Method for genotyping rare gene variants

Abstract

A method of genotyping one or more gene rare variants in a plurality of nucleic acid samples is described. The method includes running one or more assays of the plurality of nucleic acid samples using at least one microarray comprising a plurality of probe sets, and generating assay data. Heterozygous genotypes in the plurality of nucleic acid samples are invoked and evaluated to identify any rare heterozygous genotype calls from a probe set having at least one probe, wherein the number of heterozygous genotype calls for the probe set does not exceed a maximum threshold. Evaluating each identified rare heterogenotype call using a support vector machine prediction model or a supervised machine learning classification model to determine whether the identified rare heterogenotype calls include a true rare heterogenotype call, the support vector machine prediction model or the supervised machine learning classification model includes a plurality of prediction values for identifying true rare heterozygous genotype calls.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Provisional Application Serial No. 63 / 335,159, filed on April 26, 2022, and U.S. Provisional Application Serial No. 63 / 344,020, filed on May 19, 2022. The entire contents of these applications are incorporated herein by reference to the extent permitted by applicable jurisdiction. Background Art

[0003] The translation of biomarkers from discovery to routine clinical use is critical to the future of precision medicine. However, applying these discoveries and associating population-specific variants with individual traits, diseases, and potential downstream treatment options remains one of the most challenging barriers to personalized genomics. Although large population cohort studies and private sequencing initiatives have revealed thousands of single nucleotide polymorphisms (SNPs) and structural variants associated with nearly 2,000 human diseases and traits, many of these variants are not within genes and have no known biological function. A better understanding of the role of these variants requires studies optimized for populations and imputation. These studies must use carefully designed genotyping panels and reliable, rapid, and flexible genotyping platforms to integrate knowledge of disease, traits, and biological function.

[0004] In addition, the widespread use of next-generation sequencing (NGS) in clinical settings has led to the identification of many novel gene variants, including rare variants of genes. With the improvement of high-throughput sequencing technology, the genetics community is also showing a significant interest in the rare variant / common disease hypothesis. The common disease-rare variant hypothesis (CDRVH) assumes that if a disease with a genetic cause is common in a population (prevalence greater than 1% to 5%), then the genetic cause-specific gene error (gene variant or disease allele) will not necessarily be found to be common in the population, as indicated by the competitive common disease-common variant hypothesis (CDCVH), but will be composed of multiple risk alleles, each of which is rare in the population. It makes this prediction for diseases (polygenic diseases) where genetic contributions are believed to come from multiple genes at the same time. Autism is considered to be such a disease.

[0005] Genetic databases are now available to explain or associate gene variants for various health conditions. For example, ClinVar (https: / / www.ncbi.nlm.nih.gov / clinvar / ) from the National Center for Biotechnology Information (NCBI) is a freely available archive that explains the clinical significance of gene variants for reported conditions. The database includes germline and somatic variants of any size, type, or genomic location. GeneReviews TM , UniProt, expert panels, and practice guidelines. See Landrum et al., ClinVar: Public archive of interpretations of clinically relevant variants, Nucleic Acids Res. 2016 Jan 4; Vol 44 (Database Release): D862–D868 (https: / / www.ncbi.nlm.nih.gov / pmc / articles / PMC4702865 / )

[0006] Using databases such as ClinVar, rare variants of genes are associated with traits and diseases, and some researchers and healthcare providers want to design and use microarrays that detect the occurrence of such rare variants. However, genotyping rare variants is more challenging than genotyping common variants, and false heterozygous (Het) genotypes for rare variants called on microarray platforms can cause problems in data analysis for researchers and healthcare providers.

[0007] Very rare variants can be more difficult to genotype on microarray platforms because they use clustering algorithms to identify genotypes. While for common variants the location and shape of heterozygous (Het) clusters can provide strong evidence for the accuracy of the Het call, rare variants often have no other samples in the Het cluster, making the call less robust. Summary of the invention

[0008] A method for genotyping one or more rare variants of a gene in a plurality of nucleic acid samples is described. The method includes running one or more assays of the plurality of nucleic acid samples using at least one microarray comprising a plurality of probe sets, and generating assay data. Heterozygous genotypes in the plurality of nucleic acid samples are called and evaluated to identify any rare heterozygous genotype calls, wherein the number of heterozygous genotype calls for the probe set does not exceed a maximum threshold. Each identified rare heterozygous genotype call is evaluated using a support vector machine prediction model to determine whether the identified rare heterozygous genotype call includes a true rare heterozygous genotype call, the support vector machine prediction model including a plurality of prediction values ​​for identifying true rare heterozygous genotype calls.

[0009] Additional aspects and advantages of the present disclosure will be apparent to those skilled in the art from the following detailed description, in which only illustrative embodiments of the present disclosure are shown and described. As will be appreciated, the present disclosure is capable of other different embodiments, and its several details can be modified in various obvious aspects, all without departing from the present disclosure. Therefore, the drawings and description are to be regarded as illustrative rather than restrictive in nature. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0011] Figure 1 is a flow chart illustrating a method for genotyping rare genetic variants according to various embodiments of the present disclosure.

[0012] Figure 2 A genotyping clustering table according to various embodiments of the present disclosure is illustrated.

[0013] 3 is a flow chart illustrating a method for rare heterozygous adjustment (RHA) according to various embodiments of the present disclosure.

[0014] 4 is a flow chart further illustrating a method for rare heterozygous adjustment (RHA) according to various embodiments of the present disclosure.

[0015] Figure 5 A graph showing acceptable ranges of Z-scores versus intensity distribution is illustrated in accordance with various embodiments of the present disclosure.

[0016] Figure 6 Graphs and data tables showing the variation in genotype calling accuracy for representative data sets are illustrated according to various embodiments of the present disclosure.

[0017] Figure 7 Illustrated is a graph showing the results of the RHA algorithm used in one embodiment of the present disclosure.

[0018] Figure 8 is a flow chart illustrating a method of genotyping rare variants including support vector machine analysis according to some embodiments of the present disclosure.

[0019] Fig. 9 is a table of results for support vector machine analysis according to some embodiments of the present disclosure.

[0020] Fig. 10A Illustrated are graphs and data tables for various predictors used for support vector machine analysis according to some embodiments of the present disclosure.

[0021] Fig. 10B Illustrated are graphs and data tables for various training set sizes for support vector machine analysis according to some embodiments of the present disclosure.

[0022] Fig.11 is a table of test results for a support vector machine prediction model according to some embodiments of the present disclosure.

[0023] Fig.12 Illustrated is a graph of test result data for a support vector machine prediction model according to some embodiments of the present disclosure.

[0024] Fig.13 is a flow chart illustrating a method of genotyping rare variants including support vector machine analysis according to some embodiments of the present disclosure.

[0025] Fig.14 Illustrated are test result tables and test result data graphs for a support vector machine prediction model according to some embodiments of the present disclosure, where heterozygous calls for alleles are generated from probe sets with only one probe for each allele on a microarray.

[0026] Fig.15 is a high-level block diagram of an exemplary device that can be used to implement the systems, devices, and methods described herein.

[0027] Although the present invention is described with reference to the above drawings, the drawings are intended to be illustrative and other embodiments are consistent with the spirit of the invention and are within the scope of the invention. DETAILED DESCRIPTION

[0028] To provide a more thorough understanding of the present invention, the following description sets forth many specific details, such as specific configurations, parameters, examples, etc. However, it should be recognized that such description is not intended as a limitation on the scope of the present invention, but is intended to provide a better description of exemplary embodiments.

[0029] Throughout the specification and claims, unless the context clearly dictates otherwise, the following terms shall take the meanings expressly associated herein:

[0030] Although the phrase "in one embodiment" as used herein may refer to the same embodiment, it does not necessarily refer to the same embodiment. Therefore, as described below, the various embodiments of the present invention can be easily combined without departing from the scope or spirit of the present invention.

[0031] As used herein, the term "or" is an inclusive "or" operator, and is equivalent to the terms "and / or" unless the context clearly dictates otherwise.

[0032] The term "based on" is not exclusive and allows for being based on additional factors not described unless the context clearly dictates otherwise.

[0033] As used herein, and unless the context indicates otherwise, the term "coupled to" is intended to include direct coupling (where two elements coupled to each other are in contact with each other) and indirect coupling (where at least one additional element is located between the two elements). Therefore, the terms "coupled to" and "coupled with..." are used synonymously. In the context of a networked environment where two or more components or devices are able to exchange data, the terms "coupled to" and "coupled with..." are also used to mean "communicatively coupled with...", possibly via one or more intermediate devices.

[0034] Furthermore, throughout the specification, the meanings of “a”, “an” and “the” include plural references, and the meaning of “in” includes “in” and “on”.

[0035] Although some of the various embodiments provided herein constitute a single combination of elements of the present invention, it should be understood that the subject matter of the present invention is considered to include all possible combinations of disclosed elements. Therefore, if one embodiment includes elements A, B, and C, and another embodiment includes elements B and D, the subject matter of the present invention is also considered to include other remaining combinations of A, B, C, or D, even if not explicitly discussed herein. In addition, the transition term "comprising" means having parts or components, or these parts or components. As used herein, the transition term "comprising" is inclusive or open-ended, and does not exclude additional, unlisted elements or method steps.

