System and method for identifying mutation and phenotype association

Abstract

Aspects of the inventive concepts generally relate to systems and methods of mutation processing, and more specifically, to systems and methods of identifying associations between phenotypes and mutations. An exemplary method generally includes receiving one or more input features including phenotype data and mutation data, generating a candidate seeker (CE) score indicating a probability of association between a phenotype and a mutation through a machine learning model based on the one or more input features, and outputting an indication of an association between the phenotype and the mutation based on the CE score.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 357,803, filed on July 1, 2022, entitled “SYSTEMS AND METHODS FOR IDENTIFYING ASSOCIATIONS BETWEEN MUTATIONS AND PHENOTYPES,” the entire contents of which are incorporated herein by reference.

[0003] Government-backed recognition

[0004] This invention was made with government support from the National Institutes of Health under Grants AI125581 and AI100627. The government has certain rights in this invention. Background Art 1. Technical Field Various aspects of the present inventive concepts relate generally to systems and methods for mutation management, and more particularly to the identification of associations between phenotypes and mutations.

[0006] 2. Discussion of related technologies Phenotype refers to a set of observable characteristics that result from the interaction between genotype and environment. In some cases, a genetic mutation may be the causative factor of a phenotype. Mutations generally refer to changes in the sequence of deoxyribonucleic acid (DNA). Mutations can occur as a result of DNA replication during cell division, ionizing radiation, mutagens, or viral infection. Summary of the invention

[0007] Certain aspects of the disclosed technology may provide a mutation processing method. The method generally includes receiving one or more input features including phenotype data and mutation data, generating a candidate explorer (CE) score indicating a probability of an association between a phenotype and a mutation based on the one or more input features through a machine learning model, and outputting an indication of an association between the phenotype and the mutation according to the CE score.

[0008] In some embodiments, the associated indication includes a candidate state of the association based on the CE score and an algorithm score indicating the likelihood that the mutation is causal. The mutation data may include a damage score indicating the likelihood that the protein function associated with the mutation is impaired. In addition, the method may also include generating a damage score by using another machine learning model trained with known harmful mutations and neutral mutations. In addition, the one or more input features may also include a necessity score indicating the likelihood that homozygous mice of a robust knockout allele of a gene associated with the mutation will die before weaning age. Additionally or alternatively, the method may also include generating a necessity score by using another machine learning model trained with genes known to be non-essential for survival and genes known to be essential for survival. The one or more input features may also include features associated with an algorithm score indicating the likelihood of a mutation causing the cause.

[0009] In addition, in some cases, the one or more input features may also include linkage data generated using automated meiotic mapping (AMM). The method may also include, when the two or more mutations co-segregate, determining which of the two or more mutations is a more robust causal candidate for the phenotype by omitting shared zygosity instances of the two or more mutations, wherein the CE score is generated based on the determination. In addition, the one or more input features also include at least one of the following: the number of phenotypes whose algorithm scores for the mutation meet the threshold, the algorithm score indicating the likelihood that the mutation is causative; the average number of AMM operations that result in a p-value that meets the threshold for each allele of the gene associated with the mutation; the algorithm score of the mutation or phenotype; the number of AMM operations that result in a p-value that meets the threshold for the gene associated with the mutation; the damage score of the mutation, the damage score indicating the likelihood that the protein function associated with the mutation is impaired; the number of pedigrees in a superpedigree associated with the gene, and whether the p-value resulting from the AMM operation of the superpedigree meets the threshold; the number of phenotypes whose p-values ​​of the superpedigree meet the threshold; the number of pedigrees that contribute to the p-value of the superpedigree that meets the threshold; the number of pedigrees in the superpedigree amount; percentage of fluorescence-activated cell sorting (FACS) screens with p-values ​​meeting the mutation threshold; minimum p-value from AMM manipulations; percentage of variant allele (VAR) mice whose screen results overlap with B6 mice; whether superpedigree AMM manipulation results meet the thresholds for null and missense alleles; whether superpedigree AMM manipulation results meet the threshold for null alleles; percentage of VAR mice whose screen results overlap with those of reference allele (REF) mice; difference between AMM manipulation results from heterozygous (HET) mice and VAR mice; number of female REF mice used for AMM manipulations; percentage of body weight screens with p-values ​​meeting the mutation threshold; number of female HET mice used for AMM manipulations; and / or difference between AMM manipulation results from REF mice and VAR mice.

[0010] In addition, certain aspects of the disclosed technology may provide a mutation processing device. The device may generally include: a memory; and one or more processors coupled to the memory, which are configured to: receive one or more input features including phenotype data and mutation data, generate a CE score indicating the probability of association between the phenotype and the mutation through a machine learning model based on the one or more input features, and output an indication of the association between the phenotype and the mutation based on the CE score.

[0011] In some embodiments, the association indication includes an algorithmic score indicating the likelihood that the association candidate state and the mutation are causative based on the CE score. The mutation data may also include a damage score indicating the likelihood of impaired protein function associated with the mutation. In addition, one or more processors may also be configured to generate a damage score by using another machine learning model trained with known deleterious mutations and neutral mutations. In addition, one or more input features may also include a necessity score indicating the likelihood that homozygous mice of a robust knockout allele of a gene associated with the mutation are lethal before weaning age.

[0012] In some cases, the one or more processors may be further configured to generate an essentiality score by using another machine learning model trained with genes known to be non-essential for survival and genes known to be essential for survival. The one or more input features may also include a feature associated with an algorithm score indicating the likelihood that the mutation is causal. In addition, the one or more input features may also include linkage data generated using automatic meiotic mapping (AMM). In addition, when two or more mutations are co-segregated, the one or more processors may be further configured to determine which of the two or more mutations is a more robust causal candidate for the phenotype by omitting shared zygosity instances of the two or more mutations, wherein the one or more processors may be configured to generate a CE score based on the determination.

[0013] Certain aspects of the disclosed technology provide a non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to: receive one or more input features including phenotypic data and mutation data, generate a CE score indicating a probability of an association between a phenotype and a mutation through a machine learning model based on the one or more input features, and output an indication of an association between the phenotype and the mutation based on the CE score.

