16S rRNA metagenome clustering and diversity estimation using locality sensitive hashing
© Rasheed et al.; licensee BioMed Central Ltd. 2013
Published: 23 October 2013
Advances in biotechnology have changed the manner of characterizing large populations of microbial communities that are ubiquitous across several environments."Metagenome" sequencing involves decoding the DNA of organisms co-existing within ecosystems ranging from ocean, soil and human body. Several researchers are interested in metagenomics because it provides an insight into the complex biodiversity across several environments. Clinicians are using metagenomics to determine the role played by collection of microbial organisms within human body with respect to human health wellness and disease.
We have developed an efficient and scalable, species richness estimation algorithm that uses locality sensitive hashing (LSH). Our algorithm achieves efficiency by approximating the pairwise sequence comparison operations using hashing and also incorporates matching of fixed-length, gapless subsequences criterion to improve the quality of sequence comparisons. We use LSH-based similarity function to cluster similar sequences and make individual groups, called operational taxonomic units (OTUs). We also compute different species diversity/richness metrics by utilizing OTU assignment results to further extend our analysis.
The algorithm is evaluated on synthetic samples and eight targeted 16S rRNA metagenome samples taken from seawater. We compare the performance of our algorithm with several competing diversity estimation algorithms. We show the benefits of our approach with respect to computational runtime and meaningful OTU assignments. We also demonstrate practical significance of the developed algorithm by comparing bacterial diversity and structure across different skin locations.
New genomic technologies allow researchers to determine DNA sequences of organisms existing as communities across different environments , . The collective sequencing of organisms without culturing and cloning each organism individually is known as "metagenomics". Metagenome samples consist of several DNA sequences originating from all organisms in the examined environment. Through metagenomics, it is possible to study the vast majority of microbes on earth and systematically investigating, classifying, and manipulating the entire genetic material extracted directly from environmental samples. Metagenomics enables scientists to conduct a survey of different microorganisms present in a specific environment, such as water, soil and human body [1, 3, 4]. By comprehensive study of nucleotide sequence, structure, regulation, and biological functions within the community, the roles played by microbial communities can potentially be examined.
However, sequencing technologies do not provide the whole genome of different co-existing organisms, but produce short contiguous subsequences called sequence reads from random positions of the entire genome. One of the grand challenges in the study of metagenome data involves reconstructing the different microbial genomes from a mixture of sequence reads. This is referred to as the metagenome assembly problem. Due to the high species complexity and the short length of sequencing reads from current sequencing technologies, the genome reconstruction goal becomes more difficult to attain if not impossible. Also within a community, microbes (or organisms) vary in abundance, diversity, complexity, genome lengths and may have not been individually sequenced before. Likewise, the current sequencing technologies produce large volume of sequence reads, and reads that may have varying error idiosyncracies . As such,the metagenome assembly problem is complex and challenging  and is often subject to further analysis as a collection of short reads.
Targeted metagenome or 16S rRNA gene sequencing has been widely used for the analysis of genetic diversity estimation, enabling deep views into hundreds of complex microbial communities simultaneously. 16S sequences are marker genes, which exist in most microbial species but have variations in their sequences that allow them to be separated into different taxonomical groups . Several metagenome analysis projects use sequencing of 16S genes as a first step in estimating the diversity within a sample. Various computational methods have been developed for the rapid analysis of large sets of reads obtained from targeted metagenome (16S marker genes) or "whole" metagenome studies . Clustering methods have been developed to compare metagenome samples by grouping similar metagenome sequences into bins [8–10]. Other methods use classification techniques to categorize metagenome samples into different phylogenies .
In general, two sets of approaches are considered for species diversity estimation in metagenome analysis. The first approach consists of comparative or sequence similarity based methods that rely on homology to separate sequences into different taxonomic levels and classes using annotated database [12–14]. These methods align reads or contigs using global and local sequence alignment algorithms (characters) to identify regions of similarity between sequences . The second approach contains unsupervised clustering methods that identify groups of similar sequences within metagenome samples. This grouping of similar sequences is commonly known as "binning" problem. Different groups in a particular sample is also referred to as Operational Taxonomic Units (OTUs) and the number of OTUs gives an approximation of species diversity in a sample [16–18]. OTU-based approaches are not constrained due to the absence of a complete coverage in taxonomic databases. Several environmental samples contain microbial organisms that have never been laboratory cultured, and as such do not exist in genomic databases.