[0036] In the following disclosure, reference may be made to servers, services, interfaces, engines, modules, clients, peers, portals, platforms, or other systems formed by computing devices. It should be understood that the use of these terms is considered to represent one or more computing devices having at least one processor (e.g., ASIC, FPGA, DSP, x86, ARM, ColdFire, GPU, multi-core processor, etc.), which are configured to execute software instructions stored on computer-readable tangible non-transitory media (e.g., hard disk drives, solid-state drives, RAM, flash memory, ROM, etc.). For example, a server may include one or more computers that operate as web servers, database servers, or other types of computer servers in a manner that performs the described roles, responsibilities, or functions. It should also be understood that the disclosed computer-based algorithms, processes, methods, or other types of instruction sets may be embodied as a computer program product that includes a non-transitory tangible computer-readable medium storing instructions that cause a processor to perform the disclosed steps. Various servers, systems, databases, or interfaces may exchange data using standardized protocols or algorithms, possibly based on HTTP, HTTPS, AES, public-private key exchange, web service APIs, known financial transaction protocols, or other electronic information exchange methods. The data exchange may take place over a packet-switched network, a circuit-switched network, the Internet, a LAN, a WAN, a VPN, or other types of networks.

[0037] As used in the description herein and throughout the claims that follow, when a system, engine, server, device, module, or other computing element is described as being configured to perform functions on data in a memory, the meaning of "configured to" or "programmed to" is defined as one or more processors or cores of the computing element being programmed by a set of software instructions stored in the memory of the computing element to perform the set of functions on target data or data objects stored in the memory.

[0038] It should be noted that any language for a computer should be understood to include any suitable combination of computing devices or network platforms, including servers, interfaces, systems, databases, agents, peers, engines, controllers, modules, or other types of computing devices that operate individually or collectively. It should be understood that the computing device includes a processor that is configured to execute software instructions stored on a tangible non-transitory computer-readable storage medium (e.g., a hard drive, FPGA, PLA, solid-state drive, RAM, flash memory, ROM, etc.). The software instructions configure or program the computing device to provide roles, responsibilities, or other functions, as discussed below with respect to the disclosed device. In addition, the disclosed technology may be embodied as a computer program product that includes a non-transitory computer-readable medium that stores software instructions that cause the processor to perform the disclosed steps associated with the implementation of a computer-based algorithm, process, method, or other instruction. In some embodiments, various servers, systems, databases, or interfaces use standardized protocols or algorithms, possibly based on HTTP, HTTPS, AES, public-private key exchange, web service API, known financial transaction protocols, or other electronic information exchange methods, to exchange data. Data exchange between devices may be performed over a packet-switched network, the Internet, a LAN, a WAN, a VPN, or other types of packet-switched networks; circuit-switched networks; cell-switched networks; or other types of networks.

[0039] In various embodiments, the systems and methods described herein can be used to detect one or more types of biological components of interest. These biological components of interest can be any suitable biological targets, including but not limited to DNA sequences (including cell-free DNA), RNA sequences, genes, oligonucleotides, molecules, proteins, biomarkers, cells (e.g., circulating tumor cells) or any other suitable target biomolecules.

[0040] In various embodiments, such biological components can be used with various polymerase chain reaction (PCR) methods and systems in applications such as multiplex PCR, viral detection and quantification standards, genotyping, sequencing verification, mutation detection, genetically modified organism detection, fetal diagnosis, rare allele detection, and copy number variation.

[0041] Embodiments of the present disclosure generally relate to systems and methods for genotyping one or more rare genetic variants in a plurality of biological samples.

[0042] Although generally applicable to digital quantification such as PCR, it will be appreciated that any other suitable quantification method may be used according to the various embodiments described herein. Suitable PCR methods include, but are not limited to, for example, digital PCR, allele-specific PCR, asymmetric PCR, ligation-mediated PCR, multiplex PCR, nested PCR, qPCR, genome walking, and bridge PCR.

[0043] As used herein, thermal cycling can include the use of, for example, a thermal cycler, isothermal amplification, thermal regulation, infrared-mediated thermal cycling, or helicase-dependent amplification.

[0044] According to various embodiments, detection of the target can be, but is not limited to, for example, fluorescence detection, cation or anion detection, pH detection, voltage detection, or current detection, alone or in combination.

[0045] It should be understood that the disclosed technology provides many advantageous technical effects, including automated methods for genotyping one or more rare gene variants in biological samples using analyte detection (e.g., PCR) equipment. The technology described herein uses logic to automate various processes, including processes currently performed manually. In addition, the disclosed technology has been designed to support data accuracy and allow data algorithms and complex arrangements to be processed at scales and speeds that cannot be achieved manually.

[0046] It should also be appreciated that the following description is not intended as an extensive overview and that, therefore, the concepts may be simplified for the sake of clarity and conciseness.

[0047] All publications, patents, and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference.

[0048] Embodiments of the invention discussed herein utilize microarray analysis. A microarray is a substrate to which are attached millions of single-stranded DNAs complementary to the DNA to be detected. Microarrays can come in several styles; for example, where the substrate is a small plastic or glass slide, such as a microscope slide or plastic beads. There can be from a few thousand to a million probes, each consisting of thousands of single strands with the same sequence. The probes are attached to the substrate of the microarray. Microarrays are typically species-specific or genotype-specific, although the same array can sometimes be used for closely related species, such as humans and chimpanzees. In microarray analysis, it is usually necessary to know what sequence to study, because the microarray requires known sequences to be used as probes. Data capture cannot usually occur without the probes. In addition, genetic variation affects the intensity of complementary DNA hybridization.

[0049] The probes or probe sets need to be selected to provide adequate sensitivity (i.e., the ability to detect rarely expressed transcripts in a complex background) and specificity (i.e., the ability to discriminate the measurement among transcripts with high sequence similarity), as well as high coverage (i.e., the ability to include all relevant transcripts in the experiment). It is desirable to avoid sequences that are ambiguous (i.e., hybridize to multiple transcripts) or highly similar to non-target transcripts (i.e., cross-hybridize). In addition, redundancy (i.e., several probes or probe sets targeting the same transcript) can increase the accuracy of the measurement, but at the same time can reduce coverage. Furthermore, successful applications also require correct and up-to-date annotation of the probes or probe sets (i.e., the association of the probes with the target transcripts).

[0050] Figure 1 1 is a flow chart illustrating a method 100 for genotyping rare genetic variants according to various embodiments. In step 110, a microarray such as a conventional array is designed, or a pre-designed array is obtained for genotyping analysis. In step 120, the target (sample under study) is prepared. In step 130, the assay is prepared. In step 140, the array is processed.

[0051] In one embodiment of the present invention, amplify and randomly fragment total genomic DNA from sample.These fragments are purified, resuspended and hybridized with complementary probes.After hybridization, the combined target is washed under stringent conditions to remove non-specific background.Each allele in the target DNA is queried via a multicolor connection event.After connection, the array is dyed and imaged.The intensity of the image of each allele (A or B) of the expected target is roughly proportional to the amount of each allele in the DNA sample.

[0052] The dye intensity for each probe is summed up by a scanning microscope, which essentially takes a picture of the microarray at the wavelength of the marker. The raw data is the reflected intensity of the marker at each pixel. This intensity is expected to be proportional to the amount of material in the labeled sample.

[0053] The basic data generated during the array processing step 140 may include a digitized picture of the array giving the label intensities.

[0054] Genotype calling and data analysis are performed at step 150. In some embodiments of the invention, genotyping analysis can be performed by genotyping analysis software manufactured by Thermo Fisher Scientific, such as Applied Biosystems TM Array Power Tools (APT), such as APT version 2.11.3 or other versions, or Applied Biosystems TM Axiom TMAnalysis Suite Software (AxAS), such as AxAS version 5.1, 5.1.1 or other versions. In some embodiments of the present invention, the TM Axiom TM The conventional arrays and catalog genotyping arrays manufactured can provide highly accurate genotyping for conventional variants and rare variants. It may be challenging to use genotyping arrays for extremely rare variants (e.g., those with a minor allele frequency (MAF) less than 0.01%). Therefore, it is necessary to improve the accuracy of calling such variants with a genotyping algorithm customized for rare variants or very rare variants. Such algorithms for adjusting rare heterozygote calls may be included in post-analysis processing step 160, and / or used in combination with genotyping analysis software or algorithms such as Applied Biosystems TM Array Power Tools (APT), such as APT version 2.11.3 or other versions, or Applied Biosystems TM Axiom TM Analysis Suite Software (AxAS), such as AxAS version 5.1, 5.1.1 or other versions, including Axiom GT1 genotyping. The genotyping algorithm for rare heterozygote (Het) adjustment related to embodiments of the present invention discussed herein can be used together with other genotyping analysis software or algorithms of other manufacturers known in the art. Such software will allow the user to view and export data analysis results at step 170.