[0014] Other embodiments are also described and described herein. In addition, although multiple embodiments are disclosed, other embodiments of the disclosed technology will also become apparent to those skilled in the art in the following detailed description, which illustrates and describes exemplary embodiments of the disclosed technology. As will be realized, the disclosed technology can be modified in various aspects, all of which will not depart from the spirit and scope of the disclosed technology. Therefore, the drawings and detailed description should be regarded as exemplary, not restrictive. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 An exemplary computing device consistent with certain aspects of the present inventive concepts is shown.

[0016] Figure 2 is a diagram illustrating input and output features of a Candidate Explorer (CE) system according to certain aspects of the present inventive concepts.

[0017] Figure 3 is a graph showing a polynomial regression analysis of CE scores and mean percentages of validated mutation-phenotype associations according to certain aspects of the present inventive concepts.

[0018] Figure 4 is a graph illustrating a CE score receiver operating characteristic (ROC) curve according to certain aspects of the present inventive concepts.

[0019] Figure 5A is a table showing CE performance of flow cytometry phenotypes according to certain aspects of the present inventive concepts.

[0020] Figure 5B is a case showing the co-localized mutation fraction according to certain aspects of the present disclosure.

[0021] Figure 6 is a table showing examples of input features of a machine learning model of a CE system according to certain aspects of the present inventive concepts.

[0022] Figure 7 is a table showing determination algorithm score rules according to certain aspects of the present inventive concepts.

[0023] Figure 8 is a graph illustrating a ROC curve of algorithm scores according to certain aspects of the present inventive concepts.

[0024] Fig. 9 is a table showing flow cytometry screening parameters according to certain aspects of the present inventive concepts.

[0025] Fig. 10A is a graph of the number of good / excellent phenotypic associations versus gene count according to certain aspects of the present inventive concepts.

[0026] Fig. 10B Shown are the number of good / excellent gene associations plotted versus flow cytometry parameters according to certain aspects of the present inventive concepts.

[0027] Fig. 10C The numbers and percentages of essential and non-essential genes according to certain aspects of the inventive concepts are shown.

[0028] Fig.11 is a flow chart illustrating exemplary operations of mutation processing according to certain aspects of the present inventive concepts.

[0029] After reading the entire disclosure, it will be apparent to those skilled in the art that the steps shown in the above figures may be performed in an order different from that described, and one or more of the steps shown in these figures may be optional.

[0030] Detailed Description Certain aspects of the inventive concept relate to methods and systems for using machine learning algorithms to identify chemically induced mutations that cause a screened phenotype. For example, if a gene is independently targeted for knockout or reconstitution mutations, the Candidate Explorer (CE) system can determine the probability that the mutation is validated as causing the screened phenotype. The CE system (also referred to as "CE") uses some parameters (e.g., 67 parameters) in the mapping data, including gene, mutation, genotype, allele, and phenotype information, to determine the CE score and validation probability.

[0031] Forward genetic studies use meiotic mapping to demonstrate that specific mutations, often induced by germline mutations, are responsible for specific phenotypes. Particularly in small pedigrees, co-segregation of multiple mutations, occasional ignorance of mutations, and lack of homozygosity can lead to misjudgments of causality. Certain aspects of the inventive concept provide systems for improving the identification of mutations that can be identified in mice that cause immune phenotypes. The CE system can use machine learning to integrate features of gene mapping data into a single numerical score that can be mathematically converted to a probability of validation for any putative mutation-phenotype association.

[0032] The CE system can be used to evaluate putative mutation-phenotype associations generated by screening for damaging mutations in mouse genes (e.g., approximately 55%) for effects measured by flow cytometry of immune cells in the blood. The CE system identifies more than half of the genes in which mutations are likely to cause phenotypic changes in Mus musculus. （ Mouse ） CE data can be used to identify causal variants within quantitative trait loci. Quantitative trait loci are regions of DNA associated with specific phenotypic traits. Clinical geneticists can use CE to link causal variants to rare inherited immune diseases, even without linkage information. CE displays integrated mutation, phenotypic, and linkage data.

[0033] Figure 1 An exemplary computing device 100 according to certain aspects of the present inventive concept is shown. The computing device 100 may include a processor 103 for controlling the overall operation of the computing device 100 and its related components (including input / output devices 109, communication interfaces 111, and / or memory 115). A data bus may interconnect the processor 103, the memory 115, the input / output devices 109, and / or the communication interface 111.

[0034] Input / output (I / O) devices 109 may include a microphone, keyboard, touch screen, and / or stylus through which a user of computing device 100 may provide input, and may also include one or more of a speaker for providing audio output and a video display device for providing text, audiovisual, and / or graphical output. Memory 115 may store software for providing instructions to processor 103 to allow computing device 100 to perform various operations. For example, memory 115 may store software used by computing device 100, such as operating system 117, application programs 119, and / or associated internal databases 121. The various hardware storage units in memory 115 may include volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Memory 115 may include one or more physical persistent memory devices and / or one or more non-persistent memory devices. The memory 115 may include, but is not limited to, random access memory (RAM), read-only memory (ROM), electronically erasable programmable read-only memory (EEPROM), flash memory or other storage technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by the processor 103.

[0035] The communication interface 111 may include one or more transceivers, digital signal processors, and / or additional circuits and software for communicating over any wired or wireless network using any of the protocols described herein. The processor 103 may include a single central processing unit (CPU), which may be a single-core or multi-core processor (e.g., dual-core, quad-core, etc.), or may include multiple CPUs. The processor 103 and related components may enable the computing device 100 to execute a series of computer-readable instructions to perform some or all of the processes described herein. Although in Figure 1115 or other components in the computing device 100 may include one or more caches, such as a CPU cache used by the processor 103, a page cache used by the operating system 117, a disk cache of a hard drive, and / or a database cache for caching the contents of the database 121. For embodiments that include a CPU cache, one or more processors 103 may use the CPU cache to reduce memory latency and access time. The processor 103 may retrieve data from or write data to the CPU cache instead of reading / writing data to the memory 115, which may increase the speed of these operations. In some examples, a database cache may be created to cache certain data in the database 121 in a separate, smaller database in a memory separate from the database, such as in RAM or on a separate computing device. For example, in a multi-tier application, a database cache on an application server does not need to communicate with a backend database server over a network, thereby reducing data retrieval and data operation time. These types of caches and other types of caches may be included in various embodiments and may provide potential advantages in certain embodiments of the software deployment system, such as faster response times and less dependence on network conditions when transmitting and receiving data.