The OTU assignment can be used to estimate several species diversity estimates such as Chao1 index , Shannon diversity index  and Abundance-based Coverage Estimator (ACE) index . These OTU assignments and diversity estimates facilitate the process of comparative metagenomics i.e., comparing the genomic content of different community samples. Mothur, DOTUR and ESPRIT are the most widely used methods for OTU estimation [22, 23]. QIIME  is an open source software package for OTU estimation, taxonomic assignment, statistical analysis and comparison of microbial communities and is primarily used for analyzing high-throughput 16S metagenomic data, generated on a variety of platforms.
One of the purpose of developing clustering methods is to handle the large output of metagenome sequencing projects. We propose a new, scalable metagenomic sequence clustering algorithm (LSH-Div) for targeted metagenome sequences (or called 16S rRNA metagenomes) that utilizes an efficient locality sensitive based hashing (LSH) function  to approximate the pairwise sequence operations. LSH algorithm maps the original dimension of input sequences into reduced dimensional space by randomly choosing subset of sequence positions. LSH-based method is also used by Buhler et. al.  to detect all pairs of similar segments within long genomes. Our LSH-function is enriched to use gapless, subsequences of fixed length (w-mer), which helps us to reduce the number of false positives and improve cluster quality. The LSH-Div algorithm follows a greedy and iterative framework for assigning OTUs (or clusters) to each sequence. LSH-Div further uses these OTU assignment results and introduces standard species richness metrics which bring more explanation to clustering results.
We assess LSH-Div algorithm on eight environmental samples and one synthetic sample taken from sea water. These samples contain varying complexities of microbial community which give in-depth representation of microbes in sea water. Our evaluation metrics include quality of OTUs, species diversity estimation, computational speedup and pairwise sequence similarity of sequences within each OTU. We demonstrate that LSH-Div is computationally efficient and also produces meaningful and accurate results in comparison to the state-of-the-art OTU estimation algorithms. This work is an extension of our previous work on species diversity estimation  and clustering .
In this section, we briefly explain three state-of-the-art OTU estimation algorithms. Mothur  and DOTUR  use a PHYLIP-formatted pairwise distance matrix  for computing the species richness metrics. In ESPRIT , the computation is reduced by using k-mer distance between sequences.
MOTHUR and DOTUR are hierarchical clustering approaches that take as input an all-pairwise similarity/distance matrix. The distance matrix calculated by performaing sequence alignment between every pair of sequences within an input metagenome sample. Both approaches support three merge criterion to perform hierarchical clustering given by nearest-neighbor, furthest-neighbor or use of group average. These approaches also use a distance cutoff as an input parameter. This cutoff varaible d, defines the maximum allowable distance between sequences within the final cluster (also OTU), such that sequences within a given cluster are at most d% distance apart.
DOTUR reports bio-diversity metrics (e.g., Chao1, Shannon and ACE) with varying distance cutoffs. On the other hand, MOTHUR generates statistics such as count of OTUs, singletons, doubletons and number of sequences in the largest OTU. As a computational expenses, both approaches require memory space for loading the entire all-pairwise distance matrix.
ESPRIT is another sequence clustering algorithm that avoids performing the computationally intensive all-pairwise distance calculation. Instead, the algorithm computes a k-mer based distance between pairs of input sequences and achieves computational efficiency. For each sequence, ESPRIT builds a complete genomic alphabet profile of selected k-mer and uses that representation to calculate all pairwise distances . In order to reduce further computational complexity, ESPRIT also uses various heuristics. For example, if two sequences are identical or one sequence is a subset of the other sequence, then only longer sequence is kept and the number of occurrence for identical sequences are recorded. This technique reduces the total number of sequences and therefore reduces the time taken by k-mer pairwise distance matrix calculation. Using the pairwise input matrix, ESPRIT then performs standard hierarchical clustering.
Locality sensitive hashing
where parameter p is the maximum allowable mismatch between the two pairs of strings and P  is the probability which is computed over random choices of the k indices, i1, . . . , i k .