[0055] The genotyping array of commercially available targeted rare variants is proved to be a kind of reasonable alternative although order-checking is still forbidden for a large number of studies.The technical challenge of array-based method is to derive the task of genotype category (homozygous or heterozygous) by clustering intensity data points.The performance of clustering tools for common polymorphisms has been well established, and when carrying out with large-scale rare variants (wherein data points are sparse for the genotype comprising rare alleles), their performance is more unknown.In one embodiment of the invention, in Axiom GT1 genotyping process, use the clustering algorithm with very high accuracy to carry out genotyping.

[0056] Figure 2The genotype cluster graph according to various embodiments is illustrated.In one embodiment of the present invention, AxiomGT1 is a tuned version of BRLMM-P (Affymetrix, 2007) clustering algorithm, which adapts the pre-positioned genotype cluster position (prior) to sample data and calculates three a posteriori cluster positions.The genotype cluster position for the data with A allele and B allele is defined by the 2D mean and variance for AA, AB and BB genotype cluster distribution.The prior can be universal, meaning that the same pre-positioned position is provided for each SNP; or the prior is SNP-specific, meaning that different pre-positioned positions are provided on the basis of each SNP.As shown in genotyping cluster graphs 210 and 220, clustering is carried out in two dimensions.The X dimension is called "contrast", and the Y dimension is called "size". They are the logarithm-linear combination of two allele signal intensities.For allele A and B, the contrast is log2(A / B), and the size is (log2(A)+log2(B)) / 2. The genotype cluster graph produced by Axiom Analysis Suite and Ps_Visualization uses size / contrast to or with the formula marking axis for size and contrast.Genotype call is carried out by identifying the genotype intensity cluster (AA, AB or BB) to which each sample most likely belongs.Sample is called by these genotypes in the cluster graph to be colored and shaped: BB call is red circle, AB call is green circle, and AA call is yellow circle.It should be noted that data can be colored according to other sample attributes, wherein there are different options in Ps_Visualization and SNP cluster graph.The array used in the embodiment of the present invention uses the clustering algorithm with very high accuracy to call SNP genotype as AA, AB or BB, and the clustering algorithm works well when multiple clusters are well filled, such as in graph 210, but when only a cluster is well filled, SNP genotype call may be less accurate, as shown in graph 220, wherein false positive het call may be more likely to be used for showing only the AB genotype call of an AB genotype.

[0057] Therefore, in one embodiment of the invention, the Axiom Rare Heterozygous Adjusted (RHA) genotyping algorithm optimizes the Axiom GT1 genotyping for very rare variants. Although the position and shape of the heterozygous (het) clusters provide strong evidence for the het calls of common variants, rare variants usually have a single sample in the het cluster, making the call less robust. In-depth analysis of the replicate probe signal distribution on the Axiom microarray shows that this distribution can be used to significantly improve the accuracy of heterozygous calls.

[0058] In one embodiment of the invention, the RHA algorithm is performed after the initial genotyping algorithm (e.g., Axiom GT1) to determine which predictions of rare heterozygous (Het) genotypes may be false. The genotype predicted by the false Het as determined by RHA is then set to "no call". After checking hundreds of different array results, the main point of the algorithm is that most false Het predictions occur due to a small amount of intensity higher or lower than expected on the microarray surface. These unexpected intensities can cause the summary probe set intensity of homozygous samples to be located away from the main homozygous genotype cluster in "signal contrast" to "signal size" (clustering) space, and enter a position that can be consistent with the heterozygous genotype cluster, thereby causing the AxiomGT1 genotyping algorithm to mistakenly predict Het.

[0059] In one embodiment of the invention, genotyping is performed using an initial genotyping algorithm (e.g., Axiom GT1) on an exemplary health resource such as the UK Biobank, which tracks the health and well-being of 500,000 people, all genotyped at Applied Biosystems by Thermo Fisher Scientific TM Axiom TM The whole exome sequencing (WES) for the participant subset is also available, for example, the initial WES release version for UK Biobank contains data for about 50,000 participants. In one embodiment of the invention, the Axiom GT1 algorithm for genotyping UK Biobank has incorporated the detection and elimination of unexpected intensity into its genotype call, but does not use higher stringency for low minor allele frequency (MAF) variants. However, it is particularly challenging to genotype the rare variants of Het in the population, because when the number of true Het is too low (or zero in the most extreme case) and the Het cluster cannot be correctly anchored, false Het prediction is more likely to occur. The RHA algorithm improves rare Het prediction by incorporating additional information in the test for unexpected intensity. These include: (1) the intensity difference of the replicate probe sequence on the array for the inferred Het sample, and (2) the position of the signal distribution of the inferred Het signal relative to the homozygous sample. In one embodiment of the present invention, the RHA algorithm eliminates (sets to NoCall) most false positive heterozygous calls, while true positive heterozygous calls are virtually untouched.

[0060] The RHA used in one embodiment of the present invention is summarized in the exemplary flow chart 300 of Figure 3. At step 310, the RHA algorithm is applied only to probe sets with two duplicate probes and a small number of Hets. Hets on probe sets that do not meet these requirements are retained at step 320.

[0061] In the UK Biobank release used in one embodiment of the invention, samples are genotyped in batches of approximately 5,000 samples, and RHA is applied only when the number of Het predictions in a particular batch is between 1 and max_nAB, as shown in step 330. In some embodiments of the invention, the RHA algorithm checks for heterozygous calls from probe sets that have two replicates on the array and have max_nAB with three or four or fewer heterozygous (Het) calls in the batch. Max_nAB is a parameter of the algorithm. Several metrics used by the algorithm also apply to probe sets with one replicate.

[0062] At step 340, all Het calls that meet the criteria at step 330 for very rare variants are tested for a pattern of "unexpected intensities," defined as inconsistency in intensity between the two replicate probe sequences for a Het. If there are no unexpected intensities, the prediction at step 320 is considered true, and the Het genotype call is retained.

[0063] Otherwise, the prediction undergoes further testing to determine whether the individual probe intensities are "Het-like". All Het calls for rare variants are tested for "unexpected intensity" patterns, which are defined as inconsistencies in intensity between the two replicate probe sequences for a Het. If there are no unexpected intensities, the prediction is considered true and the Het genotype call is retained.

[0064] In one embodiment of the invention, after summarizing the intensity of various replicas for the same probe, the AxiomGT1 genotyping algorithm aggregates the samples in the "signal contrast" to "signal size" (aggregation) space to obtain genotypes. Due to the position of the probe sequence on the array, different replicas of the probe sequence for the same probe set may have slightly different intensity distributions. The slight inhomogeneity on the array surface may cause the intensity of the probe to be lower or higher than its due intensity.

[0065] In the case of RHA as an additional enhancement to the initial genotyping algorithm (such as Axiom GT1), if a Het prediction is a rare Het (Het cluster size <= max_nAB) and meets one of the two criteria for unexpected intensity, the Het prediction is marked as suspicious and may require further inspection at step 340:

[0066] a) One of the probe replicates of a probe set was eliminated by the Axiom GT1 algorithm due to unexpected intensity. For such probes, the genotyping algorithm uses only one of the replicates to genotype the sample. However, sometimes a single remaining intensity can cause the position of a homozygous sample to shift in cluster space, which is large enough and in the appropriate direction for the genotyping algorithm to call a Het.

[0067] b) The difference between the intensities measured on different probe replicates is higher than expected. In order to predict and classify probe intensities as unexpected, the SCI (Sequence Contrast Index) metric is defined.

[0068] SCI is calculated separately for the A and B alleles using I1 and I2 (normalized intensities for the two probe replicates) and can be easily generalized to more replicates:

[0069] SCI(A)=abs(I1(A)-I2(A))I1(A)+I2(A) (1);

[0070] SCI(B)=abs(I1(B)-I2(B))I1(B)+I2(B) (2)

[0071] For identical intensity, SCI is equal to 0, and as the difference between intensity increases, SCI approaches 1.Because two replicas of probe are unlikely to experience the unexpected intensity of the same type simultaneously, low SCI shows that the intensity measured on two probe replicas is reliable.If SCI is high for at least one of allele, then mark Het prediction is used for further inspection.The cutoff value for SCI is empirically derived, and has been shown to be applicable to different arrays and batch sizes.

[0072] Otherwise, or if the probe set has a single replicate and SCI cannot be calculated, the prediction undergoes further testing to determine whether the individual probe intensities are "Het-like". Not all Het predictions with unexpected potential intensities are false. In order to minimize the loss of true positive Hets, suspicious Het predictions can undergo further inspection. Ideally, the intensity of the driving Het prediction is compared with the expected intensity distribution for the Het genotype, as done for higher minor allele frequency (MAF) variants. However, this requires many examples of true Hets for a given variant, which are usually unavailable for low MAF variants.

