Detection of genomic sequences in microbial genomes by whole genome sequencing
Through local sequencing depth estimation and adaptive threshold strategy, the false positive and false negative problems in detecting microbial genome sequences in the next generation sequencing technology are solved, and the high-precision identification of the SARS-CoV-2 virus cloning complex is achieved, which improves the Accuracy and reliability of the detection.
Patent Information
- Application Number
- CN202380031801.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2022-04-05
- Filing Date
- 2023-04-04
- Publication Date
- 2024-11-15
AI Technical Summary
Next-generation sequencing technology has problems with false positive and false negative results when detecting specific genomic sequences in the microbial genome. Especially when sequencing depth changes, it is difficult to accurately identify clonal complexes of microorganisms such as SARS-CoV-2 viruses. .
By calculating the estimate of the target genome sequence at its local sequencing depth, combined with the count of other genome sequences in the neighborhood, an adaptive thresholding strategy is used to detect the presence of target genome sequences in the microbial genome, reducing the false positive and false negative rates.
It realizes the high-precision detection of specific genomic sequences in the microbial genome under the changes in sequencing depth, which improves the accuracy and reliability of the detection and reduces the occurrence of false positive and false negative results.
Smart Images

Figure CN118974833A_ABST
Abstract
Description
Field of the Invention
[0001] The present invention relates to the field of microbial molecular analysis, and in particular to identifying genome sequences in sequenced genomes using whole genome sequencing technology.
[0002] Advantageously, the present invention is suitable for identifying the subspecies taxonomic unit to which a clonal complex of a microorganism, such as the SARS-COVID-2 virus, belongs. Background Art
[0003] Next generation (whole genome or targeted) sequencing ("NGS"), such as that performed by Illumina and Oxford Nanopore Technologies platforms, has greatly changed the microbial analysis of patient, food or environmental samples by overcoming common microbial analysis workflow deficiencies, such as low agnosticism of the microorganisms in the sample, workflow duration or workflow complexity. If NGS technology shows many advantages, it requires a complex bioinformatics pipeline to process the reads generated by the sequencing platform. In fact, in general, the pipeline relies on specific genomic sequence detection, such as determining the clonal complex affiliation of a virus or the antibiotic sensitivity of a bacterium, which can be a challenging task because the genomic sequence may not be so present in the read or due to high sequencing error rates.
[0004] To confirm the presence of a specific sequence in a genome, a natural approach is to first calculate the number of copies of the sequence in the reads and the average sequencing depth of the genome, and then conclude that the sequence exists if the number of copies of the sequence exceeds a predetermined fraction of the average sequencing depth, as described in patent applications WO2021180771 and WO2021180768. The concept behind this pipeline relies on the fact that if the number of copies of a specific sequence is found to be limited compared to the total number of sequences generated by the sequencing platform, it means that these copies come from sequencing errors. It can therefore be concluded that the specific sequence is not actually present in the genome.
[0005] However, in addition to the fact that the sequencing depth varies from genome to genome due to the WGS technology itself, an enrichment step aimed at increasing the number of certain genome parts can also be performed before the sequencing step. For example, in SARS-CoV2 variant identification, sequencing usually occurs after a preliminary step of genome amplification by large-scale multiplex PCR. Therefore, the sequencing depth can vary greatly from genome to genome. On the one hand, if the actual sequencing depth of the target sequence position in the genome is much lower than the average depth, the pipeline is likely to give a false-negative result. On the other hand, if the actual sequencing depth of the target sequence position in the genome is much higher than the average depth, the pipeline is likely to give a false-positive result. Summary of the invention
[0006] The aim of the present invention is to provide a robust method for detecting specific genomic sequences (hereinafter referred to as "target genomic sequences") in a microbial genome in a set of reads generated by an aNGS workflow, even when the sequencing depth is highly variable.
