timeline. Major breakthroughs in optical mapping over the past two decades are displayed.

2.1. History of optical mapping

In 1993, Schwartz et al. [25] demonstrated the use of optical mapping by constructing ordered restriction maps of Saccharomyces cerevisiae chromosomes. First, the stained and stretched DNA molecules were fixed in agarose gel and then nicked by a selected restriction enzyme. Next, specialized fluorescence microscopy was used to observe and record the specific patterns of single-molecule DNA restriction fragments. However, it was later found that the thickness of agarose gel can affect image capture by scattering light and attenuating signals.

To solve this problem, in 1995 Cai et al. [30] and Meng et al. [31] fixed DNA molecules on glass surfaces that had been treated with specific chemicals. Consequently, the DNA stretch rate was improved to 60% of its contour length, compared with 30% on agarose [29]. The number of cut sites was also increased. To increase throughput, in 1998 Jing et al. [32] used the “fluid fixation” effect in their design of the first automated approach to generate restriction maps. The improvement of microfluidics and chemistries led to further advances in optical mapping, and in 2007, Jo et al. [33] reported a nanoslit approach that had higher accuracy and throughput in restriction map production.

Throughout this time, companies such as OpGen, Bionano Genomics, and NABsys have invested intensively in single-molecule whole-genome mapping and have extended its application from microorganisms to more complex organisms. OpGen and Bionano Genomics have commercially released different platforms to produce optical maps, such as Argus from the former and Irys and Saphyr from the latter. However, with the acquisition of the Argus platform by BGI in 2011, OpGen’s optical mapping services were terminated. Bionano Genomics currently dominates the optical mapping market and continues to invest in technology development. Yet, given the pending release of the HD-mapping system from NABsys, it is likely that some of the Bionano Genomics market share will be taken by NABsys.

2.1.1. OpGen optical mapping

OpGen was licensed its technology from New York University and the University of Wisconsin and was the first company to commercialize optical mapping technology [34], represented by its release of the Argus platform and its accompanying analysis software, MapSolver [34], [35], [36]. OpGen applied the original restriction-based method for map construction to produce maps of 200 kb in length [10], [29], [36]. At first, Argus targeted simpler organisms, such as bacteria and yeast. Later, OpGen expanded its support to higher organisms across animals and plants, including humans, domestic goat and legume Medicago [34], [36], [37], [38]. A global alignment strategy was applied in the design of MapSolver to help detect complex genome rearrangements [36], [39]. However, with the sale of the Argus system, OpGen discontinued its optical mapping services, and its optical mapping market share was then assumed by its competitors.

2.1.2. Bionano optical mapping

Bionano Genomics is another optical mapping technology provider, originally named BionanoMatrix. The company rose to popularity in 2012 after the release of its massively parallel Irys platform, which applies nanochannel-array fluidics technology to produce optical maps [40]. This technology greatly enhances the throughput and the accuracy of molecule length estimation by use of a more uniform linearization [14], [40], and the use of nicking enzymes to create only single strand breaks preserve molecule contiguity more than OpGen technologies do. The Irys platform came bundled with the IrysView desktop application for data visualization and management. Bionano also provides its proprietary RefAligner for map alignment and Assembler for de novo map assembly, together with an IrysSolve Scripts package to assist data analysis. In 2016, two maps generated from the use of different nicking enzymes were added to the Irys workflow to complement the limited label density and increase map coverage. However, double strand breaks at “fragile sites” where two nicking sites locate closely (~400 bp) on opposite strands still hamper the contiguity of optical maps.

Bionano recently developed Direct Label and Strain (DLS) technology to improve map contiguity to the chromosome-level [41]. The release of the DLS protocol was coupled with the new Saphyr platform, which has improved optics and throughput. DLE-1, the first enzyme in the DLS family, functions akin to a methyltransferase by labelling DNA without damaging it [29], [42]. DLE-1 eliminates the “fragile site” problem [43]. Currently, the Saphyr platform guarantees a production of 1300 Gb of raw data per flow-cell for human samples. The resolution of Bionano optical mapping has been increased up to 500 bp, and there have been attempts to develop algorithms to enable a resolution of 10 bp [40], [44]. Bionano has also updated their data analysis software to a web application supported by Bionano Solve and Bionano Access, to streamlining bioinformatics analysis and data management.

The Bionano Tools are freely accessible by the public. There is constant advancement of instruments and chemistries, and the software is continuously updated with new features by Bionano Genomics, whilst the core components (RefAligner and Assembler) remain proprietary.