[0073] Instead, the predicted Het intensity is compared with the intensity distribution for the major Hom. The major Hom genotype consists of two copies of the major allele and zero copies of the minor allele, while the Het genotype consists of one copy of each allele. These differences should be reflected in the intensity measurement of each allele in the two alleles, and can therefore be used to distinguish the two genotypes. The advantage of this method is that for low minor allele frequencies, the vast majority of genotypes are major Hom, which provides us with a large number of examples to characterize the major Hom intensity distribution for a given replica, as shown in step 350.

[0074] Although the intensity distribution is different for different probe sets, and sometimes different for different replicates, the range of positions relative to the mean and variance of this distribution suitable for Hets is generally common to all probe sets. The Het predicted potential intensities are therefore converted to Z scores at step 360.

[0075] In Figure 4, the analysis used in embodiments of the present invention is continued on an exemplary flow chart 400. The suspected probe intensities determined at step 410 can be compared relative to the appropriate major Hom intensity distribution and checked to determine whether these Z scores fall within an "acceptable" range as shown in step 430 (different ranges are appropriate depending on whether the intensity is measured for the major allele or the minor allele). In this way, it is possible to check whether the intensity measured on each probe-replicate is consistently Het-like.

[0076] If the prediction passes the second test, then in step 420, the Het genotype call is retained, otherwise the genotype of the probe set on the sample is set to "no call" (NC) at step 440. When the distribution of the replicate probe signals for the sample and probe set considered indicates that the call is a false positive, the Het is selectively adjusted to "no call". The number of het calls that are checked and set to "no call" depends on the array, more specifically on the number of rare variants on the array. In one embodiment of the invention, in a batch of 96 samples of genotyping on a high-density array, less than one hundred calls are usually adjusted. Typically, the rare het adjustment affects an average of less than one het per sample.

[0077] Figure 5 An exemplary illustration of the acceptable range of Het intensity Z scores for the intensity distribution of the main Hom sample used in the embodiments of the present invention is shown. Based on the summary statistics (mean and standard deviation) of the corresponding intensity distribution of the main Hom on a specific probe set, the Z score for the intensity of the predicted Het is calculated using the following formula:

[0078]

[0079] In one embodiment of the invention, only major Hom genotypes with reliable potential strength are used to calculate these summary statistics. In addition, Z scores are calculated only when the major allele can be determined with confidence (at least 90% of the samples are major Hom) and the number of high-quality major Hom genotypes is high enough (at least 30). Finally, if it is not possible to obtain such summary statistics for the major Hom, all Hets with potential abnormal strength are set to "no call".

[0080] Figure 5 An exemplary acceptance region for the Z-score for the major allele intensity is shown at 500. Since the major Hom has two copies of the major allele, while the Het has only one copy of the allele, a negative Z-score is expected. However, too low an intensity value and its associated Z-score may indicate that the probe is not working and is therefore not accepted. Figure 5 Graph 510 of shows an exemplary acceptance region for the Z score of the minor allele intensity. The major Hom has zero copies of the minor allele, and therefore their intensity on this allele is background noise, while the Het has one copy of this allele. Therefore, it is expected that the Z score for the minor allele intensity will be positive. On the other hand, an excessively high Z score may indicate that the intensity measured on the probe is abnormally high.

[0081] Embodiments of the present invention can be designed and trained on a variety of independent datasets. An important metric for rare het calls is the positive predictive value (PPV) of such calls. PPV is the fraction of correct het calls among all het calls, or TruePositive / (TruePositive+FalsePositive). For example, Figure 6 The number of samples in each data set is shown in chart 610, as well as the positive predictive value (PPV) for six representative data sets before and after applying RHA genotyping to clusters with a single het, all genotyping was performed with 1000 genome samples with public genotypes, where the application of RHA clearly increased the PPV to above 90% for all six representative data sets as shown by the blue squares and arrows of chart 610. Table 620 shows data for the six representative data sets, including initial PPV, PPV after RHA, and TP%, where TP% is the percentage of true Hets retained after RHA. TP% is an important metric, which is the number of true positive calls retained by the RHA algorithm. Figure 6The graph 610 in FIG. 610 further shows in green dots that on average more than 99% of true positive calls are retained after RHA application. Table 620 also shows other raw data, such as the initial number of true positives (TP# and TP# after RHA), and the initial number of false positive het calls (FP# and FP# after RHA).

[0082] Similar results were obtained when an embodiment of the present invention was applied to UK Biobank data. In one embodiment of the present invention, real data can be taken from whole exome sequencing (WES) data for 200K individuals from UK Biobank. UK Biobank data is a health resource that tracks the health and well-being of 500,000 human participants. The genetic data of the participants was genotyped on the Applied Biosystems Axiom UKBiobank and UKBileve microarrays from Thermo Fisher Scientific.

[0083] Figure 7 Illustrated is a chart showing the results of the RHA algorithm used in one embodiment of the present invention for UK Biobank data. Before describing the validation results for UK Biobank data, the minor allele frequency (MAF) bin is defined, as described in Weedon et al., (2019) "Assessing the analytical validity of SNP-chips for detecting very rare pathogenic variants: implications for direct-to-consumer genetic testing", https: / / www.biorxiv.org / content / 10.1101 / 696799v2.

[0084] According to Weeton et al., based on the Axiom array genotypes in the UK Biobank data, the UK Biobank probe sets can be restricted to Axiom calculated MAF (AcMAF) bins. The bins can be defined as shown in the table below. Note that bins 1-3 have extremely low allele frequencies. For example, bin 1 has an extremely low AcMAF, where a maximum of 9 individuals out of almost 500,000 individuals genotyped on the UK Biobank array have the minor allele. The variant in bin 3 is expected to have less than 5 to 10 individuals with het calls out of 50,000 individuals, so even bin 3 has an extremely low MAF. As Figure 7 As shown in Figure 700, RHA genotyping significantly improves PPV across all MAF ranges. PPV and sensitivity were calculated based on exome sequencing data for approximately 50,000 UK BioBank samples. Statistics were calculated for all heterozygous genotypes of probe sets from various calculated MAF groups and applied to two replicate probe sets with het cluster sizes up to 4 within the corresponding batch. Probe sets were filtered based on rare het algorithm predictions. Bins 4 and 5 have few to no rare variants and are therefore not affected by the RHA genotyping algorithm. For completeness, in Figure 7 Bin 4 is shown in the graph 700 of , while bin 5 having a MAF greater than 1% is not shown because all values ​​are effectively 100% before and after the rare het adjustment.

[0085] warehouse 1 2 3 4 5 AcMAF 0％-0.001％ 0.001％-0.005％ 0.005％-0.01％ 0.01％-1％ 1％-50％

[0086] Verification or ground truth data (e.g., known sequences to be studied) can be obtained by whole exome sequencing or whole genome sequencing. The results of the genotyping algorithm in one or more embodiments of the invention discussed herein can be compared with Exome sequencing data available for approximately 50,000 participants in the UK Biobank cohort.

[0087] The embodiments of the invention discussed herein utilize an improved algorithm for genotype calling. Experimental data demonstrate that the Axiom microarray achieves excellent PPV for very rare variants, thereby removing false het calls with high accuracy while keeping true calls virtually intact. For very rare variants (less than 0.01% MAF, e.g., less than 1 expected het in 5,000 individuals), it was observed that the Axiom array genotype gave excellent performance for these markers. oI2n , especially when evaluated with samples carrying rare HETs to confirm performance.

[0088] When variants were restricted to those probe sets that generated at least one true positive het in the Exome sequencing data, a PPV of 98% and an average sensitivity of 96% were achieved (data not shown). This excellent performance was observed on the most challenging bin, AcMAF bin 1.

[0089] In one embodiment of the invention, additional levels of quality control beyond RHA are needed to further reduce false Het calls for very rare variants, even if RHA has greatly reduced false Het calls. In such cases, supervised machine learning classification models, such as support vector machine prediction models, can be used. Support vector machine prediction models are described below with respect to embodiments of the invention, although other classification models known in the art can be envisioned within the scope of the invention, including but not limited to random forests or multi-layer perception (MLP) classifiers.

[0090] The support vector machine (SVM) predictor utilized in the embodiments of the invention described herein starts with the rare heterozygous adjustment (RHA) algorithm output. Thus, the SVM predictor provides additional quality control and predicts whether a putative Het that passes the RHA test (not set to NoCall) is an actual heterozygous (Het) genotype (true positive) or an actual homozygous (Hom) genotype (false positive). The support vector machine (SVM) predictor utilized in the embodiments of the invention described herein starts with the rare heterozygous adjustment (RHA) algorithm output.

[0091] Figure 8 800 is a flow chart illustrating a method 800 for genotyping rare variants including support vector machine analysis according to some embodiments of the present invention. In step 810, genotype calling is performed as described above, such as using the AxiomGT1 genotyping algorithm. In step 820, rare heterozygote adjustment (RHA) genotyping is performed on the 2-replica probe set as described above. In one embodiment of the present invention, if the rare Het genotype identified from the Axiom GT1 genotyping generated by the 1-replica probe set is evaluated, step 820 may not be performed.