[0007] To this end, one object of the present invention relates to a method for detecting a target genomic sequence at a predetermined position in a microbial genome, comprising:
[0008] - preparing a sample comprising at least one copy of the genome of said microorganism;
[0009] - sequencing the sample to generate a first set of digital genome sequences corresponding to the genome of the microorganism;
[0010] -And for each target genomic sequence, the computer performs the following steps:
[0011] A. calculating the local sequencing depth of the microbial genome in the neighborhood of the position of the target genome sequence;
[0012] B. if the copy number of the target genomic sequence is higher than a predetermined fraction of the calculated local sequencing depth,
[0013] then detecting that the target genome sequence exists in the microbial genome,
[0014] Wherein, the step of calculating the local sequencing depth comprises the following steps:
[0015] A1. Detecting a second set of digital genome sequences in the first set of digital genome sequences, wherein the second set comprises digital genome sequences from at least one reference genome of the microorganism;
[0016] A2. Generate a third set of digital genome sequences, which includes:
[0017] i. a first set of digital genomic sequences belonging to the neighborhood of said position of said target genomic sequence;
[0018] ii. a second set of digital genomic sequences belonging to the neighborhood of said position of said target genomic sequence;
[0019] A3. Counting the number of copies of each digital genome sequence in the third set of digital genome sequences; and
[0020] A4. Calculate the local sequencing depth equal to the maximum value of the counted copy number.
[0021] In the sense of the present invention, microorganisms are viruses, bacteria, yeasts or fungi. In the sense of the present invention, the genome of bacteria, yeasts or fungi comprises the chromosome and the plasmid part.
[0022] In other words, the detection according to the present invention relies on the cooperation of the following concepts:
[0023] i. If the number of copies of the target genome sequence in the read exceeds a certain fraction of the local sequencing depth, the target genome sequence is confirmed to exist in the microbial genome. By doing so, the variation in the actual sequencing depth of the genome is taken into account, thereby reducing the false positive rate and false negative rate observed when using the average sequencing depth;
[0024] ii. The estimate of local sequencing depth depends on the count of the target genome sequence in the read, and also on the count of other target genome sequences in its neighborhood, as well as the count of similar genome sequences that may come from mutations of the target genome sequence observed in the microbial genome. The estimate of local sequencing depth is ultimately defined as the highest number of counts found in the neighborhood of the target sequence, which more strongly reflects the actual composition of the genome of the organism at the position of the target sequence than the average sequencing depth of the entire genome.
[0025] iii. In addition, due to sequencing errors, the counts may not perfectly reflect the actual local sequencing depth. Therefore, if the target sequence is not the most abundant sequence in its neighborhood, it cannot be simply said that the target sequence is not present in the genome. By using the fraction of the estimated local sequencing depth, and thus the fraction of the most abundant sequence in the neighborhood of the target sequence, sequencing errors can be explicitly taken into account. It is important to note here the synergistic relationship between local sequencing depth and considering error rate, as the latter cannot be managed without first defining the local sequencing depth.
[0026] According to one embodiment of the present invention, the target genome sequence is a k-mer, which is a component of a variable of a model for predicting that the microorganism belongs to a clonal complex. Specifically, the microorganism is a coronavirus, and the length of the k-mer is between 15 and 25.
[0027] According to one embodiment of the present invention, the genomic sequences of the second set of genomic sequences are regularly spaced.