[0036] Forward genetics begins with a phenotype, typically induced by a random germline mutagen, and ends with the discovery of a causative mutation. Certain aspects provide techniques for rapidly identifying causative mutations in mice carrying N-ethyl-N-nitrosourea (ENU)-induced germline mutations. Certain aspects provide techniques involving mutagenizing male inbred strains (e.g., C57BL / 6J) mice (G0) and breeding them on a C57BL / 6J background to produce first generation (G1) male pedigree founders, second generation (G2) females, and third generation (G3) male and female mice. The exomes of all G1 pedigree founders can be sequenced to achieve greater than 99% coverage of the target exome at 10X. The identified variants (e.g., relative to a C57BL / 6J reference genome) are genotyped in G2 and G3 mice prior to phenotypic screening. Using multiple phenotypic screens, G3 mice can be tested for phenotypic variation relative to a population of C57BL / 6J mice or a control population of G3 mice. Demonstration of linkage between a mutant phenotype detected in a screen and a specific mutation is done by automated meiotic mapping (AMM) performed by the linkage analyzer algorithm (or program or software), which tests the null hypothesis for each mutation in the pedigree (e.g., "Mutation A is not associated with phenotypic manifestation in screen α"). Conversely, if the frequency of a mutation associated with a mutant phenotype is higher than would be predicted by chance alone, then it is very likely that the mutation confers that phenotype. Rejection of the null hypothesis with a p-value less than or equal to 0.05 (with Bonferroni correction for multiple comparisons) is considered suggestive of causality. Validation using independently generated alleles can be performed to confirm association.

[0037] Experience with thousands of mutation-phenotype associations identified through AMM and confirmed or excluded by testing CRISPR / Cas9 targeted alleles suggests that the p-value determined by AMM is not the only indicator of causality. The p-value is a measure of the probability that the observed difference could have occurred simply by chance. The lower the p-value, the greater the statistical significance of the observed difference.

[0038] Mutations associated with a phenotype with a p-value less than 0.05 are sometimes not causative. Many other factors, such as the nature of the mutation (benign, damaging, null), the necessity of the gene for preweaning survival, pedigree size, the number of homozygotes tested, the phenotypic effect, the variance characteristics of the data from the relevant screen, the number of different phenotypes caused by the mutation, the presence or absence of co-segregating mutations, and the observation of other alleles with similar effects, can influence the correct choice of a true causative mutation. The CE system described in this paper can estimate the probability of validation of any putative mutation-phenotype association involved in the AMM.

[0039] Analyze changes in immune cell populations, particularly B cells, T cells, conventional and plasmacytoid dendritic cells (DCs), macrophages, neutrophils, natural killer (NK) cells, and NK1.1 + T cells. Flow cytometry analysis of peripheral blood leukocytes from G3 mutant mice carrying ENU-induced mutations allows detection and measurement of cell populations and subpopulations. In some cases, the CE system was used to assess 87,795 mutation-phenotype associations (e.g., P <0.05), from which we identified more than 1270 genes with high and certain probability of verifiable importance in leukocyte development or maintenance. Many of these genes were previously not known to be important in immune function.

[0040] The CE system helps researchers predict whether a mutation associated with an AMM phenotype is a true causative mutation. The CE system evaluates mutation-phenotype associations that pass specific base filters to identify traditionally good candidates. Default filters for the data that can be used include p-value less than 0.05 (Bonferroni correction), number of mice in the pedigree tested ≥10, and number of homozygous reference mice screened ≥2; however, users can also set more stringent criteria. At the core of the CE system is a supervised machine learning algorithm that outputs a numerical score (CE score), a categorical assessment (candidate status), and a validation probability for each mutation-phenotype association based on the input phenotypic data (e.g., from screening), mutation data, genetic data, and meiotic mapping data.

[0041] Reference Figure 1, the processor 103 and / or the memory 115 may be used to implement the CE system. For example, the processor 103 may include a circuit 120 for receiving one or more input features (e.g., receiving at least one of a phenotypic feature, a linkage data feature, a mutation feature, a gene feature, or an algorithm score). The processor 103 may also include a circuit 122 for generating a CE score based on one or more input features. For example, the circuit 122 may be a trained machine learning model. The machine learning model may be trained based on phenotypic evaluations of mice carrying a target null allele or a replacement allele of a candidate gene. The processor 103 may also include a circuit 124 for outputting an indication of an association between a phenotype and a mutation based on the CE score.

[0042] The memory 115 may be coupled to the processor 103 and may store code that, when executed by the processor 103, performs the operations described herein. For example, the memory 115 may include code 130 for receiving one or more input features, code 132 for generating a CE score, or code 134 for outputting an association indication.

[0043] As described above, the CE system may include a machine learning model. The machine learning model may be trained using an objective function. For example, a candidate solution may be provided to the model and evaluated based on a training data set. By comparing the solution to the training data set, an error score (also known as model loss) may be calculated. The machine learning model may be trained to minimize the error score. For example, the machine learning model may be trained to implement a CE system, including a memory 115 and a processor 103. The CE system may be trained based on phenotypic evaluations of mice carrying target null or replacement alleles of candidate genes. During the prediction process, CE is performed four times a day due to the dynamic state of the database, and all the defining features of the original pedigree screening data are used to estimate the probability of candidate validation. CE can be used to query mutation-phenotype associations identified in flow cytometry screening and in bone radiology screening (dual-energy X-ray absorptiometry (DEXA) scanning).

[0044] Figure 22 is a schematic diagram showing input and output features of a CE system according to certain aspects of the present inventive concept. The CE machine learning system 212 can be a supervised machine learning algorithm that outputs a numerical score (e.g., a CE score 214), a classification assessment (e.g., a candidate state 218), and a validation probability 216 for each mutation-phenotype association based on various input features. The input features may include input phenotypic data (e.g., phenotypic features 202), mutation data (e.g., mutation features 206), gene data (e.g., gene features 208), meiotic mapping data (e.g., linkage data features 204), and algorithm scores 210. Mutation features may include a damage score that indicates the likelihood that the protein function associated with the mutation is impaired. The damage score may be generated using an ML system 230, which may use known deleterious mutations and neutral mutations, as described herein. Gene features 208 may include a necessity score that indicates the likelihood that a homozygous mouse for a robust knockout allele of a gene associated with the mutation is lethal before weaning. The necessity score may be generated using an ML system 240, which may be trained using genes known to be non-essential for survival and genes known to be essential for survival. The algorithm score can be a score generated according to a set of rules 260 associated with empirical observations, as described in more detail herein. Meiosis mapping data can be generated using the automatic meiosis mapping (AMM) described herein. The generated CE score 214 can be used to determine the validation probability 216. The CE score and the algorithm score can be used to generate a candidate state 218 (e.g., a candidate for a mutation-phenotype association being excellent, good, potential, or poor).