2.1.3. Nabsys mapping

In contrast to OpGen and Bionano Genomics, NABsys is not like a traditional optical-mapping technology provider. NABsys eschews image capture and conversion in their technology and instead directly detects the sequence-specific tags using an electronic nanodetector and uses the resulting data to produce maps [45]. NABsys claims that their technology has greater sensitivity and accuracy than existing optical mapping technologies. The resolution of NABsys maps is claimed to be much greater than that of its competitors and is capable of resolving physical distances between tags of fewer than 300 bp [45]. Because the NABsys HD-Mapping platform has not been commercially released, it is difficult to publicly assess its performance. Nevertheless, NABsys mapping is a potential option for use in genome assembly, structural variation (SV) detection, and strain typing.

3. Workflow and algorithms used in optical mapping

Optical mapping workflows generally commence with experiment design, followed by high-molecular-weight DNA extraction, DNA labelling, DNA molecule loading and image capture, and then data conversion, bioinformatics analyses, and validation (Fig. 2). Standard data processing involves molecule-map quality control, map alignment, de novo assembly, and scaffolding (Fig. 3). Downstream analyses include but are not limited to genome-assembly improvement, SV detection, strain typing, and comparative genomics.

Fig 2.

Workflow used in optical map production and analysis. Generally, the workflow can be divided into three parts: wet laboratory preparation, optical map generation, and bioinformatics analysis and validation. The essential steps are listed in this figure to show how optical mapping works.

Fig 3.

Optical-mapping data analysis. The major bioinformatics analysis steps are illustrated from in silico genome digestion, data quality control, de novo map assembly, map alignment, scaffolding, and SV detection.

3.1. In silico digestion

It is important to select the correct enzyme and the label density required for an optical mapping experiment, because a low label density may limit the resolution of maps and reduce the coverage of useful molecules, whilst a high label density may confound data grouping [46]. During genome digestion in silico, a draft genome assembly is usually used to check the label density generated by a selected enzyme in a genome (Fig. 3). This process is based on a sequence motif finding and can be performed with several tools. For instance, Bionano Genomics provides Bionano Knickers, Label Density Calculator, and “fa2cmap_multi_color.pl” in the Bionano Solve package to assist with in silico genome digestion. OMTools [47] and Nucleomics (https://github.com/Nucleomics-VIB/bionano-tools) also implement such a function in their packages to assist with enzyme selection.

3.2. Data conversion

Converting images or electronic signals into labels of physical distances is a prerequisite for optical mapping data analysis (Fig. 2). Usually, this process is performed after image capture by Semi-AutoVis in the OpGen Argus platform, AutoDetect in the Bionano Irys platform, or Bionano Access in the Bionano Saphyr platform. The conversion process can be summarized as extracting skeletal segments, tiling restriction maps, grouping segments, and handling sizing errors [48]. The algorithms used in each tool must deal with ambiguities in sizing and end-point location because they can be easily affected by image signal intensities and noise [46]. Moreover, the thermal motion of DNA under imaging may blur fluorescent patterns, which then require correction via kymograph alignment. Weighted Path Align is an algorithm that may enable kymograph alignment of nanochannel images in linear time [49].

3.3. Data correction

The raw restriction maps produced by various platforms may be error prone. This is mitigated by the use of raw-map error-correction to enable efficient use of data. This correction is performed by software such as cOMet, which corrects raw optical-mapping data by detecting maps with significant overlaps using k-mer seeds (Table 1) [50]. However, cOMet is demanding of computing resources, particularly for processing of large genomes. To solve this problem, a spaced-mer method, Elmeri, was developed [51]. On a human data test-set, it was found that Elmeri outperformed cOMet in both running speed and accuracy and used substantially fewer CPU hours than cOMet (15 CPU hours vs. 10 CPU days). Furthermore, the quality of raw maps generated with Elmeri show more than four times improvement of alignment scores over those generated by cOMet.

Table 1.

State-of-the-art tools used in optical mapping analysis.