[0028] The present invention also relates to a system for detecting a target genome sequence at a predetermined position in a microorganism genome, comprising a computing unit configured to perform the above steps A and B. The present invention also relates to a computer-readable medium storing instructions for executing a method executed by a computer, the method comprising the above steps A and B. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The present invention will be better understood by reading the following description, given by way of example only, in conjunction with the accompanying drawings, in which like reference characters indicate the same or similar elements, and in which:
[0031] Figure 1 shows the adaptive threshold definition for a given k-mer (circled with bold outline) based on the number of occurrences of its neighboring k-mers (circled with dashed outline);
[0032] Figure 2 The performance and support of the model selected by cross-validation are shown with k = 21;
[0033] Figure 3 The k-mers selected in the model naturally capture lineage-defining mutations;
[0034] Figure 4 The confusion matrix obtained based on the assembled test set is given, involving 2000 genomes for each evolutionary branch;
[0035] Figure 5 The confusion matrices obtained when making predictions from all available reads (endpoint strategy) based on the ONT test dataset, with or without relying on the adaptive threshold method, are given;
[0036] Figure 6 The effect of the adaptive threshold according to the present invention on samples misclassified using the average threshold strategy is shown;
[0037] Figure 7 These minimum sequencing depths are given along with the differences in performance achieved for each clade. DETAILED DESCRIPTION OF THE INVENTION
[0039] A. Performing embodiments of the invention to detect SARS-CoV-2 viral clone complexes
[0040] Now, one embodiment of the present invention is described with respect to predicting the clonal complex to which the SARS-COV-2 virus infecting a patient belongs using the readouts generated by the ONT platform.
[0041] A.1 Introduction
[0042] Since the start of the COVID-19 pandemic in late 2019, many variants of the original SARS-CoV-2 strain have frequently emerged (van Dorp et al., 2020). Some of these variants exhibit highly altered viral properties, such as increased transmissibility and associated disease severity, or reduced effectiveness of treatments or vaccines, or reduced performance of diagnostic tools, or failure of other public health and social measures. Therefore, surveillance and evaluation of these various SARS-CoV-2 variants have become critical to managing the ongoing COVID-19 pandemic (WHO, 2021). To prioritize global surveillance and research, the World Health Organization (WHO) and the Centers for Disease Control and Prevention (CDC) have identified specific variants of concern (VOI) and variants of concern (VOC) that pose a greater risk to global public health and warrant additional measures to prevent their spread (WHO, 2021; CDC, 2021; Parums, 2021). Sequencing played an important role in identifying SARS-CoV-2 as the causative agent of COVID-19 (Wu et al., 2020) and has now become key to understanding the dynamics of the virus and monitoring its spread. Two major reference tools, Pangolin (O'Toole et al., 2021) and Nextclade (Aksamentov et al., 2021), are used to identify SARS-CoV-2 variants from assembled genomes. Both tools focus on detecting point mutations within various proteins, including the spike protein.
[0043] Next-generation sequencing platforms, commonly used for routine whole-genome sequencing, have been successfully used for SARSCoV-2 sequencing. This includes the mainstream Illumina platform and Oxford Nanopore Technologies (ONT). Although both technologies allow multiple samples to be sequenced together (multiplexing), Illumina sequencing is still too time-consuming and costly to be used for timely diagnosis. The ONT compact instrument provides faster turnaround time by analyzing sequencing reads in real time (Magi et al., 2017). This real-time feature is critical in many clinical applications, such as for treatment adjustment, where the targeted virus within the same host may be silenced and evade treatment due to the presence of the SARS-CoV-2 virus. However, the error rate of ONT reads is high (e.g., the error rate of the R9.4 flow cell is ∼5%), so appropriate bioinformatics pipelines are required to obtain accurate results, especially for the detection of point mutations (Delahaye and Nicolas, 2021). Finally, for SARS-CoV-2, polymerase chain reaction (PCR) that produces amplicons has been used to amplify SARS-CoV-2 genetic material before sequencing library preparation. Currently, the most widely used protocol for SARS-CoV-2 sequencing is the tiling amplicon-based method designed by the ARTIC Network (Quick et al., 2017). In this approach, primer sets targeting amplicons of different lengths, typically 400-2000 bases, are designed to optimally cover the SARS-CoV-2 genome. Several primer sets are now commercially available and used for SARS-CoV-2 sequencing (Itokawa et al., 2020; Tyson et al., 2020). However, differences in primer efficiency and amplicon yield have been observed depending on the primer kit used. This, in turn, leads to inconsistent genome sequencing depth (Charre et al., 2020; Liu et al., 2021).