[0045] As described herein, based on germline retargeting of genes (e.g., 514 genes), the CE system can be trained using the CE training set. The CE training set (e.g., a machine learning model used to train the CE system 212) can contain validated (e.g., 1903 validated) and excluded (e.g., 3013 excluded) mutation-phenotype associations (4916 evaluations in total). Germline retargeting can be performed using CRISPR / Cas9 in mice of a pure reference background (C57BL / 6J or C57BL-6N) to generate knockout alleles of candidate genes. Alternatively, when there is evidence that homozygous lethality of the null allele exists (e.g., using the essentiality scores described herein), or when it is suspected that the N-ethyl-N-nitrosourea (ENU) mutation will cause hypermorphic, neomorphic, or antimorphic effects, the original ENU allele can be recreated by CRISPR / Cas9 targeting (designated as a "replacement" allele). Mice carrying targeted germline knockout or replacement alleles can be expanded to form mouse pedigrees containing reference allele homozygotes (REF), heterozygotes (HET), and variant allele homozygotes (VAR). Compound heterozygous mice with variant alleles of two or more genes can be generated. New mouse pedigrees carrying CRISPR-targeted alleles can be subjected to phenotypic screens in which the original ENU mutation can be used as a hit. In some aspects, a CRISPR-targeted mutation can be considered validated based on the following criteria: (1) the same phenotype is observed with the same direction of change as the original ENU allele, and the p-value is better than 0.01; (2) the same phenotype is observed with the opposite direction of change as the original ENU allele, and the p-value is better than 0.001; or (3) the phenotype is re-observed (e.g., not seen in the original screen), and the p-value is better than 0.001.

[0046] Figure 3300, which shows a polynomial regression analysis of the average percentage of CE scores and verified mutation-phenotype associations according to certain aspects of the present inventive concept. Each data point represents a set of mutation-phenotype associations. The percentage of verified associations (such as the y-axis of FIG. 300) and the CE score range (such as the x-axis of FIG. 300) are plotted in units of 0.01 (such as 0.35 to 0.36, 0.37 to 0.38, and so on), with n=4916 mutation-phenotype associations and 514 CRISPR / Cas9 targeted genes. The CE score (e.g., ranging from 0 to 1) class probability is related to a polynomial function of the actual verification probability of the CRISPR targeted allele, as determined by regression analysis. Combined with the algorithm score, the CE system uses it to specify one of four possible candidate states (excellent, good, potential, or poor) for each mutation-phenotype association. In some aspects, excellent candidates correspond to CE scores ≥ 0.39 and algorithm scores ≥ -0.5; good candidates correspond to CE scores ≥ 0.39 and -4.5 ≤ algorithm scores < -0.5; potential candidates correspond to CE scores ≥ 0.39 and algorithm scores < -4.5 or CE scores < 0.39 and algorithm scores ≥ -0.5, and poor candidates correspond to CE scores < 0.39 and algorithm scores < -0.5.

[0047] In some aspects, good or excellent candidates for CRISPR / Cas9 targeting and further study can be selected. Figure 3 As shown, the CE score is not strictly proportional to the probability of validation, and some "good" or "excellent" candidates may not be validated. Conversely, "potential" and "poor" candidates may sometimes be validated as true positive associations. As more alleles are obtained and tested (i.e., approaching saturation), true candidates may receive higher CE scores and may therefore eventually be validated.

[0048] Figure 4 400 is a graph of a CE score receiver operating characteristic (ROC) curve according to certain aspects of the present inventive concept. The performance of the CE prediction model established using the training set can be evaluated by repeating the 10-fold cross validation method. The area under the curve (AUC) of the ROC curve is 0.943, where the critical value can be set to 0.39, corresponding to the point where the ROC curve has the smallest distance from the upper left corner.

[0049] Figure 5A 5 is a table showing CE performance of flow cytometry phenotypes according to certain aspects of the present inventive concepts 500. As shown therein, a good or better CE ranking may correspond to a precision of about 80% (e.g., correctly calling validated candidates "true", a false discovery rate of 20%) and a recall rate of 87% (e.g., true positive rate).

[0050] Figure 5B501 is a table showing the performance of CE in scoring co-localized mutations according to certain aspects of the present inventive concepts. When two or more mutations co-segregate, the CE system can identify which mutation is the causative mutation (e.g., determined by software, as described herein). Figure 5B As shown in Figure 2, among 961 such samples, CE can identify 76.5% of the causative mutations as the mutation with the highest CE score on average, and CE generally performs better when there is less co-segregation of mutations. With further training, CE performance will continue to improve with the increase in the total amount of screening data (e.g., with an increase in the number of alleles and the overall density of allele series).

[0051] Multiple alleles of a given gene may be subject to a given phenotypic screen, resulting in multiple mutation-phenotype associations for the same gene and phenotype. Each mutation-phenotype association is independently given an estimate of the allele validation probability (AVP) of the associated mutation, derived from a multinomial regression analysis of the CE score and the average percentage of validated mutation-phenotype associations (e.g. Figure 3 In addition, a composite estimate of the probability that one or more mutations (e.g., N mutations) within a gene can be validated as the source of a phenotype (e.g., the gene validation probability (GVP)) is given by the following formula: In a given screen, AVPs of alleles that cause phenotypic changes in the same direction are also included in the calculation.

[0052] Figure 6 600 is a table of machine learning model input features for a CE system according to certain aspects of the present inventive concept. The CE prediction model may include 67 input data features, including 34 phenotypic features (e.g. Figure 2 phenotypic characteristics in 202), 20 linkage analysis characteristics (such as Figure 2 Linkage analysis features 204), 9 mutation features (such as Figure 2 Mutational signatures in 206), 2 gene signatures (e.g. Figure 2 208) and 2 other features (e.g. Figure 2 Algorithm score 210).