[0092] In step 830, a support vector machine (SVM) prediction model can be trained and tested. For the SVM prediction model, in one embodiment of the present invention, real data can be taken from the second release of whole exome sequencing (WES) data of 200K individuals from the UK Biobank table. In the real data, 8988 are true positives (TP), where Axiom GT1 using RHA calls Het, and the exome true call is also Het. In the real data, 2764 are false positives (FP), where Axiom GT1 using the RHA algorithm calls Het, but the WES true call is homozygous. The real data is then divided as follows: 80% of the real data is used as training data for training SVM predictors, and 20% of the real data is used as test data. In one embodiment of the present invention, five predictors can be used. The five predictors are as follows, four numeric predictors and one Boolean predictor (e.g., yes / no): minMinorSdDist (should be positive), maxMajorSdDist (should be negative), maxSCI (lower is better), confidence score (lower is better), and AlleleSpecific (AS) (e.g., yes AS or no AS; where no AS is better). In step 840, the SVM prediction model is run to predict whether the Het tested by RHA is a heterozygous (true positive) genotype or a homozygous (false positive) genotype.

[0093] The five predictors are defined as follows:

[0094] MaxSCI: SCI (sequence contrast index) is calculated separately for the A and B alleles using I1 and I2, the normalized intensities of the two probe replicates for that allele: maxSCI = Max(SCI(A), SCI(B), where

[0095]

[0096] MaxSCI requires two or more replicates for a given probe, whereas other predictors can be applied to probes with a single replicate.

[0097] minMinorSdDist and maxMajorSdDist: Convert the array intensity of the putative Het potential to a Z score (MinorSdDist, MajorSdDist) relative to the appropriate major Hom intensity distribution. Specifically: for each replicate probe sequence, i = 1 to 2,

[0098] MinorSdDist_i = [IntensityOfThePutativeHet_i - Mean(HomIntensity_i)] / Std(HomIntensity_i), in the Axiom channel for the minor homozygous allele;

[0099] ο MajorSdDist_i = [IntensityOfThePutativeHet_i - Mean(HomIntensity_i)] / Std(HomIntensity_i), in the Axiom channel for the major homozygous allele;

[0100] οminMinorSdDist=min(MinorSdDist_i);

[0101] οmaxMajorSdDist=max(MajorSdDist_i);

[0102] Confidence score: by Axiom TM Confidence scores for Axiom genotype calls generated by the GT1 algorithm, or other suitable genotyping algorithm as described herein.

[0103] AlleleSpecific (AS) Y / N): An indicator of whether the probe set for rare variants was designed using an AlleleSpecific probe set or a Non-AlleleSpecific probe set.

[0104] In one embodiment of the present invention, the support vector machine prediction model is run using the R kernel. Support vector machine (or SVM) is a machine learning technique for classification tasks. In short, SVM works by identifying the best decision boundary that separates data points from different groups (or classes), and then predicts the class of new observations based on the separation boundary. Depending on the situation, different groups may be able to be separated by linear straight lines or by nonlinear boundary lines. The support vector machine method can handle both linear class boundaries and nonlinear class boundaries. It can be used for two-class classification problems and multi-class classification problems. In real life data, the separation boundary is usually nonlinear. Technically, the SVM algorithm uses kernel transformation to perform nonlinear classification. The most commonly used kernel transformations are polynomial kernels and radial kernels. Support vector machine (or SVM) is a useful classification technique. In order to set up a nonlinear SVM classifier, a polynomial kernel function or a radial kernel function, or other appropriate kernel transformation techniques can be used.

[0105] The package automatically selects optimal values ​​for the model tuning parameters, where optimal is defined as the value that maximizes the model accuracy. To build a nonlinear SVM classifier, either a polynomial kernel function or a radial kernel function can be used, as discussed above. Likewise, the caret package can be used to easily compute polynomial SVM nonlinear models and radial SVM nonlinear models.

[0106] In one embodiment of the invention, the caret R package is used to build an SVM classifier. In one embodiment of the invention, an R package can be loaded, such as tidyverse to help facilitate data manipulation and visualization, and caret to help facilitate machine learning workflow. In one embodiment of the invention, UK BioBank data are used as training data. Variables can be normalized so that their scales are comparable. This can be done automatically by setting option preProcess=c("center", "scale") before setting up the SVM classifier. The SVM radial basis function classifier can be fitted as follows. The package automatically selects the optimal value for the model tuning parameter, wherein the optimal is defined as the value that maximizes the model accuracy.

[0107] svm_Radial<-train(Concordance_with_200Kexome~.,data=train_ukbb_ss,method="svmRadial",

[0108] trControl=trctrl,

[0109] preProcess=c("center","scale"),

[0110] tuneLength=10)

[0111] trctrl<-trainControl(method="repeatedcv",number=10,repeats=3,savePredictions=TRUE)

[0112] The SVM prediction model can be viewed as follows:

[0113] Support Vector Machine with Radial Basis Function Kernel

[0114] 9401 samples

[0115] 5 predictors

[0116] 2 classes: "0", "1"

[0117] Preprocessing: Centered(5), Scaled(5)

[0118] Resampling: Cross-validation (10-fold, repeated 3 times)

[0119] Summary of sample sizes: 8461, 8461, 8461, 8461, 8461, 8461, …

[0120] Resampling results across tuning parameters:

[0121] C Accuracyκ 0.25 0.9286606 0.7928662 0.50 0.9288734 0.7940837 1.00 0.9294050 0.7961062 2.00 0.9293694 0.7962098 4.00 09292277 0.7959212 8.00 0.9277740 0.7918608 16.00 0.9268166 0.7893567 32.00 0.9262140 0.7877711 64.00 0.9251860 0.7849598 128.00 0.9239804 0.7818771

[0132] The tuning parameter "Σ" was kept constant at a value of 0.6126916.

[0133] Use accuracy to select the best model using the maximum value.

[0134] The final values ​​for the model were Σ=0.6126916 and C=1.

[0135] Fig. 9Table 900 of test results for UK Biobank data according to one embodiment of the present invention is shown. In one embodiment of the present invention, real data is taken from the second release of whole exome sequencing (WES) data for 200,000 individuals from UK Biobank. Test data is performed on the RHA post-probe set. In the WES data, 8988 are true positives (TP), where the Axiom GT1 algorithm calls Het, and the exome true call is Het. In the WES data, 2764 are false positives (FP), where the Axiom GT1 algorithm calls het, where the exome true data is homozygous. As discussed above, the real data is then divided as follows: 80% of the real data is used as training data for training SVM predictors, and 20% of the real data is used as test data. Therefore, the WES real data used as test data has 1798 heterozygous calls (true positives) and 553 homozygous calls (false positives) for 2351 individuals. Overall, the SVM predictor model increased the positive predictive value (PPV) by 18% (from 76.4% following Axiom GT1 and RHA to 94.33% using the SVM predictor model).

[0136] Therefore, in Fig. 10A In some embodiments of the invention shown, applying the 5-predictor SVM predictor model as described above to the rare post-RHA Het genotypes generated by the "2-replicate" probe set increased the post-RHA PPV by 18% (76.4% to 94.33%) while decreasing the true Het retention percentage by only 2.5% (100% to 97.55%). In one embodiment of the invention, two predictors (ConfidenceScore and AlleleSpecific) can be decreased compared to the original 5-predictor model without substantially affecting the PPV and TrueHetsRetained%. Therefore, in one embodiment of the invention, only three predictors are required to run the SVM prediction model: maxSci, MaxMajorSdDist, and MinMinorSdDist. As shown in Table 1000A and Chart 1010A, decreasing any of the three predictors, maxSci, MaxMajorSdDist, and MinMinorSdDist, decreases the PPV and sensitivity.

[0137] In one embodiment of the invention, as discussed above with respect to the UK BioBank data, the input to the support vector machine prediction model includes data points corresponding to at least 9400 training set heterozygous genotype calls. Of course, in other embodiments of the invention, data sources with training sets of heterozygous genotype calls of varying sizes may be used as input to the support vector machine prediction model. Fig. 10B As shown, embodiments of the present invention can utilize an input size for a support vector machine prediction model that includes a set of data points with a minimum size of 940 training set heterozygous genotype calls to obtain optimal results. For example, Fig. 10B Graph 1000B (at data points 1001B and 1002B) and table 1010B show that a minimum training set size of 940 is required for high PPV (>94.0%) and high sensitivity (>97.1%).