Naturally, the LSH-based procedure will lead to either false positives (FP) or false negatives (FN) . When the two strings, s x and s y are mapped to same hash values even if they are not similar i.e., h(s x ) = h(s y ), then a FP has occured. This is because of the random sampling procedure that chose only those k nucleotide positions that were identical in the two sequences. When the two sequences s x and s y are similar but not exactly same, then a FN can occur when the hash function samples those k indices where the sequences diverge. This leads to h(s x ) ≠ h(s y ) even if two sequences are similar in terms of practical purposes. FNs can be substantially reduced by iterating the sampling procedure, multiple times. l denotes the parameter for the number of iterations.
Figure 1 shows the LSH-Div algorithm procedure.
In this section, we describe the parameters of LSH-Div algorithm in detail.
Number of sampled indices (k) and subsequence length (w)
The hash function provides a mapping from the 4 n -dimensional space into 4 k -dimensional space. Using small values for k will lead to small number of partitions, and large number of false positives, since the number of sampled bits or indices will not be enough to disambiguate dissimilar reads.
The use of gapless, subsequence of length w enhances the filtering quality as we compare a w-mer rather than single nucleotide at a given index position. For pairs of strings to hash to the same key value, requires an exact matching of k * w nucleotides. This makes the matching process more stringent and will lead to a reduction in the number of false positives. The selection of k sampled indices along with subsequence length w is shown in Figure 2.
Number of hash functions or iterations (l)
The use of multiple hash functions within the LSH-Div framework reduces the number of false negatives i.e., two sequences that are similar will have a higher chance of being accepted by our filter due to repeated sampling of k indices. The results of multiple iterations is combined using a union set operation i.e., a pair of strings are considered to be similar, as long as we see the strings hash to the same values in one of the l iterations (or l samplings). However, using large values of l will lead to dissimilar sequences being mapped to the same hash value and will lead to an increased number of false positives. A good choice of l and k allows LSH-Div algorithm to produce sensitive partitioning of the sequence data.
Percentage of mismatches (p)
In order to allow for pairs of strings that are not exactly identical, but similar (i.e., a few nucleotide mismatches) to pass the LSH-based hamming distance filter, the mismatch factor parameter p is implemented as the percentage of allowable mismatches. When p is set to 0%, the LSH-based function will consider strings to be equivalent if and only if all the k nucleotides or k w-mers are exactly identical in both the strings. As an example, when k is set to 64 indices and p is set to 10% mismatch, then pairs of strings that differ by at most 6 nucleotides or 6 w-mers will be considered equivalent. In our current approach, each w-mer can be considered a symbol. We do not allow for mismatches within a w-mer and is considered as a part of the future directions of this work.
LSH-Div framework consists of different modules such as the hashing module, OTU assignment module and application module. The algorithm begins by first generating LSH functions for all the input sequences in set . h(s) is defined using the parameters of k and w defined earlier. After computing all the different hash values, the algorithm enters into the OTU assignment module.
The assignment process starts by choosing a random sequence s x from set and assigns the first OTU to that sequence. Then for every other sequence sy in set , LSH-Div performs hamming distance calculation denoted by Sim(,), and computes similarity between using the hash values. Using the Sim(,) function we assign every sequence s y in the cluster (started by s x ) that differs from s x by a few fraction of nucleotides. Exactly similar strings are be accepted by the hashing function. Sequences, once assigned are removed from the set . We repeat the above assiged procedure for assigning OTU labels to all the sequences in the input. To improve the performance of LSH-Div, whole OTU assignment module is repeated l times. A new LSH function is created for every iteration. Finally, a union operation is performed on the OTU assignments obtained on each successive iterations.
Previous approaches (discussed above) have are computationally intensive and need a computation of all-pairs of sequence distances for OTU/cluster assignment. In comparison, the LSH-Div algorithm does not require all pairwise distances computation for OTU assignment.
Computation of pairwise distance cut-offs
An input parameter that is implemented in all diversity estimation approaches includes the distance cutoff parameter. The cutoff parameter provides a maximum allowable distance between all sequences within an OTU. As an example a distance cutoff of 0.05 implies that a sequence within an OTU is at the most 5% distant from every other sequence within the same OTU. For LSH-Div, we first generate the OTU assignments for each sequence. We then compute all pairwise distances (using either the global sequence alignment or k-mer distance) between all pairs of sequences within the OTU. We then report the maximum pairwise distance computed, which allows for validation of the distance cutoff parameter. This procedure allows us to compare our approach to MOTHUR and DOTUR, both of which use the alignment distance cutoff and ESPRIT, which uses the k-mer distance cutoff.