[0053] Table 600 provides examples of input features. For example, the phenotypic features may include at least one of the following: the percentage of VAR mice whose screening results overlap with the screening results of B6 mice, the percentage of VAR mice whose screening results overlap with the screening results of REF mice, the difference between HET and VAR results, the direction of the results (whether the mean of the VAR screening results is greater or less than the mean of the REF screening results), the difference between REF and VAR results, the number of female HET mice, the number of female REF mice, the number of male REF mice, the number of male HET mice, the number of male VAR mice, the number of female VAR mice, the characteristics of the phenotype (such as fluorescence activated cell sorting (FACS) T cells), the group characteristics of the phenotype (e.g., FACS screening or bone screening), the number of abnormalities in REF mice, the number of abnormalities in HET mice, the number of abnormalities in VAR mice, the difference between REF and B6 results, the difference between REF and HET results, whether the variance of REF is large (e.g., whether it is above a threshold), whether the variance of HET is large (e.g., whether it is above a threshold), whether the variance of VAR is large (e.g., above a threshold), Whether the average age of mice with the gene mutation / phenotype is older than the average age of all mice tested with the phenotype, whether the average age of VAR mice is younger than the average age of REF mice, the number of pedigrees of the gene / phenotype, the direction of the position superpedigree results of the gene mutation / phenotype, the number of significant single pedigrees in the significant position superpedigree of the mutation / phenotype (for example, a significant pedigree refers to a linkage analysis of a pedigree or superpedigree by AMM, where the p-value of the mutation-phenotype association is <0.05), The number of pedigrees included in the significant position super-pedigree result of the mutation / phenotype, the direction of the genetic super-pedigree result (null alleles) for the phenotype, the direction of the genetic super-pedigree result (null + missense alleles) for the phenotype, whether the untrimmed data has a corresponding trimmed result (for example, when only the VAR result is greater than the REF result, where the trimmed result is the original data normalized for cell viability), how close the VAR result is to the B6 result, how close the HET result is to the B6 result, how close the REF result is to the B6 result, or whether the REF and B6 results are different.The linkage characteristics may include at least one of the following: the average number of linkage analyzer runs with a p-value < 0.00005 for each allele of the gene, the number of phenotypes with a significant selective gene super-pedigree result for the gene, the number of linkage analyzer runs with a p-value < 0.00005 for the gene, the number of pedigrees in the selective gene super-pedigree and whether the result is significant for the gene / phenotype, the number of pedigrees that contribute to a significant gene super-pedigree result (null allele), the number of pedigrees in a significant gene super-pedigree result (null allele), the minimum p-value of a single linkage analyzer result for the mutation / phenotype, the percentage of weight screening with a p-value < 0.0001 for the mutation, the percentage of FACS screening with a p-value < 0.0001 for the mutation, whether the gene super-pedigree result is significant for the phenotype ( null+missense), whether the p-values ​​in the original and normalized analyses for this gene mutation / phenotype are significant, whether the minimum p-value applies to a recessive mode of inheritance (rather than dominant or additive inheritance), whether this phenotype is driven by other mutations, the percentage of DSS screens with a p-value < 0.0001 for this mutation, the number of FACS phenotypes in FACS phenotypes with a p-value < 0.0001 for this mutation, the number of Dejerine-Sottas syndrome (DSS) phenotypes with a p-value < 0.0001 for this mutation, the number of weight phenotypes with a p-value < 0.0001 for this mutation, whether the positional superpedigree result is significant for this mutation / phenotype, whether the genetic superpedigree result is significant for this phenotype (null allele), or whether the genetic superpedigree result is significant for this phenotype (missense allele).

[0054] A mutation signature (e.g., mutation signature 206) may include at least one of the following: the damage score of the mutation, the number of alleles the gene has, whether the mutation is an autosomal mutation, whether the mutation co-localizes with another mutation of the phenotype, whether the mutation co-localizes with a verified mutation of the phenotype, whether the mutation co-localizes with an excluded mutation of the phenotype, whether the mutation co-localizes with a mutation with a higher damage score, the number of splice variants of the gene containing the mutation, or the ratio of the number of named mutations to the number of accidental mutations of the amino acid change. Gene features may include at least one of the following: the p-value of the lethal phenotype or the probability that the gene is an essential gene (e.g., based on the E-score calculated as described herein). Other features (e.g., algorithm score features) may include at least one of the following: the number of phenotypes for which the algorithm score of the mutation is greater than or equal to -0.5 or the algorithm score of the mutation / phenotype. Although various features are shown in Table 600, only a subset of these features may be used, such as the features shown in bold font.

[0055] In some aspects, the impairment score and the essential score (E-score) are derived from separate machine learning programs. The rule-based algorithm score is derived from a computational execution of a fixed algorithm.

[0056] A damage score (e.g., from 0 to 1) is a mutational signature that has important biological implications. The damage score represents the probability that a protein has impaired function and is determined by a machine learning algorithm that integrates independent prediction scores from the database of nonsynonymous function predictions (dbNSFP) in humans (e.g., 37 points) and the probability that a mouse mutation causes a protein that is damaged to phenotypic variation. Higher scores indicate that the mutation is more likely to be deleterious and, therefore, more likely to be causative (although not always). Damage score prediction models can be implemented using machine learning models trained on known deleterious mutations (e.g., 871 mutations) and known neutral mutations (e.g., 1797 mutations). Mutations with known effects (e.g., 666 mutations) can be used to test the performance of the established model, resulting in an ROC curve with an AUC of 0.852. A deleterious mutation is a genetic change that increases susceptibility or predisposition to a disease or disorder. A neutral mutation is a mutation that is neither beneficial nor detrimental to the organism's ability to survive and reproduce.

[0057] The E-score (e.g., ranging from 0 to 1) is a gene feature that indicates the likelihood that mice homozygous for a robust knockout allele of a gene will be lethal before weaning age (e.g., 4 weeks postpartum). The E-score is calculated by a machine learning algorithm that combines various independent features of the gene, including gene conservation, protein-protein interaction networks, expression phase, and viability / proliferation capacity of human cell lines in which the gene is mutated. The machine learning model used to generate the E-score (e.g., also referred to as an E-score prediction model) can be trained on lethal / survival mutations. The E-score prediction model can be trained on a monthly basis. The training dataset can include known non-essential genes (E-score = 0) (e.g., 3538 non-essential genes) and known essential genes (E-score = 1) (e.g., 2070 essential genes), which are determined based on annotations in the Mouse Genome Informatics (MGI) database and the observed effects of CRISPR-targeted null mutations generated in C57BL / 6J mice. The cutoff can be set to greater than 0.5 for essential genes and less than 0.5 for non-essential genes to inform gene targeting efforts where knockout alleles or replacements identical to the original ENU allele are created to validate the phenotype. Genes known to have an effect on viability (e.g., 1041 genes) can be used to test the performance of the established model, which could produce an ROC curve with an AUC of 60.894.