[0044] The inventors propose a robust method based on ONT sequencing that requires only a limited amount of sequencing data, enabling ultra-fast variant identification with high accuracy. This method relies on the k-mer representation and sparsity of the SARS-CoV-2 genome to promote machine learning models to identify concise genomic features that can be efficiently and accurately detected from the reads. To improve the accuracy of the model, the inventors' method specifically addresses the high error rate of ONT sequencing and the inconsistent sequencing depths obtained using the ARTIC scheme by relying on local estimates of sequencing depth to optimally detect k-mers embedded in the features.
[0045] A.2. Dataset
[0046] Two datasets were assembled for this study. The first dataset contains assembled SARS-CoV-2 genomes from the GISAID database (Khare et al., 2021). It was used to learn and validate machine learning models. The second dataset consists of ONT sequencing reads and matched assembled genomes. It is used to illustrate the feasibility of predictions directly from real-time reads without any loss in performance compared to predictions from assembled genomes.
[0047] A.2.1. Assembled genome
[0048] At the heart of the method is a machine learning model trained to identify SARS-CoV-2 variants. In this work, it was considered the Nextstrain nomenclature to define variants and relied on assembled genomes taken from the GISAID database to learn and validate the model. More precisely, the method relies on a database of May 14, 2021, consisting of 1:508:656 genomes representing 13 clades. From this database, 2000 genomes were selected for training and 2000 for independent validation.
[0049] A.2.2.ONT reading
[0050] To demonstrate the feasibility of ultra-fast identification of SARSCoV-2 variants from ONT readings, a smaller data set was assembled, with 100 genomes for each clade. ONT readings were downloaded from Sequence Read Archive (SRA), and corresponding assemblies were extracted from the GISAID database for comparison. None of the genomes included in this data set is part of the data set for training and verifying the machine learning model. Table 1 provides the number of genomes assembled for each clade for training and verifying the model and the number of ONT genomes (readings) for testing the model. The first column gives the clade name, followed by the total number of genomes available in the corresponding GISAID export, the 2nd and 3rd columns correspond to the number of genomes for training and verifying the model, and the 4th column corresponds to the ONT reading data set for testing the model.
[0051] A.3. Methods
[0052] A.3.1. Genome sequence presentation
[0053] The inventors' method relies on the k-mer presentation of the SARS-CoV-2 genome sequence. The DBGWAS software is used to calculate the k-mer presentation, as described in patent applications WO2021180771 and WO2021180768 by Jaillard et al. (2020). In short, for a given value of k, a De Bruijn graph is constructed from all assembly sets within the training dataset. The nodes of this graph correspond to k-mers, which are compressed into so-called unitigs of completely conserved regions, resulting in k-mers of variable length. To further reduce the dimensionality, equivalent unitigs are grouped together as features. The final presentation of a given genome is obtained by a binary vector encoding the presence or absence of all these features and used as input to the machine learning model.
[0054] Table 1
[0055]
[0056]
[0057] A.3.2. Model training with assembly set
[0058] A lasso-penalized multinomial logistic regression model was trained to identify genomic variants as defined by the Nextclade nomenclature from the k-mer based representation described above.
[0059] Multinomial logistic regression is a widely used generalized linear model for solving multi-classification problems. In our case, it consists of a vector x∈{0,1} p The genome represents a set of K linear functions f k (.), defined as:
[0060]
[0061] where K is the number of Nextstrain clades considered (13 in this study), p is the number of features involved in the k-mer representation, and are model coefficients. The model defines the probability of belonging to each K clade as:
[0062]
[0063] k∈{1,…,K}, the final prediction corresponds to the clade with the highest probability.