Key feature	Tool	Description	URL	Ref
Data simulation	OMSim	A tool for producing synthetic Bionano optical maps	https://github.com/biointec/omsim	[144]
	BMSIM	A tool for producing and mimicking real Bionano molecules	https://github.com/pingchen09990102/BMSIM	[43]
Data correction	cOMet	A tool for raw map error correction using k-mer seeds	https://github.com/kingufl/cOMet	[50]
	Elmeri	A tool for raw map error correction using spaced-mer seeds	https://github.com/LeenaSalmela/Elmeri	[51]
Map alignment	RefAligner	An aligner for map alignment owned by Bionano Genomics using a dynamic programming framework	https://bionanogenomics.com/support/software-downloads	NA
	OMBlast	An aligner for map alignment using a modified seed-and-extend method	https://github.com/TF-Chan-Lab/OMBlast	[56]
	OMMA	An aligner for multiple map alignment in a population-scale study	https://github.com/TF-Chan-Lab/OMTools	[62]
	Maligner	An aligner for raw map alignment based on a dynamic programming framework for large eukaryotic genomes and an indexed method for identifying discordances in the reference	https://github.com/LeeMendelowitz/maligner	[57]
	OPTIMA	An aligner for whole genome alignment using a novel seed-and-extend global alignment method	https://github.com/verznet/OPTIMA	[39]
Genome assembly	Assembler	An assembler for de novo map assembly owned by Bionano Genomics	https://bionanogenomics.com/support/software-downloads	NA
	Novo&Stitch	A tool for assembly reconciliation and scaffolding	https://github.com/ucrbioinfo/Novo_Stitch	[80]
	BiSCoT	A tool for scaffold contiguity improvement based on map alignment	https://github.com/institut-de-genomique/biscot	[79]
Assembly evaluation	BionanoAnalyst	A tool for reporting and visualizing potential errors in the genome assemblies based on discordances in the alignment	https://github.com/AppliedBioinformatics/BioNanoAnalyst	[76]
	misSEQuel	A tool for reporting and resolving mis-assemblies using optical maps and paired-end reads	https://www.cs.colostate.edu/seq/missequel	[85]
SV detection	OMSV	A tool for SV detection based on map alignment from OMBlast and RefAligner using a complete error model	http://yiplab.cse.cuhk.edu.hk/omsv	[117]
	Structome	A tool for SV detection using Bionano Irys data and map alignment	https://github.com/RyanONeil/structome	NA
Data management and visualization	IrysView	A tool bundled with the Bionano Irys platform for data analysis, data management and visualization on a window system	https://bionanogenomics.com/support/software-downloads	NA
	Bionano Access	A tool package bundled with the Bionano Saphyr platform for data analysis, data management and visualization on multiple operating systems	https://bionanogenomics.com/support/software-downloads	NA
	MapOptics	A tool for viewing comparisons of in silico maps, empirical optical maps, and hybrid scaffolds	https://github.com/FadyMohareb/mapoptics	[75]
	OMTools	A toolkit for data processing, map alignment and visualization	https://github.com/TF-Chan-Lab/OMTools	[47]
Pipeline wrapper	runBNG	A wrapper of Bionano IrysSolve Scripts package for data processing and analysis on Linux systems	https://github.com/AppliedBioinformatics/runBNG	[74]
	Irys-scaffolding	A refined pipeline for genome assembly based on Bionano RefAligner and Assembler	https://github.com/kstatebioinfo/Irys-scaffolding	[26]
	OMWare	A refined pipeline for genome assembly based on Bionano RefAligner and Assembler	https://github.com/sharpa/OMWare	[73]

3.4. Map alignment

Alignment is an essential step in the analysis of optical map data. It can be performed by grouping data to quickly identify maps homologous to target genome regions (Fig. 3). Alignment reveals discordances between query and reference maps, which can provide useful information for SV detection or mis-assembly identification [48]. However, the alignment of optical mapping data is more complicated and challenging than sequence alignment because a greater diversity of errors (e.g. missing and spurious labels) can be introduced during enzyme digestion or labelling. DNA breaks can occur during sample preparation. Image noise can also introduce false-positive and false-negative label signals [46]. Moreover, imprecise scaling can lead to errors in molecule length and label distance estimation. To cope with these problems, algorithms are used in sequence alignment, such as local and global alignment and dynamic programming, accompanied by scoring approaches.

In 2006, Valouev et al. [52] introduced a technique to assist optical map alignment, in which the nature of errors in optical mapping are embedded in a scoring system in a dynamic programming framework. Later, SOMA was developed, based on dynamic programming with a heuristic scoring scheme [53]. During benchmarking, it was found that SOMA performed better than Valouev et al.’s technique [53], [54]. Similarly, the Bionano RefAligner also adopts a dynamic programming framework to perform local alignment between maps. RefAligner then penalizes unaligned labels at alignment ends followed by applying a likelihood to select for global alignment [26].

Another method, TWIN, uses dynamic programming and attempts to accelerate alignment by using an FM-index method [55]. However, the error intolerance of TWIN limits its application [56], [57]. OPTIMA (Table 1), a recently developed application, uses a seed-and-extend global–local alignment method with a technology-agnostic statistical model to evaluate alignment performance [39]. Maligner, on the other hand, combines a fast and stringent dynamic programming framework and an index-based method to align single-molecule maps or in silico maps to a reference. A new alignment-score normalization method has also been implemented in Maligner [57]. OMBlast was recently designed to align maps using a seed-and-extend approach that indexes reference as k-tuples to save memory. After seeding, OMBlast recursively extends the alignment with different scaling factors. Its performance is considerably better than that of other tools [56].