[0058] Assessment of mutation-phenotype associations can be performed using a human-developed algorithm that outputs a point-based score called the algorithm score (e.g., ranging from -13.5 to 3.5). The algorithm score appears twice in the important features that contribute to the CE algorithm and provides an overall assessment of the likelihood of the mutation being causal.

[0059] Figure 7 700 is a table showing rules for determining an algorithm score according to certain aspects of the present inventive concepts. The algorithm includes a set of rules based on empirical observations. For each feature that supports or opposes the authenticity of a mutation-phenotype association, respectively, the algorithm score is increased or decreased. The features used in the algorithm score calculation are similar to those used in the CE machine learning algorithm, but are static (e.g., not affected by new training data), and the performance of the rule-based algorithm itself cannot reach the performance of the CE prediction model. Each mutation-phenotype association starts with an algorithm score of zero, which is calculated based on Figure 7 The rules described are adjusted.

[0060] Figure 8 800 is a graph showing a ROC curve 802 of the algorithm score. As shown, the AUC of the ROC curve 802 is 0.733, which is lower than the performance of the CE prediction model with an AUC of 0.943. Other input features of the CE algorithm, such as the linkage data feature 204, can be generated by an algorithm called an algorithm driver, which evaluates linked and unlinked candidate mutations to determine the best candidate mutation. A group of linked mutations sometimes cannot undergo meiotic separation; therefore, there may be more than one mutation as a candidate mutation to determine the causality of the phenotype. In other cases, as a matter of chance, a homozygote for a non-causal, non-linked mutation may also be a homozygote for a causal mutation. Typically, this occurs when the number of homozygotes for a non-causal mutation is small. The algorithm driver omits all cases where two mutations share zygosity and recalculates the p-value to test for deviations from the null hypothesis in recessive, additive, and dominant transmission models and determine which mutation is a more robust candidate mutation for causality. Such a mutation is designated as a "driver" state. CE can identify causal variants from a group of colocalizing variants, giving them significantly higher CE scores, based on "driver" status and other factors (e.g., which variant is most damaging, which variant is most necessary for survival to weaning age, which variant has evidence that other alleles have a similar phenotype).

[0061] Finally, allele series probed by phenotypic screens provide important clues to causality and are considered in CE assessment. If multiple alleles of the same gene are associated with the same phenotype, this strongly indicates that a mutation in that gene causes the observed phenotype. Superpedigrees (complexes of multiple pedigrees assayed in the same screen) are of three types. Genetic superpedigrees pool different (non-identical) alleles in a given gene that have been subjected to the same screen. Positional superpedigrees pool only identical alleles that may have arisen from: 1) chance mutations at the same nucleotide; 2) propagation of a single mutation across multiple G1 progeny of a single G0 mouse; or 3) background mutations present in the mutagenized population and shared by multiple G0 mice. Selective genetic superpedigrees only include alleles that are associated with a p-value < 0.05 in a given phenotypic screen and have a common direction of action, thus being intentionally biased against mutational effects. Because many (but not all) ENU-induced mutations are superpedigrees, a selective genetic superpedigree targeting a set of mutations in a particular gene may strongly implicate that gene in the phenotype being explored in the associated screen. The number of pedigrees (and alleles) tested is also important; for very large genes, hundreds of alleles may have been tested, and finding a two or three allele fraction in a particular screen may be due to chance alone. The CE system takes this into account when calculating the probability of causality.

[0062] Fig. 9 900 is a table showing flow cytometry screening parameters according to certain aspects of the present invention. As shown, flow cytometry screening detected 42 parameters of peripheral blood cells, measured the frequency of various immune cell populations and the expression levels of several cell surface markers. In the flow cytometry screening, AMM tested 7109669 mutation-phenotype associations, of which 87795 passed the default initial filter, allowing CE to analyze. These putative mutation-phenotype associations come from 39685 mutations in 14809 genes, which are present in 142653 G3 mice from 3987 pedigrees. Restricted to good or excellent candidates, the number of mutation-phenotype associations has been reduced to 7676, and these mutations come from 2336 mutations in 1279 genes, distributed in 1634 pedigrees.

[0063] Fig. 10A , 10B , and 10C show gene-phenotype association characteristics of genes having at least one good / excellent mutation-phenotype association according to certain aspects of the present inventive concepts. Fig. 10A is a graph 1000 showing the number of good / excellent phenotypic associations plotted against gene count. Fig. 10B Plots showing the number of good / excellent gene associations versus flow cytometry parameters. Fig. 10C The number and percentage of essential and nonessential genes are shown.

[0064] Various observations can be made regarding gene-phenotype associations. First, Fig. 10A As shown, among the 1279 genes, mutations in most genes (872 genes, 68.2%) may result in three or fewer good / excellent phenotypic associations, and 533 genes (41.7%) have only one good / excellent phenotypic association. In contrast, only 30 genes (2.3%) may have at least 20 good / excellent phenotypic associations, of which 26 are well-known immune regulatory genes. Second, the number of good / excellent gene associations can vary greatly depending on the cell type affected, such as Fig. 10B As shown in Figure 2, B cell and T cell phenotypes were associated with the most genes, while conventional and plasmacytoid DC phenotypes were associated with very few genes. Fig. 10C As shown, 449 genes (35.1%) known or predicted to be essential for survival (E-score > 0.55 in this case) could be associated with at least one flow cytometric phenotype, suggesting that many genes important for development may also have postnatal functions in leukocytes.

[0065] Regardless of current candidate status, 1354 mutations in 667 genes rated as good / excellent by CE and suspected or confirmed to cause flow cytometry phenotypes can be assigned allele names and annotated as phenotypic mutations in the Mutagenetix database. Although the named alleles are likely causative, it is not certain that the unnamed alleles are also causative; in fact, 27% of the named alleles have an AVP ≤ 0.5. Some unnamed alleles are designated as "linked" to or "driven" by another mutation in the same pedigree. This may indicate that they are not causative, but it is not always guaranteed, and in some cases two named alleles are linked, suggesting that both mutations may be declared causative (e.g., even though they may co-segregate). CRISPR / Cas9 targeting technology can provide evidence for this dual causality.