[0064] To estimate the model coefficients and select a limited number of relevant features simultaneously, standard Lasso-penalized maximum likelihood estimation was used. More formally, from a set of n genomes (x i ,yi ) starts, where y i ∈{1,…,K} represents the actual evolutionary branch of the ith genome and x i ∈{0,1} p Its k-mer representation solves the following optimization problems:
[0065]
[0066] in and -logP(Y=y i |x i ) is defined according to Equation 1:
[0067]
[0068] The regularization parameter λ controls the trade-off between prediction accuracy and model sparsity and is chosen via cross-validation to maximize model accuracy. In practice, we rely on R software, more precisely the glmnet package (Friedman et al., 2010), to estimate the model coefficients and optimize the regularization parameter.
[0069] A.3.3. Prediction from the assembly set
[0070] The prediction step simply amounts to detecting the presence of features (a set of unitigs) selected by the model in the sequencing data. To do this from an assembled genome, a three-step process is required. Since unitigs are of variable length (between 21 and 41 in this study), we first use MUMmer ( et al., 2018) detect all k-mers of length 21, and then call a unitig present when all its constituent k-mers are detected without errors (no mismatches). Finally, when multiple unitigs define a given feature, all unitigs are required to be present.
[0071] A.3.4. Prediction from readings
[0072] The strategy used for assembly-based prediction was adjusted to allow predictions to be made directly from reads, including a threshold for the number of occurrences of each k-mer, while keeping the rest of the procedure unchanged.
[0073] As mentioned above, different approaches can be used to define the k-mer detection threshold. The first natural strategy is to rely on a unique threshold T, defined as a fraction α of the average sequencing depth. This average sequencing depth is calculated by dividing the total number of sequenced base pairs (N bps) divided by the length of the reference SARS-CoV-2 genome MN908947 (equal to 29903 bp, see, e.g., Wu et al. (2020)):
[0074]
[0075] However, as mentioned in the introduction, the ARTIC protocol used for SARS-CoV-2 sequencing includes an enrichment step, which causes the sequencing depth to vary across genomes. In addition, the sample composition can also lead to inconsistent sequencing depth, especially in the presence of contamination.
[0076] To address this limitation, the present invention proposes an alternative strategy that relies on adaptive thresholds. This approach relies on local estimates of sequencing depth to define k-mer specific detection thresholds. For local estimates of sequencing depth, one embodiment of the present invention advantageously utilizes k-mers used by the fastv software (Chen et al., 2020), which is designed to assess the quality of sequencing data using k-mers generated at regular intervals along a given genome. Therefore, these k-mers generated by fastv are regularly spaced along the SARS-CoV-2 genome, thus providing a natural surrogate for estimating local sequencing depth. More precisely, embodiments of the present invention proceed as described below:
[0077] 1. First, fastv k-mer and the k-mer involved in the model were mapped on the reference SARS-CoV-2 genome MN908947 (Wu et al., 2020), and for each k-mer k i Its position pos(k i ) stored on the reference genome.
[0078] 2. For a given window size δ, and for each k-mer k i , the neighborhood is defined according to Equation 2 Integrate the model's k-mer and fastv k-mer, which is k away from the k-mer i Within the distance δ:
[0079]
[0080] in Represents the union of the fastv k-mer set and the k-mer set involved in the prediction model.
[0081] 3. Local sequencing depth is estimated as iThe maximum number of observed k-mer occurrences (occ) within a dataset, since both fastv k-mers and k-mers from the model can capture different mutations relative to the reference genome.
[0082] 4. Finally, as shown in Equation 3, with k-mer k i The relevant threshold T i is defined as a fraction α of the local sequencing depth estimate. Note, however, that depending on the size of the neighborhood δ, a k-mer may remain isolated. In this case, we rely on the median of all k-mer occurrences (features and fastv) to estimate the local sequencing depth.
[0083]
[0084] In fact, α and δ are two hyperparameters that can be advantageously optimized to improve performance.
[0085] Figure 1 The example illustrates the different steps for defining such an adaptive k-mer detection threshold based on k-mers derived from a reference genome and model k-mers.
[0086] A.3.5. Real-time / streaming analysis
[0087] When identifying variants from ONT reads, two prediction settings are advantageously considered. The endpoint setting corresponds to making a single prediction from all sequencing reads available at the end of a sequencing experiment, as is done in traditional Illumina-like sequencing workflows.