Results and discussion
In this section we report on a thorough set of experimetal results. We evaluate LSH-Div on several metrics. Specifically, we assess the performance of LSH-Div algorithm with respect to the number of OTUs, different species diversity metrics and run time. We also perform comparative evaluation.
Number of OTUs and species diversity estimation by LSH-Div algorithm.
OTUs Cutoff Distance 0.03
OTUs Cutoff Distance 0.05
OTUs Cutoff Distance 0.10
Figure 5 shows the run times for the different methods on each of the eight samples. As discussed earlier, ESPRIT, Mothur and DOTUR require O(N2) comparisons to compute the pairwise distance matrix, where N is the number of sequences in the input set. Mothur and DOTUR use the more expensive Needleman-Wunsch global alignments, whereas ESPRIT uses a k-mer based distance function for every sequence pair. ESPRIT also achieves efficiency by removing redundant sequences and also filters sequence pairs that are not very similar. In evaluation of the competing methods, the original environmental samples FS312 and FS396 did not work for ESPRIT, Mothur and DOTUR. These samples required 15GB of memory, and as such were reduced in size by a filtering procedure performed by the ESPRIT method. We report the run time results for the trimmed as well as the original input sequences. We can see a clear performance advantage achieved by LSH-Div in comparison to the other approaches.
Evaluation of ESPRIT on global alignment pairwise distance
ESPRIT and LSH-Div results for different cut-offs (Sample 112R)
k -mer distance
NW / k -mer
Significance and impact
Metagenome projects aim to determine the species similarity and diversity across different ecological samples. We use our algorithm to provide an enriched analysis of the variation of microbial species across different skin locations (a real environmental dataset ). This dataset contains 112,283 near-full-length 16S rRNA sequences sampled across 21 different skin locations, from ten healthy human controls. These skin sites were selected because they show a predisposition for bacterial infections. The skin sites are categorized as: (i) sebaceous or oily, (ii) moist (typically skin creases) and (iii) dry, flat surfaces. As given by Grice et. al. , the sebaceous sites include locations like glabella (between the eyebrows), alar crease (side of the nostril), external auditory canal (inside the ear), retroauricular crease (behind the ear), occiput (back of the scalp), manubrium (upper chest) and back. Moist sites include the nare (inside the nostril), auxiliary vault (armpit), antecubital fossa (inner elbow), interdigital web space (between the middle and ring fingers), inguinal crease (side of the groin), gluteal crease (topmost part of the fold between the buttocks), popliteal fossa (behind the knee), plantar heel (bottom of the heel of the foot), toe web space and umbilicus (navel). Dry sites include the volar forearm (inside of the mid-forearm), hypothenar palm (palm of the hand proximal to the little finger) and buttock. We use the information about skin locations and "genera" as ground truth in our study. The sequence lengths varied from 1280 to 1370 nucelotides (1300 average).
Clustering of metagenome sequences (16S) tends to group sequences that are closely related in the taxonomy/phylogenetic tree (genera) together. Using the LSH-Div algorithm, we cluster all the metagenome sequences. Sequences in this dataset are taken from different skin locations where each skin location is a metagenome sample i.e., genomic sequences of microbes co-existing on the specific skin locations. After binning sequences across all the skin locations, each sequence is associated with one of the OTU labels. Using the OTU label as features for a skin location, we compute the Jaccard index between all pairs of skin locations. Jaccard index measures the proportion of shared clustering labels (species) between the pair of skin locations. A higher index value indicates that the two skin locations are more similar to each other.
We developed an efficient and accurate species diversity estimation algorthm referred to as LSH-Div. The key features of our algorithm include the use of a randomized, locality-sensite hashing approach that reduces the complexity of computing expensive global alignments across all pairs of input sequences. The algorithm also incorporates the use of w-mers that reduces the number of false positives, and does not increase the run time but leads to better OTU estimation results. LSH-Div was evaluated on synthetic datasets, real world datatsets and after a thorough study, we were able to demontrate that LSH-Div is computationally efficient in comparison to the best OTU estimation algorithms. It also reports the key diversity estimation metrics that are widely used by different biologists. The LSH-Div code is written in Python programming language and is available publicly with the GNU GPL license.