Overlapping unassembled raw maps in the optical map assembly step is challenging but necessary for building consensus optical maps. The index-based Kohdista is a recent solution that formulates the alignment problem as an automaton path-matching problem to index and query restriction maps [58]. Kohdista has been tested on E. coli data to demonstrate its fast alignment performance. It is likely that Kohdista will be used or modified to perform high-quality pairwise restriction-map alignments of large eukaryotic organisms.

Multiple alignment is also useful in comparative analysis to identify similar regions that may have functional, structural, or evolutionary relationships. Multiple alignment of nucleotide sequences has been achieved with a number of algorithms, such as those used in ClustalW [59], Muscle [60], and MAFFT [61]. To solve the multiple alignment problem in optical mapping, Leung et al. [62] published a two-step pipeline based on optical mapping by multiple alignment to separate genomic segments into collinear blocks, thereby enabling population-scale comparison of complex genomic features.

3.5. Map assembly

De novo single-molecule map assembly is useful for identification of novel genomic features, such as large insertions in a genome (Fig. 3). It can also be used to detect genetic features of a species without a reference genome.

The concept of de novo single-molecule map assembly is similar to that of assembly in sequence contexts. The fragment assembly problem was first modelled mathematically by Lander and Waterman, suggesting that two reads should be merged if the overlap exceeds a threshold and that, therefore, the short-read lengths can be compensated by deeper coverage [63], [64]. Contig construction can be well approached using graphs, where fragments are represented as nodes, connected by edges when the overlap justifies a merge. The overlap-layout-consensus (OLC) paradigm [65] and the de Bruijn graph (DBG) [66] are the common methods to construct sequence assemblies, where the former performs pairwise alignment for all combinations and the latter computes the overlap information implicitly between neighbouring artificially chopped fragments of a fixed size k called k-mer [64]. A major difference between the two approaches lies in repeat handling. OLC places repeats as separate nodes and DBG collapses them into single nodes, resulting in higher computation and memory demand in pairwise comparison involving repeats in OLC graphs [66].

While DBG offers more efficient performance in sequence assembly, it is not yet readily adopted in de novo whole genome map assembly due to the error properties of optical mapping data that affect its alignment principles as discussed above [67]. The first optical map assembly algorithm was described by Anantharaman et al. [68] in 1997, modelling the errors using Bayesian probability and determining the best alignment pair by dynamic programming. This algorithm uses a greedy heuristic global search for islands of connected fragments to combine and hence is sensitive to data quality. While the algorithm has been demonstrated in several microorganism genomes, the search space remains unscalable [67]. Valouev et al. [69] implemented the OLC graph by first sorting pairwise alignment with a confidence score to construct a layout graph. Although this solution is scalable, it is time-consuming when performing error correction and could drop potential connectivity during strict graph correction [67]. In 2011, Goldstein et al. [70] briefly proposed an idea of using a DBG algorithm Germinate and Grow to assemble restriction maps tested on the Medicago truncatula genome (~500 Mb). The algorithm first uses geometric k-mer hashing to identify potential error-free nodes in the DBG to assemble subsets of restriction maps as seed maps. Then those assembled restriction maps are iteratively extended and refined to cover more parts of the genome. While the algorithm has been reported to be used in the local map assembly of a domestic cow genome, in regions that are largely discordant with the sequence reference from prior alignment, and whole genome map finishing, the tool has not been officially published [71]. Among the latest development, Li et al. [67] proposed an iterative algorithm (IOMA) for optical map assembly, where the consensus map from one OLC contig construction is taken forward as input for the next iteration, until the alignment coverage stops increasingly generate consensus genome maps. Currently, the iterative OLC-based Bionano assembler is the de facto tool for single-molecule optical map assembly, with a customizable fixed number of extensions (Table 1) [72]. Some pipelines, such as AssembleIrysCluster [26] and OMWare [73] are based on the Bionano assembler.

3.6. Data visualization and other processing

Data visualization is an important way to check the quality of data analysis. IrysView, which is bundled with the Bionano Irys platform, is an application that helps with data visualization by enabling plotting of raw data metrics, repeat content, and map alignment by enabling raw-image viewing. However, its reliance on a Microsoft Windows (Microsoft Inc, USA) operating system limits its application [74].