[0138] Other embodiments of creating an SVM using training data and running the SVM using test data are shown below:

[0139] #Create SVM using training data

[0140] library(caret)

[0141] trctrl<-trainControl(method="repeatedcv",number=10,repeats=3,savePredictions=TRUE,

[0142] classProbs=TRUE,summaryFunction=twoClassSummary)

[0143] set.seed(3233)

[0144] svm_Radial<-train(Concordance_with_200Kexome~.,data=train_ukbb_ss,method="svmRadial",

[0145] trControl=trctrl,

[0146] preProcess=c("center","scale"),

[0147] tuneLength=10,

[0148] metric="ROC")

[0149] #Run test data

[0150] test_ukbb<-test_ukbb_ss[,c(-1)]

[0151] test_pred<-predict(svm_Radial,newdata=test_ukbb,type="prob")

[0152] Fig.11 The test result table of 2-replica probe set is illustrated, and the test result table shows various probability cutoff values ​​(Het_ProbabilityCUT), and above these probability cutoff values, the SVM genotype is called heterozygous (Het), otherwise it is called homozygous (Hom). TP (true positive) is that the SVM genotype is predicted to be Het, and the WES (real data) genotype is Het. FP (false positive) is that the SVM genotype is predicted to be Het, and the WES genotype is Hom. FN (false negative) is that the SVM genotype is Hom, and the WES genotype is Het. TN (true negative) is that the SVM genotype is Hom, and the WES genotype is Hom." red arrow " 1110 is the PPV of Het through RHA (for example, Het that is not set to NoCall), and wherein Het_ProbabilityCUT is zero, meaning that the SVM prediction model will always call Het, so the PPV is the same as the PPV of all Het through RHA. "Yellow arrow" 1120 is the probability cutoff selected by the SVM for binary prediction, where the SVM prediction model has an equal chance of predicting Het or Hom. In this case, the true Het call retention percentage (HetsRetained% = TP / (TP+FN)) is still as high as 97.72%, while the positive prediction value (HetPPV = TP / (TP+FP)) has risen from 76.48% to 94.01%.

[0153] Fig.12 1200 in Figure 1200 shows that PPV can be increased with almost no loss of true Het retention percentage, so that applying the SVM prediction model to the genotyping results of the microarray will increase the accuracy of rare heterozygous genotype calls. The trade-off between PPV and true Het retention percentage can be tuned using the Het probability value as described above. Graph 1210A shows that PPV increases with Het probability cutoff value, and graph 1210B shows that true Het retention percentage decreases with Het probability cutoff value. This is because as shown in graph 1220A, the probability of Het is high for true heterozygous calls, and as shown in graph 1220B, the probability of Het is low for true homozygous calls.

[0154] In some embodiments of the invention, support vector machine prediction models can be directly applied to genotyping data (eg, genotyping data from Axiom GT1) that have not been individually adjusted for rare heterozygous calls. Fig.13 An exemplary flow chart 1300 of such a process used in an embodiment of the present invention is shown. In one embodiment of the present invention, the SVM prediction model can be applied to rare Hets produced by 1-replica probe sets. For example, for UK Biobank microarray genotyping data, it is inferred that Het genotypes can be obtained from UK Biobank Het calls from Het genotype clusters with <=4 Het calls, and produced by a tiling probe set (called "1 replica" probe set) with only one feature. It should be noted that in certain versions of the Axiom GT1 genotyping process performed as step 1310, these Het calls are all completed without applying the Axiom GT1 RHA algorithm version, because in some embodiments of the present invention, Axiom GT1 RHA may not be applied to 1-replica probe sets. In one embodiment of the present invention, a separate SVM model can also be developed, which can be trained on a single replica.

[0155] In one embodiment of the invention, the real data is taken from the whole exome sequencing (WES) data (second release) for approximately 200,000 individuals from UK Biobank. In one embodiment of the invention, 10% of the 213,596 het calls from the WES real data as described above are randomly sampled for training and testing the SVM prediction model in step 1320, although more or less data can be sampled according to the needs of computational performance. As in the previous SVM prediction model described herein, the WES real data can be divided into 80% training data and 20% test data. In one embodiment of the invention, the counts in the test set are: 3893 true positives (TP), 379 false positives (FP). The counts in the training set are: 15490 TP and 1597 FP. In one embodiment of the invention, three predictors may be applied: (1) MinorSdDist, where the formula is as defined above, e.g., [IntensityOfthePutativeHet-Mean(HomIntensity)] / Std(HomIntensity); (2) MajorSdDist, where the formula is as defined above, e.g., [(IntensityOfthePutativeHet-Mean(HomIntensity)] / Std(HomIntensity); and (3) nAB_inBatch, which is the number of hets in a Het genotype cluster (the larger the better). Once the SVM prediction model has been trained and tested, it may be operated on experimental data at step 1330 to predict whether a rare heterozygous call generated by a genotyping assay is a true heterozygous (true positive) genotype or a homozygous (false positive) genotype.

[0156] Fig.14Table 1400 in shows the SVM test set metric for 1-replica probe set, wherein different Het probability value cutoffs (Het_PVALUECUT) are listed for the UK Biobank data used in one embodiment of the present invention. In one embodiment of the present invention, for each Het call generated by a genotyping algorithm (e.g., Axiom GT1), the SVM prediction model generates the probability that the genotype is Het, without a separate rare heterozygous adjustment step as described above. The rule for Het_PVALUECUT is that if the Het probability is greater than the Het_PVALUECUT value, the SVM call is Het, otherwise the SVM call is homozygous. Chart 1410 shows a graph of PPV to true Het retention percentage for different probability cutoffs as shown in Table 1400. Note that the PPV for 1-replica probe set Het starts with a much higher value (91.4%) than the 2-replica probe set Het, and the PPV for the 2-replica probe set is about 76%.

[0157] These test results occur because in the Axiom array design used in the embodiments of the present invention discussed herein, only probe sets with high cluster resolution (e.g., 2-replica FLD>6) are selected to tile the 1-replica probe set, and FLD represents the Fisher linear discriminant (FLD) = Min(i = aa, bb){|Mab-Mi| / sd}.

[0158] where Maa, Mbb, and Mab are the centers of the homozygous (aa, bb) and heterozygous (ab) clusters in the contrast dimension (X-axis), and sd is the square root of the pooled variance across the three distributions.

[0159] Graph 1410 shows that the 1-replica SVM prediction model experimental results increased the PPV by 3%, from 91.4% produced by the standard Axiom GT1 algorithm to 94.4%, while only reducing the true Het retention percentage by 1%.

[0160] In one embodiment of the invention, the RHA algorithm output can determine whether a het call is "defective", and the "defective" value is stored in an output file. "Defects" refer to physical defects that can be caused by spots and scratches on the microarray surface, which can cause differences in the intensity values ​​measured on the microarray surface. Most of the het calls determined by RHA are flawless. In some embodiments of the invention, the SVM prediction value may not be so accurate when determining whether a het call is a true het for a flawed het call. Therefore, in one embodiment of the invention, a flawed het call is set to a probability of zero, so this currently has a negligible effect on the true het retention percentage value. In one embodiment of the invention, it is possible to trigger a setting or switch in the user interface, such as how to handle a flawed het call. In some embodiments of the invention, a method similar to the above method can be used to develop an SVM prediction model for flawed hets.

[0161] The systems, devices, and methods described herein can be implemented using digital circuit systems or using one or more computers utilizing well-known computer processors, memory units, storage devices, computer software, and other components. Typically, a computer includes a processor for executing instructions and one or more memories for storing instructions and data. A computer may also include or be coupled to one or more mass storage devices, such as one or more magnetic disks, internal hard disks and removable disks, magneto-optical disks, optical disks, and the like.

[0162] The systems, devices, and methods described herein can be implemented using computers operating in a client-server relationship. Typically, in such a system, the client computer is remote from the server computer and interacts via a network. The client-server relationship can be defined and controlled by computer programs running on the respective client and server computers. Examples of client computers may include desktop computers, workstations, portable computers, cellular smart phones, tablets, or other types of computing devices.

[0163] The systems, devices, and methods described herein may be implemented using a computer program product tangibly embodied in an information carrier, such as in a non-transitory machine-readable storage device, for execution by a programmable processor; and comprising Figure 1 Figure 3, Figure 4, Figure 8 and Fig.13The method processes and steps described herein of one or more of the steps of the steps can be implemented using one or more computer programs that can be executed by such a processor. A computer program is a set of computer program instructions that can be used directly or indirectly in a computer to perform a certain activity or produce a certain result. A computer program may be written in any form of programming language, including a compiled language or an interpreted language, and the computer program may be deployed in any form, including as a separate program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0164] Fig.15 A high-level block diagram of an exemplary device that can be used to implement the systems, devices, and methods described herein is shown. Device 1500 includes a processor 1510 that is operatively coupled to a persistent storage device 1520 and a main memory device 1530. Processor 1510 controls the overall operation of device 1500 by executing computer program instructions that define such operations. The computer program instructions may be stored in persistent storage device 1520 or other computer-readable medium and loaded into main memory device 1530 when execution of the computer program instructions is desired. For example, processor 1510 may include one or more components of PCR device 1540. Thus, Figure 1 Figure 3, Figure 4, Figure 8 and Fig.13 The method steps may be defined by computer program instructions stored in the main memory device 1530 and / or the persistent storage device 1520 and controlled by the processor 1510 executing the computer program instructions. For example, the computer program instructions may be implemented as a program that can be programmed by a person skilled in the art to execute the method. Figure 1 Figure 3, Figure 4, Figure 8 and Fig.13 Thus, by executing the computer program instructions, the processor 1510 performs the Figure 1 Figure 3, Figure 4, Figure 8 and Fig.13 The method steps are defined in the algorithm. The device 1500 also includes one or more network interfaces 1580 for communicating with other devices via a network. The device 1500 may also include one or more input / output devices 1590 that enable a user to interact with the device 1500 (e.g., a display, keyboard, mouse, speaker, buttons, etc.).