Materials and implementation
In this section, we describe the datasets and species richness metrics used in the experiments.
LSH-Div was assessed on both synthetic and real environmental 16S rRNA metagenome samples. The synthetic data was obtained from a previous study conducted in . The synthetic dataset had a total of 345K sequence reads and the reads were generated from forty-three known 16S marker gene sequences. More details about the dataset can be found in .
The eight seawater-based metagenome samples were taken from Sogin et. al. . These samples use the 454-based sequencing technology to provide a global in-depth description on the diversity of microbes and their relative abundance in seawater. The description of the samples are given in . The mean length of sequence reads within thesese samples was found to be 60 bp.
We also use a real environmental skin dataset to show the application and significance of LSH-Div algorithm. The dataset covers 21 different skin locations represented by 16S rRNA sequences samples. The total number of sequences is 112,283 with an average length of 1300 bp.
Species richness estimation metrics
where S obs is the number of observed species (number of OTUs), n1 is the number of OTUs with only one sequence and n2 is the number of OTUs with only two sequences.
where S obs is the number of observed OTUs, n i = the number of sequences in OTU i and N is the total number of sequences in the sample. We denote Shannon index as H′ in this paper.
where n i is the number of OTUs with i assigned sequences, S rare is the number of OTUs with 10 or fewer assigned sequences and S abund is the number of OTUs with more than 10 assigned sequences.
The results produced by LSH-Div can also be used to compute other richness metrics. Also, the rarely occurrent OTUs can be compared against annotated databases in order to identify new species.
Hardware and software details
The LSH-Div algorithm is available on the supplementary website. It is written using the Python programming language. For experimental evaluation, a single desktop was used. The workstation had 6GB RAM memory witn an Intel-i5 2.53 GHz processor. The competing approaches were all run on the same machine using executables provided by the authors of respective software.
This article is supported by NSF Career Award IIS-1252318 awarded to HR. Based on "LSH-Div: Species Diversity Estimation using Locality Sensitive Hashing", by Zeehasham Rasheed, Huzefa Rangwala and Daniel Barbará which appeared in Bioinformatics and Biomedicine (BIBM), 2012 IEEE International Conference on. © 2012 IEEE 6392649.
The publication costs for this article were funded by NSF Career Award IIS-1252318 awarded to HR.
This article has been published as part of BMC Systems Biology Volume 7 Supplement 4, 2013: Selected articles from the IEEE International Conference on Bioinformatics and Biomedicine 2012: Systems Biology. The full contents of the supplement are available online at http://www.biomedcentral.com/bmcsystbiol/supplements/7/S4.
- Venter J, Remington K, Heidelberg J, Halpern A, Rusch D, Eisen J, Wu D, Paulsen I, Nelson K, Nelson W, et al: Environmental genome shotgun sequencing of the Sargasso Sea. science. 2004, 304 (5667): 66-74. 10.1126/science.1093857.View ArticlePubMedGoogle Scholar
- Qin J, Li R, Raes J, Arumugam M, Burgdorf KS, Manichanh C, Nielsen T, Pons N, Levenez F, Yamada T, et al: A human gut microbial gene catalogue established by metagenomic sequencing. Nature. 2010, 464 (7285): 59-65. 10.1038/nature08821.PubMed CentralView ArticlePubMedGoogle Scholar
- Gill SR, Pop M, DeBoy RT, Eckburg PB, Turnbaugh PJ, Samuel BS, Gordon JI, Relman DA, Fraser-Liggett CM, Nelson KE: Metagenomic analysis of the human distal gut microbiome. science. 2006, 312 (5778): 1355-1359. 10.1126/science.1124234.PubMed CentralView ArticlePubMedGoogle Scholar