In contrast, OMView implemented in OMTools is the first application to allow data visualization on multiple platforms [47]. OMView accepts separate input maps, pairwise and multiple alignments in standard data formats, and supports annotation visualization. OMView also marks large-scale SVs, such as insertion or deletion of bases (indels) and inversions, with symbols. Options are also provided in OMView to help visualize segment stretch factors with the use of color. Conversely, whilst OMView provides multiple panes to visualize alignments to the same reference, MapOptics provides three way-alignment view features, thus allowing view comparisons of in silico maps, empirical optical maps, and hybrid scaffolds. This is convenient for assessing the performance of hybrid scaffolding [75].

Another cross-platform software is BioNanoAnalyst, which visualizes mis-assemblies as conflicts between sequence and optical map assemblies [76]. Bionano Genomics recently renewed its data visualization tool and integrated Bionano Solve in a web application that incorporates draggable three-way alignments, conflict viewing, resolution viewing, and a Circos-like whole-genome SV-distribution plot to assist Bionano data analysis. An image displayer is installed on the new Bionano workstation computer to visualize the raw image files in a JXR format. The image displayer provides adjustable contrast and export options to help reduce artefactual problems during data conversion. Irys Extract makes further use of raw images by providing a function to crop raw images for specified molecules, which is useful for showing special label patterns [77].

4. Applications of optical mapping

Optical mapping is increasingly used in various areas of genomic research due to its improved accuracy, decreasing cost, and high throughput. Generally, optical mapping is used to improve genome assembly, to facilitate large-scale SV detection, and for strain typing. Although these applications can also be achieved by other technologies, such as genome scaffolding by using linked reads, mate-pair reads or Hi-C reads, compared to optical mapping, the read length and physical read coverage of these alternatives can be lower, and they generally depend on short-read sequencing to produce reads. Therefore, optical mapping remains one of the best options to facilitate these genomic research tasks mentioned above. With recent advances in technologies, optical mapping is also expected to improve haplotyping on different ploidy levels and to be applicable to other research, such as epigenomic studies.

4.1. Genome assembly improvement

Initially, optical mapping was developed to assist genome assembly, during which two approaches are used to piece together fragmented genome drafts. A scaffolding-based approach is commonly used, in which sequence contigs are aligned, validated, and scaffolded in an a priori assembly fashion. In an alternative approach, the long-linkage information from optical mapping data is incorporated into the de Bruijn graph for assembly building.

An early developed tool, SOMA, applies a dynamic-programming algorithm to scaffold short-read assemblies with assembled maps. SOMA models segment-length distribution and calculates the extent of matches between reference and query maps with scoring functions [53]. OMACC is another tool that further improves the accuracy of genome assembly by rescaling optical maps and applying length constraints in path selection [78]. OMACC considers the gap sizes and copy numbers of repeats in the gap between two contigs to generate better assemblies.

In addition to the hybrid pipeline provided by Bionano Genomics for Bionano scaffolding, the Irys-scaffolding pipeline is the first refined pipeline to help scaffolding [26]. This pipeline applies a threshold minimum percent of total aligned length to filter local alignments. Although the source code of the Irys-scaffolding pipeline is available on GitHub (Table 1), the alignment still relies on RefAligner, which is difficult to optimize due to the many unexplained configuration parameters provided. A pre-released Bionano SCaffolding COrrection Tool (BisCoT) was recently claimed to use pre-existing optical assembly to generate better scaffolds than the Bionano hybrid scaffolding pipeline by closing gaps. However, the accuracy of this new method is yet to be evaluated [79].

Multiple genome assemblies can be merged to improve the contiguity of genome assembly. Novo & Stitch is such a tool to produce the best assembly merged from several assembly versions [80]. The pipeline iteratively selects the optimal non-conflicting alignments and stitches up sequence contigs by aligning two original contigs to the new one. In contrast, OMGS takes a more streamlined scaffolding approach as it incorporates information from multiple optical maps by correcting raw maps with cOMet [81].

There are two algorithms that directly align optical maps to a DBG to help sequence-based genome assembly. Optical maps are used as a guide to connect paths and to search for correct paths instead of grouping repeats into single paths. The first algorithm was AGORA, which uses assembled restriction maps to eliminate inconsistent alternate paths [82]. Then came omGraph, an algorithm designed to use the shorter and error-prone raw restriction maps by first performing map error-correction [83]. Both algorithms were designed to support restriction maps, whilst tests and optimization are needed to apply them to non-restriction-based data, as the latter are known to have different error models [84].