[0088] However, since the ONT workflow offers the possibility to process sequencing data in real time, we evaluated the added value of this unique feature of ONT sequencing by simulating the real situation of real-time generation of reads by a sequencer. To do this, we divided the sequencing data downloaded from SRA into chunks of 1000 reads. This made it possible to study the evolution of prediction performance as sequencing depth and sequencing time increased.
[0089] A.4. Results
[0090] A.4.1. Cross-validation performance
[0091] The prediction model was trained using several k values between 11 and 31. The best k value was selected based on the cross validation results. The inventors observed that the k value had little impact on the performance and support of the model. The value of 21 was chosen because it was the best trade-off between the performance and sparsity of the model. The total number of features involved in the model was 430. Figure 2 The performance and support of the model selected by cross validation with k=21 is shown.
[0092] Figure 3 The example illustrates that the k-mers selected in the model naturally capture lineage-defining mutations. In this example, the k-mers corresponding to the 20J / 501Y.V3 clade were mapped against the reference genome of SARS-CoV-2. This set of k-mers captures the AAT at position 23063, which corresponds to the absence of the well-known N501Y mutation in the spike protein, while its presence is associated with the 20J / 501Y.V3 clade.
[0093] A.4.2. Assembly Performance
[0094] Figure 4 The confusion matrix obtained based on the assembled test set is given, involving 2000 genomes for each clade. The overall accuracy of the model is 99.7%. This performance level is very consistent with the cross-validation performance, indicating the good generalization properties of the machine learning model of the present invention and the method for detecting k-mer based on local sequencing depth.
[0095] A.4.3. Performance based on ONT readings
[0096] ONT reads from the test dataset were processed using our prediction pipeline, taking into account both endpoint and stream prediction strategies. For comparison, the prediction model was also tested on the corresponding assembly set from GISAID. For the adaptive threshold method, the neighborhood size and threshold were arbitrarily set to δ = 31 and α = 0.5, respectively. These two hyperparameters were fine-tuned by a cross-validation procedure to obtain an appropriate level of performance.
[0097] Endpoint Performance: Figure 5 The confusion matrices obtained when making predictions from all available reads based on the ONT test dataset (endpoint strategy), with or without the adaptive threshold method are shown. These results are compared with the confusion matrices obtained from the assembled genome (shown on the left).
[0098] The overall prediction accuracy for the assembled genome was 99:7%, indicating that this small dataset is representative of a validation dataset with performance levels comparable to assembly-based predictions.
[0099] Figure 6The example illustrates the effect of adaptive thresholds on samples that are misclassified using the average threshold strategy but correctly classified after adaptive thresholding. The figure highlights how the adaptive threshold strategy affects k-mer detection. In all three figures, the black dots correspond to fastv k-mers, while the blue dots correspond to model-selected k-mers. The dashed orange line corresponds to the average k-mer detection threshold. In the two enlarged figures, the model k-mers detected for this sample are highlighted in yellow. The red dots correspond to the model k-mers detected using only the average threshold strategy. Specifically, several k-mers that appear in high sequencing depth regions (shown in red, approximately position 15000) are called by the global threshold, but are ultimately no longer called using the adaptive threshold, which is consistent with the calls made from the corresponding assembled genome. The flow performance analysis shows that for both depth estimation methods, the predicted performance reaches a steady state defined by the endpoint accuracy before the end of the sequencing experiment. For the average threshold method, the minimum depth to reach a steady state is about 750×, while for the adaptive threshold strategy, the minimum depth is about 150×. The inventors observed that in flow mode, the performance obtained using the adaptive depth method is also more stable than the average depth method. To better illustrate the differences in performance and speed, we calculated the minimum sequencing depth required to achieve endpoint performance for each clade. Figure 7 These minimum sequencing depths are presented along with the differences in performance achieved for each clade. This result clearly demonstrates that not only can higher performance levels be achieved by relying on an adaptive depth approach, but also that lower sequencing depths are required to achieve endpoint performance.