[0165] Processor 1510 may include general-purpose microprocessors and special-purpose microprocessors, and may be the sole processor of device 1500 or one of multiple processors. Processor 1510 may include one or more central processing units (CPUs) and one or more graphics processing units (GPUs), which may, for example, work separately from one or more CPUs and / or perform multitasking with one or more CPUs to accelerate processing (e.g., for various image processing applications described herein). Processor 1510, persistent storage device 1520, and / or main memory device 1530 may include one or more application-specific integrated circuits (ASICs) and / or one or more field programmable gate arrays (FPGAs), be supplemented with one or more application-specific integrated circuits and / or one or more field programmable gate arrays, or be incorporated into one or more application-specific integrated circuits and / or one or more field programmable gate arrays.

[0166] The persistent storage device 1520 and the main memory device 1530 each include a tangible non-transitory computer-readable storage medium. The persistent storage device 1520 and the main memory device 1530 may each include a high-speed random access memory, such as a dynamic random access memory (DRAM), a static random access memory (SRAM), a double data rate synchronous dynamic random access memory (DDR RAM), or other random access solid-state memory devices, and may include non-volatile memory, such as one or more magnetic disk storage devices such as built-in hard disks and removable disks, magneto-optical disk storage devices, optical disk storage devices, flash memory devices, semiconductor memory devices, such as erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM), digital versatile disk read-only memory (DVD-ROM) disks, or other non-volatile solid-state storage devices.

[0167] Input / output devices 1590 may include peripheral devices such as printers, scanners, display screens, etc. For example, input / output devices 1590 may include a display device (such as a cathode ray tube (CRT), plasma or liquid crystal display (LCD) monitor) for displaying information to a user, a keyboard, and a pointing device (such as a mouse or trackball through which a user can provide input to device 1500).

[0168] Any or all of the functionality of the systems and devices discussed herein may be performed by processor 1510 and / or incorporated into a device, such as PCR device 1540. Additionally, PCR device 1540 and / or device 1700 may utilize one or more neural networks or other machine learning or deep learning techniques performed by processor 1710 or other systems or devices discussed herein.

[0169] Those skilled in the art will recognize that a specific implementation of an actual computer or computer system may have other structures and may contain other components, and Fig.15 is a high-level representation of some of the components of such a computer for purposes of illustration.

[0170] The foregoing description should be understood to be illustrative and exemplary in every respect, rather than restrictive, and the scope of the invention disclosed herein is not determined by the description, but by the claims interpreted according to the full scope allowed by the patent law. It should be understood that the embodiments shown and described herein are merely illustrative of the principles of the present invention, and that various modifications may be implemented by those skilled in the art without departing from the scope and spirit of the present invention. Various other feature combinations may be implemented by those skilled in the art without departing from the scope and spirit of the present invention.

[0171] the term

[0172] As understood by those of ordinary skill in the art, the terms used herein with reference to the embodiments of the invention disclosed in this document should be given their ordinary meanings unless explicitly indicated otherwise or the context indicates otherwise.

[0173] Allele: Each of two or more alternative forms of DNA found at the same location on a chromosome

[0174] Allele A and Allele B: For a SNP or Indel, the two alternatives that can be observed and measured in a given sample are designated as "Allele A" and "Allele B"

[0175] Array: A DNA microarray used to genotype known genetic variants (SNPs and indels) in a population

[0176] Cluster space: X and Y dimensions defined by signal contrast and signal size

[0177] Exome: The approximately 1% to 2% of the human genome that codes for proteins

[0178] Genotyping: A method used to determine the bases (A, G, T, or C) or indels present at specific locations in a person's DNA.

[0179] Het call: abbreviation for heterozygous call

[0180] Hom call: abbreviation for homozygous call

[0181] Heterozygous: Two different alleles at a given locus in an individual

[0182] Homozygous: Two identical alleles at a given locus in an individual

[0183] Indel: A type of variant that contains one or more bases inserted or deleted compared to a reference genome

[0184] Major homozygotes: Two identical alleles at a given locus, representing the most common allele at a given locus in the population of interest

[0185] Minor homozygotes: two identical alleles at a given locus, representing the less common allele at the given locus in the population of interest

[0186] Average positive predictive value for a set of variants and individuals: the average of the positive predictive values ​​for each variant.

[0187] nAB: the number of samples called heterozygous by AxiomGT1 in cluster space.

[0188] Negative predictive value: the proportion of normal alleles confirmed by the reference standard (true negative / (true negative + false negative))

[0189] Positive predictive value: the proportion of variant alleles confirmed by the reference standard (true positives / (true positives + false positives))

[0190] Total positive predictive value: The positive predictive value of all variants and samples

[0191] Probe Set: A specific group (eg, a set of three or more) of DNA sequences on a microarray that detects the presence of two or more alleles at a given locus (thus interrogating specific known polymorphic or non-polymorphic positions in the genome).

[0192] Sensitivity: The proportion of variant alleles detected by the reference standard that are also found by the index test (true positives / (true positives + false negatives))

[0193] Signal contrast and signal magnitude: AxiomGT1 genotype clustering is performed in two dimensions. The X dimension is called "contrast" and the Y dimension is called "magnitude". They are the log-linear combination of the signal intensities of the two alleles. For alleles A and B, the contrast is log2(A / B) and the magnitude is (log2(A)+log2(B)) / 2

[0194] Single Nucleotide Polymorphism (SNP): A type of single nucleotide variant; a location in the genome where an individual differs from the reference human genome by a single base change (ie, a substitution of a single letter of DNA). SNPs can be rare or common in a population.

[0195] Specificity: The proportion of normal alleles detected by the reference standard that are also found to be normal by the index test (true negatives / (true negatives + false positives)).

[0196] Variant: A locus in the genome for which different alleles have been observed in different people.

Claims

1. A method for genotyping one or more rare variants of a gene in a plurality of nucleic acid samples using assay data generated from one or more assays performed on the plurality of nucleic acid samples on at least one microarray comprising a plurality of probe sets, the method comprising: evaluating, using one or more computer processors, one or more heterozygous genotype calls generated from the assay data to identify any rare heterozygous genotype calls, wherein the number of heterozygous genotype calls for the probe set does not exceed a maximum threshold; for each identified rare heterozygous genotype call, determining, using one or more computer processors, whether the identified rare heterozygous genotype call is a suspect call, and setting some of the suspect calls to a no call value; as well as For each rare heterozygous genotype call among the identified rare heterozygous genotype calls that is not set to a no call value, evaluating the identified rare heterozygous genotype calls to determine whether the identified rare heterozygous genotype calls include true rare heterozygous genotype calls, wherein the evaluating includes using a supervised machine learning classification model executed on one or more computer processors, the supervised machine learning classification model including a plurality of prediction values ​​for identifying true rare heterozygous genotype calls, wherein the supervised machine learning classification model is trained using input to the one or more computer processors executing the supervised machine learning classification model, the input including a plurality of data points, wherein each data point corresponds to a signal in a discrete set of signals for each training set nucleic acid sample in a plurality of training set nucleic acid samples.

2. The method of claim 1, wherein the supervised machine learning classification model comprises a support vector machine prediction model.

3. The method of claim 1, wherein the maximum threshold for the number of heterozygous genotype calls for the probe set is three.

4. The method of claim 1, wherein the maximum threshold for the number of heterozygous genotype calls for the probe set is four.

5. The method of any one of claims 1 to 4, wherein for each identified rare heterozygous genotype call, determining whether the identified rare heterozygous genotype call is a suspect call and is set to a no call value, the method further comprising: Testing for any unexpected strength of identified rare heterozygous genotype calls; If there were no unexpected intensities, the identified rare heterozygous call was retained; as well as If an unexpected intensity is present, further testing of one or more individual probe intensities is performed to determine whether the rare heterozygous genotype call should be set to a no call value.

6. The method of claim 5, wherein testing any unexpected strength of the identified rare heterozygous genotype calls further comprises: The sequence contrast index (SCI) metric was calculated for each allele for each rare heterozygous genotype call identified; as well as Determine whether the identified rare heterozygous genotype call is a suspected false positive heterozygous genotype call where the SCI metric exceeds a minimum threshold.

7. The method of claim 6, wherein the SCI metric for each of the two alleles A and B of the two probe replicates of the probe set comprises: For the A allele, SCI(A)=(abs(I1(A)-I2(A))) / (I1(A)+I2(A)); and For the B allele, SCI(B)=(abs(I1(B)-I2(B))) / (I1(B)+I2(B)), where I1 and I2 comprise the normalized probe intensities for the two probe replicates. 8 . The method of claim 7 , wherein a first predicted value among the plurality of predicted values ​​comprises a maximum SCI value, the maximum SCI value being a maximum value among an SCI(A) value and an SCI(B) value.