In addition to improving the contiguity of genome assemblies, optical maps can be used to detect mis-assemblies. Several specialized tools are available for this purpose. For instance, BioNanoAnalyst uses a scoring approach to detect mis-assemblies in a draft genome assembly by identifying discordances between consensus optical maps and the reference [76]. MisSEQuel detects breakpoints of errors using paired-end sequence reads and optical mapping data [85]. MisSEQuel can also be used to detect artificial genomic rearrangements in a genome caused by a sequence assembler. Chimericognizer is an alternative that uses one or more sets of Bionano optical maps to detect chimeric contigs or maps by concatenating multiple sequence assemblies [86]. The assemblies can be obtained from different assembly tools or one assembly tool with different parameter settings.

Optical mapping was successfully used in the microbial genome assembly of species such as Deinococcus radiodurans [87] and Plasmodium falciparum [88] during the early development. Later, OpGen enabled optical mapping to the genome assembly of advanced species. In a goat genome assembly, OpGen optical maps helped anchor 2090 fosmid scaffolds into 315 super-scaffolds and demonstrated that optical mapping is a useful technique in genome assembly for species with larger genomes. After that, optical mapping was gradually adopted for analysis of the genomes of animals species such as the ostrich [89] and the yellow croaker [90].

The adoption of optical mapping is further increasing with the development of long-read sequencing. In current genome-assembly pipelines, optical maps are commonly used with a combination of PacBio or Oxford Nanopore reads with Hi-C data to build reference-quality genome assemblies. This technology combination has been applied to genome analysis in various species, such as humans [91], [92], other animals [93], [94], [95], [96], and plants [97], [98], [99], [100], [101], [102], [103], [104], [105], [106], [107], [108], [109], [110]. In particular, the Genome Reference Consortium (GRC), in an effort to improve the quality of reference genomes by error correction, gap closure, and variation representation, has been producing optical mapping data for humans and other model organisms [36]. Another ambitious application of optical mapping is implemented in the Vertebrate Genome Project (VGP) [111], which is affiliated with the Genome 10K Consortium. The data and assemblies for approximately a hundred species among the 260 species targeted in Phase I are currently available at https://vgp.github.io/genomeark.

4.2. Large-scale SV detection

Genetic variations are important resources for the study of evolution and domestication. Nevertheless, the investigation of genetic variations has long been technically limited to small variants, such as single nucleotide polymorphisms (SNPs) and small indels. Large genomic SVs of kilobases or more in length have been challenging to study. In 2008, Kidd et al. [112] mapped eight human genomes from the Human Genome Structural Variation Project and refined 1695 SV locations larger than 6 kb. They used fosmid end-sequence pairs to validate 11 loci of potential SVs that are greater than 40 kb. This study set a foundation for SV study by optical mapping. Optical mapping has since been applied to various comparative genomic studies.

In 2016, Mak et al. [113] demonstrated systematic genome-wide SV detection by comparing assembled optical maps, which were generated with the Nt.BspQI nickase from a father-mother-daughter trio selected from the 1000 Human Genome Project (1KGP) [113]. This study identified 59 insertions and 156 deletions over 5 kb and 16 inversions. The detection of the integration sites of Epstein–Barr Virus (EBV) during cell-line transformation demonstrates the application of optical mapping in large-scale SV identification. In 2019, Levy-Sakin et al. [114] mapped genomes of 154 individuals among the 26 populations in the 1KGP and found that genes located in copy-number variation regions seem to agree with the evolutionary patterns.

In addition to its application in human genomes, large-scale SV detection is also important in other research. For example, the mapping of a highly rearranged liposarcoma cell line generated 72 fusion maps representing 112.3 Mb highly rearranged regions, suggesting that chained fusions and a higher level of complex genomic architecture can occur in cancer [115]. In a soybean study, optical mapping was used to reveal an inversion that causes loss of soybean seed-coat color during domestication [106]. In a primate evolutionary study, SVs were found to be enriched near genes down-regulated in human cerebral organoids compared to those of chimpanzees [116].

There are two tools used for optical mapping SV detection: Bionano SVCaller and OMSV [117]. OMSV is a tool that identifies SVs by combining molecule-to reference alignments generated from RefAligner and OMBlast. During processing, RefAligner is used to generate a homologous alignment, whilst OMBlast handles complex rearrangements by split-alignment. Variations such as extra or missing labels, large segment-size differences and complex SVs are detected separately before being combined and deduplicated.

To facilitate the annotation of identified clinically relevant SVs, an R package called nanotatoR has also been released [118]. It calculates population frequencies of SVs using data from the Database of Genomic Variants (DGV) and, to filter variants, overlaps primary gene lists from the NCBI databases to SVs identified by Bionano SVCaller. Although optical map-based SV detection tools have been developed, their application and performance in higher ploidy contexts have not been demonstrated.