- Poinar HN, Schwarz C, Qi J, Shapiro B, MacPhee RD, Buigues B, Tikhonov A, Huson DH, Tomsho LP, Auch A, et al: Metagenomics to paleogenomics: large-scale sequencing of mammoth DNA. science. 2006, 311 (5759): 392-394. 10.1126/science.1123360.View ArticlePubMedGoogle Scholar
- Hugenholtz P, Tyson G: Microbiology: metagenomics. Nature. 2008, 455 (7212): 481-483. 10.1038/455481a.View ArticlePubMedGoogle Scholar
- Charuvaka A, Rangwala H: Evaluation of short read metagenomic assembly. BMC Genomics. 2011, 12 (Suppl 2): S8-10.1186/1471-2164-12-S2-S8.PubMed CentralView ArticlePubMedGoogle Scholar
- Chakravorty S, Helb D, Burday M, Connell N, Alland D: A detailed analysis of 16S ribosomal RNA gene segments for the diagnosis of pathogenic bacteria. Journal of microbiological methods. 2007, 69 (2): 330-339. 10.1016/j.mimet.2007.02.005.PubMed CentralView ArticlePubMedGoogle Scholar
- Li W, Godzik A: Cd-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences. Bioinformatics. 2006, 22 (13): 1658-1659. 10.1093/bioinformatics/btl158.View ArticlePubMedGoogle Scholar
- Edgar RC: Search and clustering orders of magnitude faster than BLAST. Bioinformatics. 2010, 26 (19): 2460-2461. 10.1093/bioinformatics/btq461.View ArticlePubMedGoogle Scholar
- Hao X, Jiang R, Chen T: Clustering 16S rRNA for OTU prediction: a method of unsupervised Bayesian clustering. Bioinformatics. 2011, 27 (5): 611-618. 10.1093/bioinformatics/btq725.PubMed CentralView ArticlePubMedGoogle Scholar
- Rasheed Z, Rangwala H: Metagenomic taxonomic classification using extreme learning machines. J Bioinform Comput Biol. 2012, 10: 1250015-10.1142/S0219720012500151.View ArticlePubMedGoogle Scholar
- Huse S, Dethlefsen L, Huber J, Welch D, Relman D, Sogin M: Exploring microbial diversity and taxonomy using SSU rRNA hypervariable tag sequencing. PLoS genetics. 2008, 4 (11): e1000255-10.1371/journal.pgen.1000255.PubMed CentralView ArticlePubMedGoogle Scholar
- Liu Z, DeSantis T, Andersen G, Knight R: Accurate taxonomy assignments from 16S rRNA sequences produced by highly parallel pyrosequencers. Nucleic acids research. 2008, 36 (18): e120-e120. 10.1093/nar/gkn491.PubMed CentralView ArticlePubMedGoogle Scholar
- Wang Q, Garrity G, Tiedje J, Cole J: Naive Bayesian classifier for rapid assignment of rRNA sequences into the new bacterial taxonomy. Applied and environmental microbiology. 2007, 73 (16): 5261-5267. 10.1128/AEM.00062-07.PubMed CentralView ArticlePubMedGoogle Scholar
- Huang X: On global sequence alignment. Computer applications in the biosciences: CABIOS. 1994, 10 (3): 227-235.PubMedGoogle Scholar
- Schloss P, Westcott S, Ryabin T, Hall J, Hartmann M, Hollister E, Lesniewski R, Oakley B, Parks D, Robinson C, et al: Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Applied and environmental microbiology. 2009, 75 (23): 7537-7541. 10.1128/AEM.01541-09.PubMed CentralView ArticlePubMedGoogle Scholar
- Schloss P, Handelsman J: Introducing DOTUR, a computer program for defining operational taxonomic units and estimating species richness. Applied and environmental microbiology. 2005, 71 (3): 1501-1506. 10.1128/AEM.71.3.1501-1506.2005.PubMed CentralView ArticlePubMedGoogle Scholar
- Sun Y, Cai Y, Liu L, Yu F, Farrell M, McKendree W, Farmerie W: ESPRIT: estimating species richness using large collections of 16S rRNA pyrosequences. Nucleic Acids Research. 2009, 37 (10): e76-e76. 10.1093/nar/gkp285.PubMed CentralView ArticlePubMedGoogle Scholar
- Chao A: Nonparametric estimation of the number of classes in a population. Scandinavian Journal of Statistics. 1984, 265-270.Google Scholar
- Krebs C, et al: Ecological methodology. Addison Wesley Longman Menlo Park. 1999Google Scholar