[0100] As can be seen above, methods relying on k-mers and sparsity facilitate machine learning models to identify SARS-CoV-2 variants directly from ONT reads. By using local sequencing depth estimates, the amount of sequencing data required to reliably detect k-mers can be greatly reduced while maintaining a level of variant identification performance comparable to that obtained from self-assembled genomes. Specifically, it has been demonstrated that this performance can be achieved with an average sequencing depth as low as 150×, which can be achieved in less than a minute of sequencing time if samples are not multiplexed on the ONT flow cell. Surprisingly, not taking this local, adaptive strategy into account leads to even lower performance and a 5-fold increase in the amount of sequencing data required.
[0101] Variant identification can then be obtained within a few minutes of sequencing time, depending on the amount of sample multiplexing. This therefore paves the way for timely diagnosis that could improve patient care in the context of the COVID-19 pandemic.
[0102] A.5. Processing the readings to determine the clonal complex of the Sars-CoV2 sample is a computer-implemented method, for example performed by a computing unit comprising one or more (micro)processors, computer memory and RAM, and configured to store computer-readable instructions that, when run by the computing unit, perform the above-mentioned calculations. The computing unit is, for example, a personal computer, a server of a cloud-based computer system, or a computing cluster. The processing results are advantageously output on a display and / or stored in a laboratory information system.
[0103] B. Other Implementations
[0104] - The detection of k-mers in a genome has been described. As will be readily appreciated, the term "k-mer" represents a specific way of describing a genome, a portion thereof, or more generally a genomic sequence. Thus, a k-mer is a genomic sequence, and its use in the description of the above predictions does not limit the scope of the invention;
[0105] - An adaptive threshold calculated based on local sequencing depth has been described for viruses. Obviously, the method is applicable to any genome of nature, which can be from bacteria, yeast or fungi. In particular, patent application WO2021180768 describes the prediction of bacterial sensitivity to antibiotics based on specific k-mer detection. The detection described in the application is advantageously carried out based on the present invention;
[0106] - some specific values of the hyperparameters α and δ have been described. Advantageously, as described above, said hyperparameters are part of the cross-validation technique variables and their values are chosen to optimize the performance of said cross-validation;
[0107] - The use of one reference genome has been described. Advantageously, multiple reference genomes, and preferably as many reference genomes as possible, are used to generate the reference k-mers.
[0108] References
[0109] Aksamentov,I.et al.(2021).Nextclade:clade assignment,mutation calling and quality control for viral genomes.Journal of Open Source Software,6(67),3773.
[0110] CDC(2021).Sars-cov-2variant classifications and definitions,extractedfrom:https: / / www.cdc.gov / coronavirus / 2019-ncov / variants / variantclassifications.html.
[0111] Charre,C.et al.(2020).Evaluation of ngs-based approaches for sars-cov-2whole genome characterisation.Virus evolution,6(2),veaa075.
[0112] Chen,S.et al.(2020).A computational toolset for rapid identificationof sars-cov-2,other viruses,and microorganisms from sequencing data.bioRxiv.
[0113] Delahaye,C.and Nicolas,J.(2021).Sequencing DNA with nanopores:Troubles and biases.PLoS ONE,16(10October).
[0114] Friedman,J.et al.(2010).Regularization paths for generalized linearmodels via coordinate descent.Journal of Statistical Software,33(1),1–22.
[0115] Itokawa,K.et al.(2020).Disentangling primer interactions improvessars-cov-2genome sequencing by multiplex tiling pcr.PloS one,15(9),e0239403.
[0116] Jaillard,M.et al.(2020).Interpreting k-mer–based signatures forantibiotic resistance prediction.GigaScience,9(10).giaa110.