4.3. Microbial strain typing

Pathogen identification is essential in epidemiological surveillance. Precise classification at a sub-species level, known as strain typing, is necessary to provide insight into clonal evolution that helps to trace transmission routes and formulate outbreak control. Strain typing was one of the earliest common applications of optical mapping. In strain typing, the gold standard method is pulse-field gel electrophoresis (PFGE), which enables the analysis of size patterns of DNA restriction fragments, and multilocus sequence typing (MLST), which involves Sanger sequencing of several selected housekeeping gene loci to match the profile of allele combinations [119], [120]. However, MLST relies primarily on the assumption that housekeeping genes are evolutionarily conserved but lacks whole genome information, whilst PFGE does not provide the order of DNA fragments, which makes it difficult to resolve fragments of similar sizes. These drawbacks limit the resolution of these methods. Optical mapping, as a high-throughput profiling alternative, can be easily applied to capture the size and order of fragments from the whole genome.

Bacteria are known to exchange genetic materials via horizontal transfer, including cassette elements that harbor toxin and antibiotic resistance with medical implications [121]. In 2007, Latreille et al. [122] proposed to routinely use optical mapping combined with DNA sequencing for bacterial genome assembly to aid identification, until simultaneous mapping of multiple bacterial genome became possible [35]. Furthermore, optical mapping has proven useful in various cases, such as the tracking of a food-borne outbreak of an enterohemorrhagic Escherichia coli O157:H7 and distinguishing five strains of a toxigenic Vibrio cholerae O1 under the same serotype and biotype from different geographic origin [123].

4.4. Haplotype phasing

Haplotype phasing finds continuous stretches of DNA on the same chromosome and can be viewed as a means of resolving genome assemblies at a higher resolution and detecting more SVs between haploid genomes of the same organism [124]. Capturing accurate allele structures not only refines the genome assembly, but also contributes to better understanding of the population structure and evolutionary process [125]. Various approaches can be used to solve this problem, based on the use of population, family trio, or a single-sample information and different molecular methods [126].

The first comprehensive analysis of the diploid human genome was performed using a hybrid assembly of Illumina short reads, PacBio long reads, and Bionano optical maps [127]. This study achieved 99% consistency with previous trio results. The phased human genome of the Korean individual AK1 is a demonstration of this application [91]. Haplotype phasing in nonhuman species is not yet popular, and haplotype-aware assembly pipelines for nonhuman samples are available but not yet officially supported by Bionano. However, some attempts have been made to use optical mapping to perform haplotype genome phasing in buffalos [128], cattle [129], and pigs [130]. In a recent African cassava study, optical mapping was used to identify allelic variants and allele-specific expression via haplotype comparison [131].

In contrast to DNA sequencing, optical mapping can produce long single molecule maps to cover heterozygous genome regions or loci with different haplotypes that cannot be easily spanned by DNA sequence reads. Using optical mapping, it is possible to resolve different alleles without the requirement of a family-trio or population information. Nevertheless, due to the relatively low resolution of optical mapping, DNA sequencing is needed to retrieve the sequences of different haplotypes in the phased optical map alignment.

4.5. Other genomic feature detection

Although optical mapping produces maps with superior lengths, the molecule mapping rate and hence the accessibility of different genomic regions still depends on the distribution of the enzyme recognition motifs. Moreover, the motifs of standard enzyme choices often have no direct biological implications. For more options and control, scientists are looking to alternative labelling methods to target regions of interest.

Early proposals included labelling with peptide nucleic acids or triplex-forming oligonucleotides [132]. In 2019, Müller et al. [133] published a protocol for enzyme-free optical mapping by competitive binding of YOYO-1 and netropsin. Their method blocks an AT-rich region that highlights a DNA contour, preventing DNA damage by nicking. This method also offers the potential to highlight DNA damage lesions.

In terms of more direct biological relevance, a vibrant research area of optical mapping is epigenetics. Methylation of CpG islands, clusters of cytosine-guanine dinucleotide DNA motifs, is a key mechanism in mammalian imprinting. In 2008, Ananiev et al. [134] tested mapping methylation optically on E. coli and human cell lines by using a methylation-sensitive restriction enzyme. In 2018, Gabrieli et al. [135] looked for 5-hydromethylcytosine, a product of active DNA demethylation, on the Bionano platform. Later, Sharim et al. [136] advanced methylation detection by using a synthetic analog to “trick” methylases into transferring a fluorophore instead of a methyl group. Hence, the protocol is compatible with the high-throughput Bionano platform with the merit of minimizing DNA damage, similar to the DLS method discussed above.