- Chao A, Lee S: Estimating the number of classes via sample coverage. Journal of the American Statistical Association. 1992, 87 (417): 210-217. 10.1080/01621459.1992.10475194.View ArticleGoogle Scholar
- Schloss P, Westcott S: Assessing and improving methods used in operational taxonomic unit-based approaches for 16S rRNA gene sequence analysis. Applied and environmental microbiology. 2011, 77 (10): 3219-3226. 10.1128/AEM.02810-10.PubMed CentralView ArticlePubMedGoogle Scholar
- Hughes J, Hellmann J, Ricketts T, Bohannan B: Counting the uncountable: statistical approaches to estimating microbial diversity. Applied and Environmental Microbiology. 2001, 67 (10): 4399-4406. 10.1128/AEM.67.10.4399-4406.2001.PubMed CentralView ArticlePubMedGoogle Scholar
- Caporaso J, Kuczynski J, Stombaugh J, Bittinger K, Bushman F, Costello E, Fierer N, Pena A, Goodrich J, Gordon J, et al: QIIME allows analysis of high-throughput community sequencing data. Nature methods. 2010, 7 (5): 335-336. 10.1038/nmeth.f.303.PubMed CentralView ArticlePubMedGoogle Scholar
- Indyk P, Motwani R: Approximate nearest neighbors: towards removing the curse of dimensionality. Proceedings of the thirtieth annual ACM symposium on Theory of computing, ACM. 1998, 604-613.Google Scholar
- Buhler J: Efficient large-scale sequence comparison by locality-sensitive hashing. Bioinformatics. 2001, 17 (5): 419-428. 10.1093/bioinformatics/17.5.419.View ArticlePubMedGoogle Scholar
- Sogin ML, Morrison HG, Huber JA, Welch D, Huse SM, Neal PR, Arrieta JM, Herndl GJ: Microbial diversity in the deep sea and the underexplored "rare biosphere". Proc Natl Acad Sci U S A. 2006, 103 (32): 12115-12120. 10.1073/pnas.0605127103.PubMed CentralView ArticlePubMedGoogle Scholar
- Rasheed Z, Rangwala H, Barbara D: LSH-Div: Species diversity estimation using locality sensitive hashing. Bioinformatics and Biomedicine (BIBM), 2012 IEEE International Conference on: 4-7 October 2012. 2012, 1-6. 10.1109/BIBM.2012.6392649.View ArticleGoogle Scholar
- Rasheed Z, Rangwala H, Barbar´a D: Efficient Clustering of Metagenomic Sequences using Locality Sensitive Hashing. SIAM International Conference on Data Mining (SDM12). 2012, 1023-1034.View ArticleGoogle Scholar
- Felsenstein J: PHYLIP: phylogenetic inference package, version 3.5 c. 1993Google Scholar
- Rangwala H, Karypis G: Profile-based direct kernels for remote homology detection and fold recognition. Bioinformatics. 2005, 21 (23): 4239-4247. 10.1093/bioinformatics/bti687.View ArticlePubMedGoogle Scholar
- Pevzner P, Tang H, Waterman M: An Eulerian path approach to DNA fragment assembly. Proceedings of the National Academy of Sciences. 2001, 98 (17): 9748-9753. 10.1073/pnas.171285098.View ArticleGoogle Scholar
- Schloss P: The effects of alignment quality, distance calculation method, sequence filtering, and region on the analysis of 16S rRNA gene-based studies. PLoS computational biology. 2010, 6 (7): e1000844-10.1371/journal.pcbi.1000844.PubMed CentralView ArticlePubMedGoogle Scholar
- Grice E, Kong H, Conlan S, Deming C, Davis J, Young A, Bouffard G, Blakesley R, Murray P, Green E, et al: Topographical and temporal diversity of the human skin microbiome. science. 2009, 324 (5931): 1190-1192. 10.1126/science.1171700.PubMed CentralView ArticlePubMedGoogle Scholar
- Costello E, Lauber C, Hamady M, Fierer N, Gordon J, Knight R: Bacterial community variation in human body habitats across space and time. Science. 2009, 326 (5960): 1694-1697. 10.1126/science.1177486.PubMed CentralView ArticlePubMedGoogle Scholar
- Huse S, Huber J, Morrison H, Sogin M, Welch D, et al: Accuracy and quality of massively parallel DNA pyrosequencing. Genome Biol. 2007, 8 (7): R143-10.1186/gb-2007-8-7-r143.PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.