[0066] Highly represented Gene Ontology (GO) annotations associated with 667 genes with named alleles could be identified. For genes related to immune system processes (211 genes, P =9.82e-42), lymphocyte activation (113 genes, P =5.21e-39), immune system development (117 genes, P=9.73e-36) and other terms related to immune development / regulatory processes, with the highest enrichment for biological process annotations, which is consistent with the manual assessment identifying 281 of the 667 genes (42.1%) as previously known immune regulators. Through manual assessment, 386 genes represented "novel" immune important genes, each of which was used in normal flow cytometry analysis. The mutant alleles of many of these genes may not have existed in mice before, and no original immunological or other phenotypic data are available. This may be partly due to the fact that null alleles (E-score>0.5) for 146 of these 386 genes result in known or predicted lethality. The enriched GO terms associated with these 386 new immune important genes were primarily metabolic process terms, including cellular metabolic processes (232 genes, P =3.73e-12), organic matter metabolism process (240 genes, P =8.50e-12), cellular macromolecular metabolism (178 genes, P =3.59e-7) and protein metabolism (127 genes, P =0.000264). These 386 genes could be assigned to a set of broadly defined GO annotations for biological processes, regardless of abundance. Each gene could be assigned to any of the 70 parental GO terms associated with it based on its fine-grained GO annotations. Notably, 31 of the 386 genes could be associated with the term "immune system process" based on gene-gene interactions, immune system association of an ancestral gene, sequence homology to another gene associated with an immune system process, or association of an orthologous human gene with an immune system process. In addition, 300 of the 386 genes were detected by RNA sequencing to have moderate (11 to 1,000 transcripts per million) or high (greater than 1,000 transcripts per million) expression in the spleen and / or thymus.

[0067] Currently, a total of 603 genes associated with flow cytometry phenotypes have a GVP>0.5, 332 genes have a GVP>0.8, and 222 genes have a GVP>0.95. Of the genes with a GVP>0.95, 121 (55%) are known to affect flow cytometry measurements, of which 101 (45%) are novel and the flow cytometry phenotypes almost certainly arise from these genes.

[0068] The CE system can rapidly detect mutations and genes that are strongly predicted to affect (or not affect) the phenotype of interest measured in a forward genetic screen. In general, CE offers advantages to human researchers in assessing mutation-phenotype associations because it can incorporate intuitive favorable or unfavorable parameters that are not relevant to linkage analysis, and because it can be assessed more rapidly and on a large scale. Using the numerical CE scores and categorical assessments given by the CE system, it is easy to prioritize mutations for further in-depth study. In addition, causative mutations can often be discerned from a few co-localizing mutations. As millions of coding / splicing mutations are introduced one by one into the mouse genome pedigree, a wider range of alleles may be generated, and nearly all genes in which loss-of-function mutations may be present can be identified with high confidence.

[0069] In addition to being a tool for rapid identification of mutations responsible for ENU-induced phenotypes, the CE system is also very useful to mouse geneticists studying complex traits such as cooperative crosses. Meiotic mapping may restrict the phenotype to a relatively large genomic interval within which there are many candidate genes with mutational differences. If the phenotype is immunological, then knowledge of all the genes from which the flow cytometry phenotype originates is an important starting point for studying causality, where these genes can be targeted.

[0070] CE is also valuable for clinical geneticists seeking to identify causes of human disease. Patients with immunopathology and flow cytometry abnormalities (but without a "classic" disease-causing gene mutation) can be evaluated for additional genetic variants using CE. Mouse gene symbols corresponding to all mutation loci in the patient (as determined by whole-genome or whole-exome sequencing) can be entered into CE and searched as a batch. Genes with flow cytometry abnormalities found in the mouse that are similar to those in the patient can be considered prime candidates. If genetic mapping has been performed in a human family and specific chromosomal regions have been identified, candidate genes can be identified with greater confidence using CE, which also accepts human chromosomal coordinates as search input. CE can also facilitate and expedite the identification of causal variants within disease-associated loci discovered by genome-wide association studies (GWAS) by combining CE with analysis of large human genome / phenotype datasets. Each candidate gene in a GWAS-discovered locus can be queried for associated variant-phenotype associations using CE; mouse gene variants associated with phenotypes similar to the human phenotype under study will suggest causality. Furthermore, in most cases, mutant mice can be ordered immediately, providing models of human disease for laboratory studies. Since most mutations result in loss of function (e.g., not gain of function or new function) and since most mouse genes have human orthologs or homologs, many of these cases can be quickly addressed. Thus, CE is a powerful resource for addressing "missing heritability" issues related to immune abnormalities, and as noted by the 386 novel genes, genes that regulate or mediate cellular metabolic processes may be prime candidates for consideration.

[0071] Mutation-phenotype associations representing genes with one or more variant alleles can be assessed, as well as flow cytometric parameters of peripheral blood leukocytes. Flow cytometric analysis can detect and measure immune cell populations with specific functional relevance and provide insight into the developmental stages that cells undergo. Abnormal flow cytometric patterns are often associated with immune dysfunction, and many immune deficiencies and autoimmune phenotypes may not initially be detected through functional screening per se, but rather through flow cytometric analysis of peripheral blood. Human disease states exhibit similar or identical flow cytometric phenotypes, demonstrating the clinical relevance of many mouse flow cytometric abnormalities. To date, approximately 5% genome saturation has been achieved in screening 42 flow cytometric parameters, from which 1004 genes with good / excellent phenotype associations have been identified that were not previously associated with immune function (from GO analysis of 1279 genes, 275 of these genes were found to be associated with "immune system processes"). Therefore, even with a high false discovery rate of 20%, approximately 456 new immune important genes remain to be discovered.