In addition to epigenomics, the application of optical mapping has been extended to other areas. In 2020, Young et al. [137] utilized the CRISPR/Cas9 targeted-labelling technique developed by McCaffrey et al. [138] to extend their previous study on subtelomeric regions that used nick-based genome maps. By specifically tagging fluorophores to the telomeric repeats, the location of population-specific patterns identified in the subtelomeric region using traditional labelling method were verified. The Bionano team has recently announced an improved non-nicking targeted-labelling method called CRISPR-Bind to further improve the application of optical mapping [139].

5. Summary and outlook

While current advances in DNA sequencing have improved genomic research by producing more complete genome assemblies and detecting larger SVs, the use of DNA sequencing alone is often not yet sufficient to produce a genome assembly of desired reference-grade quality, particularly in eukaryotic organisms. The application of other genomic methods such as optical mapping usually help to overcome the shortfalls in sequencing data to achieve a chromosome-level genome assembly for species with large and complex genomes.

After three decades of development, optical mapping has emerged as an important complementary method in genomic studies. Although it cannot replace DNA sequencing due to a lack of base-by-base nucleotide information, it can quickly report genomic structural information that is not easily captured by DNA sequencing. With the ability to map long molecule lengths at low cost, optical mapping has facilitated genome assembly, large-scale SV detection, and strain typing. Nonetheless, the resolution of optical mapping is still low, which makes it unsuitable for analysis of maps shorter than 100 to 150 kb, where alignment reliability is easily compromised by insufficient labelling sites [29].

Although some strategies have been implemented to improve resolution, such as applying multiple enzymes or direct DNA labelling, most designs are based on human genomes and may not be transferable to other species. Direct labelling is a promising solution to increase mapping resolution and avoid double-strand breaks. However, only the enzyme DLE-1 is commercialized for the Bionano Saphyr system. Due to the genome divergence between different species, one sequence motif pattern may not suit all species, and thus other labelling alternatives validated by comprehensive tests and demonstrations are required.

In addition, the key algorithms implemented in Bionano Tools are not open source. Researchers cannot directly study or modify the code used in RefAligner and Assembler to improve their performance. Although other tools such as OMTools [47] and OMSV [117] have been developed to assist optical mapping analysis, no other de novo assembly tools are available to compete with Bionano Tools. The direct use of the Bionano pipeline may also introduce false positives and false negatives, particularly in non-human genomes. Parameter tuning can be useful to customize the pipeline for different studies, yet the limited instructions provided and the lack of clarity on the algorithms behind RefAligner and Assembler make the adjustment difficult. For de novo map assembly, although k-mer hashing of optical maps has been suggested to be impractical in the assembly context [67], with the development of k-mer based alignment tools such as OMTools, this possibility may be worth re-exploring.

Haplotype phasing in different ploidy levels is also important in genomic research. Although Bionano Genomics has released a haplotype-phased de novo assembly pipeline to assist haplotype studies, the pipeline is human-genome oriented and little instruction has been released to assist pipeline tweaking. Different algorithms and methods are needed in optical mapping to help achieve a global phased genome for species with large and complex genomes. With the advantages of optical mapping and an increasing number of polyploid species being studied, optical mapping is expected to overcome the complications arising from polyploidy [140]. Still, a polyploid-specific analysis pipeline remains to be developed, and it is difficult to validate the reported results.

Graphs can be an efficient data structure to represent reference sequence assemblies, where different alleles are separated into different paths, with preserved linkage information and identical sequences grouped into single paths to save space [141]. A condensed graph structure presents advantages in both saving storage and in minimizing extra computational demands during analyses [142], [143]. This might motivate the further development and application of graph-based optical mapping analysis.

In the near future, it is expected that novel methods, such as new DNA stretching, labelling, and analysis approaches will be developed to increase the accuracy of optical mapping from signal conversion, map alignment, and de novo assembly to downstream analyses. Work is still needed to explore the possibility of optical mapping in haplotype phasing and polyploid research. The low efficiency of optical mapping in data visualization and examination impedes its application to large-scale projects. An automated checking and validation method is therefore required to further improve the efficiency of optical mapping.

Overall, optical mapping has proven indispensable for genomic studies. With growing application and development efforts, it is expected that optical mapping will continue to improve genomic studies, enabling the assembly of more complete genomes and the discovery of novel variations. The consequent determination of genotype and phenotype associations from optical mapping is also becoming more practical. Although the current technologies have some pitfalls, it is expected that refined optical-mapping solutions will be developed in the near future to further enrich genomic research.

CRediT authorship contribution statement

Yuxuan Yuan: Visualization, Writing - review & editing. Claire Yik-Lok Chung: Visualization, Writing - review & editing. Ting-Fung Chan: Supervision, Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.