[0117] Jouet,A.et al.(2021).Deep amplicon sequencing for culture-freeprediction of susceptibility or resistance to 13 anti-tuberculousdrugs.European Respiratory Journal,57(3).
[0118] Khare,S.et al.(2021).Gisaid’s role in pandemic response.China CDCWeekly,3(49),1049–1051.
[0119] Liu,T.et al.(2021).A benchmarking study of sars-cov-2 whole-genomesequencing protocols using covid-19 patient samples.Iscience,24(8),102892.
[0120] Magi,A.et al.(2017).Nanopore sequencing data analysis:state of theart,applications and challenges.Briefings in Bioinformatics,(June).
[0121] G.et al.(2018).Mummer4:A fast and versatile genome alignmentsystem.PLOS Computational Biology,14(1),1–14.
[0122] Meehan,C.J.et al.(2019).Whole genome sequencing of mycobacteriumtuberculosis:current standards and open issues.Nature Reviews Microbiology,17(9),533–545.
[0123] O’Toole,et al.(2021).Assignment of epidemiological lineages in anemerging pandemic using the pangolin tool.Virus Evolution,7(2).veab064.
[0124] Parums,D.V.(2021).Revised world health organization(who)terminologyfor variants of concern and variants of interest of sarscov-2.Medical ScienceMonitor:International Medical Journal of Experimental and Clinical Research,27,e933622–1.
[0125] Quick,J.et al.(2017).Multiplex pcr method for minion and illuminesequencing of zika and other virus genomes directly from clinicalsamples.Nature protocols,12(6),1261–1276.
[0126] Tyson,J.R.et al.(2020).Improvements to the artic multiplex pcr methodfor sars-cov-2genome sequencing using nanopore.BioRxiv.
[0127] van Dorp,L.et al.(2020).Emergence of genomic diversity and recurrentmutations in sars-cov-2.Infection,Genetics and Evolution,83,104351.
[0128] WHO(2021).Tracking sars-cov-2 variants,extracted from:https: / / www.who.int / en / activities / tracking-sars-cov-2-variants / .
[0129] Wu,F.et al.(2020).A new coronavirus associated with human respiratorydisease in china.Nature,579(7798),265—269。
Claims
1. A method for detecting a target genomic sequence at a predetermined position in a microbial genome, comprising: - preparing a sample comprising at least one copy of the genome of said microorganism; - sequencing the sample to generate a first set of digital genome sequences corresponding to the genome of the microorganism; - and, for each target genomic sequence, the computer performs the following steps: A. calculating the local sequencing depth of the microbial genome in the area of the location of the target genome sequence; B. if the copy number of the target genomic sequence is higher than a predetermined fraction of the calculated local sequencing depth, detecting that the target genomic sequence is present in the microbial genome, The step of calculating the local sequencing depth includes the following steps: A1. Detecting a second set of digital genome sequences in the first set of digital genome sequences, the second set comprising digital genome sequences from at least one reference genome of the microorganism; A2. Generate a third set of digital genome sequences, which includes: iii. a first set of digital genomic sequences belonging to the area of the location of the target genomic sequence; iv. a second set of digital genomic sequences belonging to the area of the location of the target genomic sequence; A3. Counting the number of copies of each digital genome sequence in the third set of digital genome sequences; and A4. Calculate the local sequencing depth to be equal to the maximum value of the counted copy number.
2. The method of claim 1, wherein the target genomic sequence is a k-mer that is a component of a variable of a model for predicting that the microorganism belongs to a clonal complex.
3. The method of claim 1, wherein the microorganism is a coronavirus and the length of the k-mer is between 15 and 25.
4. A method according to any of the preceding claims, wherein the genomic sequences of the second set of genomic sequences are regularly spaced.
5. A system for detecting a target genomic sequence at a predetermined position in a microorganism genome, comprising a computing unit configured to perform steps A and B according to any one of the preceding claims.
6. Computer-readable medium storing instructions for executing a method executed by a computer, the method comprising steps A and B according to any one of claims 1 to 4.