9. The method of any one of claims 1 to 8, wherein a second prediction value in the plurality of prediction values ​​comprises a minimum minor standard deviation distance value, the minimum minor standard deviation distance value being the minimum of the minor standard deviation distance values ​​for all replicate probe sequences i across the probe set, wherein the minor standard deviation distance value (MinorSdDist_i) for replicate probe sequence i comprises: MinorSdDist_i=[IntensityOfthePutativeHet_i- Mean(HomIntensity_i)] / Std(HomIntensity_i), where IntensityOfthePutativeHet_i includes the normalized intensity value of the minor homozygous allele of the identified rare heterozygous genotype call at replicate probe sequence i, HomIntensity_i includes the normalized intensity value of the minor homozygous allele of the homozygous genotype call at replicate probe sequence i, Mean(HomIntensity_i) includes the mean of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set, and Std(HomIntensity_i) includes the standard deviation of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set.

10. The method according to any one of claims 1 to 9, wherein a third prediction value in the plurality of prediction values ​​comprises a maximum major standard deviation distance value, the maximum major standard deviation distance value being the maximum value among the major standard deviation distance values ​​of all replicate probe sequences i across the probe set, wherein the major standard deviation distance value (MajorSdDist_i) for replicate probe sequence i comprises: MajorSdDist_i=[IntensityOfthePutativeHet_i- Mean(HomIntensity_i)] / Std(HomIntensity_i), where IntensityOfthePutativeHet_i includes the normalized intensity value of the major homozygous allele of the identified rare heterozygous genotype call at replicate probe sequence i, HomIntensity_i includes the normalized intensity value of the major homozygous allele of the homozygous genotype call at replicate probe sequence i, Mean(HomIntensity_i) includes the mean of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set, and Std(HomIntensity_i) includes the standard deviation of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set.

11. A method according to any one of the preceding claims, wherein the plurality of predicted values ​​comprises at most three predicted values.

12. A method according to any one of claims 1 to 11, wherein the input to the supervised machine learning classification model includes data points corresponding to at least 9400 training set heterozygous genotype calls.

13. A method according to any one of claims 2 to 11, wherein the input to the supervised machine learning classification model comprises a set of data points with a minimum size of 940 training set heterozygous genotype calls.

14. The method according to any one of claims 1 to 13, further comprising: One or more assays of the plurality of nucleic acid samples are run using at least one microarray comprising a plurality of probe sets.

15. The method according to any one of claims 1 to 14, further comprising: Assay data in a computer readable form is generated from the results of one or more assays.

16. The method according to any one of claims 1 to 15, further comprising: Based on the assay data, one or more heterozygous genotypes in the plurality of nucleic acid samples are called using one or more computer processors.

17. A non-transitory computer readable medium storing instructions that, when executed by one or more computer processors, perform the process of any one of claims 1 to 16.

18. A computer system comprising one or more processors configured to perform the process according to any one of claims 1 to 16.

19. A computer-related product comprising a non-transitory computer-readable medium storing one or more instructions that, when executed by one or more processors, perform a method for genotyping one or more rare variants of a gene in a plurality of nucleic acid samples using data generated from one or more assays performed on the plurality of nucleic acid samples on at least one microarray comprising a plurality of probe sets, the method comprising: evaluating the heterozygous genotype calls using one or more computer processors to identify any rare heterozygous genotype calls from a probe set, wherein the number of heterozygous genotype calls for the probe set does not exceed a maximum threshold; for each identified rare heterozygous genotype call, determining, using one or more computer processors, whether the identified rare heterozygous genotype call is a suspect call, and setting some of the suspect calls to a no call value; as well as For each rare heterozygous genotype call among the identified rare heterozygous genotype calls that is not set to a no call value, evaluating the identified rare heterozygous genotype calls to determine whether the identified rare heterozygous genotype calls include true rare heterozygous genotype calls, wherein the evaluating includes using a supervised machine learning classification model executed on one or more computer processors, the supervised machine learning classification model including a plurality of prediction values ​​for identifying true rare heterozygous genotype calls, wherein the supervised machine learning classification model is trained using input to the one or more computer processors executing the supervised machine learning classification model, the input including a plurality of data points, wherein each data point corresponds to a signal in a discrete set of signals for each training set nucleic acid sample in a plurality of training set nucleic acid samples.

20. The computer-related product of claim 19, wherein the supervised machine learning classification model comprises a support vector machine prediction model.

21. The computer-related product of claim 19, wherein the maximum threshold for the number of heterozygous genotype calls for the probe set is three.

22. The computer-related product of claim 19, wherein the maximum threshold for the number of heterozygous genotype calls for the probe set is four.

23. The computer-related product of any one of claims 19 to 22, wherein for each identified rare heterozygous genotype call, determining whether the identified rare heterozygous genotype call is a suspect call and is set to a no call value, the method further comprising: Testing for any unexpected strength of identified rare heterozygous genotype calls; If there were no unexpected intensities, the identified rare heterozygous call was retained; as well as If an unexpected intensity is present, further testing of one or more individual probe intensities is performed to determine whether the rare heterozygous genotype call should be set to a no call value.

24. The computer-related product of claim 23, wherein testing any unexpected strengths of the identified rare heterozygous genotype calls further comprises: The sequence contrast index (SCI) metric was calculated for each allele for each rare heterozygous genotype call identified; as well as A determination is made as to whether the identified rare heterozygous genotype call is a suspected false positive heterozygous genotype call, wherein the SCI metric exceeds a minimum threshold.

25. The computer-related product of claim 24, wherein the SCI metric for each of the two alleles A and B of the two probe replicates of the probe set comprises: For the A allele, SCI(A)=(abs(I1(A)-I2(A))) / (I1(A)+I2(A)); and For the B allele, SCI(B)=(abs(I1(B)-I2(B))) / (I1(B)+I2(B)), where I1 and I2 comprise the normalized probe intensities for the two probe replicates.

26. The computer-related product of claim 25, wherein a first predicted value of the plurality of predicted values ​​comprises a maximum SCI value, the maximum SCI value being a maximum value of the SCI(A) value and the SCI(B) value.

27. A computer-related product according to any one of claims 19 to 26, wherein a second prediction value in the plurality of prediction values ​​comprises a minimum minor standard deviation distance value, the minimum minor standard deviation distance value being the minimum of the minor standard deviation distance values ​​for all replicate probe sequences i across the probe set, wherein the minor standard deviation distance value (MinorSdDist_i) for replicate probe sequence i comprises: MinorSdDist_i=[IntensityOfthePutativeHet_i- Mean(HomIntensity_i)] / Std(HomIntensity_i), where IntensityOfthePutativeHet_i includes the normalized intensity value of the minor homozygous allele of the identified rare heterozygous genotype call at replicate probe sequence i, HomIntensity_i includes the normalized intensity value of the minor homozygous allele of the homozygous genotype call at replicate probe sequence i, Mean(HomIntensity_i) includes the mean of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set, and Std(HomIntensity_i) includes the standard deviation of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set.

28. A computer-related product according to any one of claims 19 to 27, wherein a third prediction value in the plurality of prediction values ​​comprises a maximum major standard deviation distance value, the maximum major standard deviation distance value being the maximum value of the major standard deviation distance values ​​for all replicate probe sequences i across the probe set, wherein the major standard deviation distance value (MajorSdDist_i) for replicate probe sequence i comprises: MajorSdDist_i=[IntensityOfthePutativeHet_i- Mean(HomIntensity_i)] / Std(HomIntensity_i), where IntensityOfthePutativeHet_i includes the normalized intensity value of the major homozygous allele of the identified rare heterozygous genotype call at replicate probe sequence i, HomIntensity_i includes the normalized intensity value of the major homozygous allele of the homozygous genotype call at replicate probe sequence i, Mean(HomIntensity_i) includes the mean of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set, and Std(HomIntensity_i) includes the standard deviation of the HomIntensity_i values ​​over the homozygous genotype calls of the probe set.

29. A computer-related product according to any one of claims 19 to 28, wherein the plurality of predicted values ​​comprises at most three predicted values.

30. A computer-related product according to any one of claims 19 to 29, wherein the input to the supervised machine learning classification model includes data points corresponding to at least 9400 training set heterozygous genotype calls.

31. A computer-related product according to any one of claims 19 to 30, wherein the input to the supervised machine learning classification model comprises a set of data points with a minimum size of 940 training set heterozygous genotype calls.

32. The computer-related product according to any one of claims 19 to 31, further comprising: One or more assays of the plurality of nucleic acid samples are run using at least one microarray comprising a plurality of probe sets.

33. A computer-related product according to any one of claims 19 to 32, further comprising: Assay data in a computer readable form is generated from the results of the one or more assays.

34. A computer-related product according to any one of claims 19 to 33, further comprising: Based on the assay data, one or more heterozygous genotypes in the plurality of nucleic acid samples are called using one or more computer processors.

Images