[0072] When all 1279 genes with at least one good / excellent phenotypic association were extensively surveyed, it was observed that the proportion of genes with one, two, or three good / excellent phenotypic associations (68.2%) was much higher than the proportion of genes with many (≥20) good / excellent phenotypic associations (2.3%). These findings suggest that most genes affecting immune cell populations in the blood have cell type- or phenotype-specific functions. The hypothesis that the same or similar combination of phenotypes affected by two or more genes could indicate that these genes function in a common molecular pathway was investigated. Despite uniform phenotypic testing across all screens, good / excellent gene associations may not affect cell populations with the same frequency. For example, gene associations may be 4.8-fold more common for T cells than for conventional dendritic cells (DCs), 12.5-fold more common for plasmacytoid DCs, and 4.4-fold more common for neutrophils. One simple explanation is that rarer blood cell populations have fewer detectable phenotypic differences, but another possibility reflecting cell biology is that T cells are inherently less tolerant of genetic variation than conventional DCs, plasmacytoid DCs, or neutrophils, at least in terms of the numbers of these cells in peripheral blood. To gain insight into these issues, it is necessary to understand the functions of individual proteins and the pathways that regulate them.

[0073] The vast majority of mice phenotyped by flow cytometry were also phenotyped in other screens, including those measuring immune responses, innate immune responses, body weight, blood pressure, heart rate, dextran sodium sulfate (DSS) sensitivity, circadian rhythmicity, and motor coordination. Currently, data from screens measuring bone phenotypes by DEXA scanning are publicly accessible. In the future, data from other screens may also be made available to public users of CE to explain the various phenotypic consequences of each mutation. All biomedically relevant phenotypic screens may ultimately inspire studies of human phenotypes and help distinguish the mechanisms by which certain alleles cause phenotypes, as many mutations have scores in different screens (e.g., immune function and body weight, or immune function and neurobehavioral function).

[0074] Further details of materials and methods that can be used to identify associations are provided below. C57BL / 6J males at 8 to 10 weeks of age can be mutagenized with ENU. As described above, mutagenized G0 males can be crossed with C57BL / 6J females, and the resulting G1 males can be crossed with C57BL / 6J females to produce G2 mice. G2 females can be backcrossed with their G1 males to produce G3 mice, which can be phenotypically screened. Whole exome sequencing and mapping can be performed.

[0075] To generate mice carrying CRISPR / Cas9 targeted mutations, female C57BL / 6J mice were superovulated by injection of 6.5 U pregnant mare serum gonadotropin (PMSG; Millipore), followed by 6.5 U human chorionic gonadotropin (hCG; Sigma-Aldrich) 48 h later. Superovulated mice were then mated with C57BL / 6J male mice overnight. The next day, fertilized eggs were collected from the oviducts, and in vitro transcribed Cas9 mRNA (50 ng / μL) and small base-paired guide RNA (50 ng / μL) were injected into the cytoplasm or pronuclei of the embryos. Injected embryos were cultured in M16 medium (Sigma-Aldrich) at 37°C and 5% CO2. To generate mutant mice, two-cell stage embryos were transferred into ampoules from the oviducts of pseudopregnant female Hsd:ICR(CD-1) (Harlan Laboratories) (10 to 20 embryos per oviduct).

[0076] Peripheral blood can be collected by buccal bleeding from G3 mice older than 6 weeks of age. Red blood cells (RBCs) can be lysed using hypotonic buffer (eBioscience). Samples are washed once with FACS staining buffer (phosphate-buffered saline (PBS) containing 1% [weight / volume] BSA) and then centrifuged at 500 × g for 5 min. RBC-reduced samples can be stained for 1 hour at 4°C in 100 μL of a 1:200 cocktail of fluorescently conjugated antibodies against 15 cell surface markers including major immune lineages B220 (BD, clone RA3-6B2), CD19 (BD, clone 1D3), IgM (BD, clone R6-60.2), IgD (BioLegend, clone 11-26c.2a), CD3ε (BD, clone 145-2C11), CD4 (BD, clone RM4-5), CD8α (BioLegend, clone 53-6.7), CD11b (BioLegend, clone M1 / 70), CD11c (BD, clone HL3), F4 / 80 (Tonbo, clone BM8.1), CD44 (BD, clone 1M7), CD62L (Tonbo, clone MEL-14), CD5 (BD, clone 53-7.3), CD43 (BD, clone S7), NK 1.1 (BioLegend, clone OK136), and 1:20 00 Fc block (Tonbo, clone 2.4G2). Flow cytometry data can be collected on a cell analyzer (e.g., BD LSR Fortessa), and the proportion of immune cell populations in each G3 mouse can be analyzed using software for analyzing cytometer data. The resulting phenotypic data can be uploaded to a server (e.g., Mutagenetix) for automated mapping of causative alleles.

[0077] AMM can be performed as described herein. For example, the genotypes of all mutation sites in the exome of G3 mice can be determined before phenotypic screening. Whole exome sequencing can be performed on the tail DNA of G1 males using a sequencing instrument (such as Illumina HiSEq. 2500). The mutation sites that have been determined in G2 and G3 mice are then genotyped (such as using IonPGM (Life Technologies)). After the phenotypic screening is completed, the program linkage analyzer can be used to perform linkage analysis of recessive, additive and dominant genetic models for each mutation in the pedigree; the program linkage browser (Linkage Explorer) can be used to display phenotypic data scatter plots and Manhattan plots. The p value of the association between genotype and phenotype can be calculated using the likelihood ratio test of the generalized linear model or the generalized linear mixed effects model and the Bonferroni correction.

[0078] In some aspects, CE prediction models can be constructed using a random forest algorithm (eg, as implemented in the R classification and regression training (caret) package). Linkage data obtained through screening can be released in phases based on phenotype.

[0079] The injury score is an overall score that integrates the independent prediction scores (e.g., 38 scores) using a logistic regression model. 37 prediction scores are available from databases such as human dbNSFP, including scores from the following algorithms: SIFT, SIFT4G, Polyphen2-HDIV, Polyphen2-HVAR, LRT, MutationTaster2, MutationAssessor, FATHMM, MetaSVM, MetaLR, CADD, CADD_hg19, VEST4, PROVEAN, FATHMM-MKL encoding, FATHMM-XF encoding, fitCons (four scores), LINSIGHT, DANN, GenoCanyon, Eigen, Eigen-PC, M-CAP, REVEL, MutPred, MVP, MPC, PrimateAI, GEOGEN2, BayesDel_addAF, BayesDel_noAF, ClinPred, LIST-S2, and ALoFT. The dataset can use the ranked scores of each algorithm converted from dbNSFP. The 38th prediction score is the probability of protein damage caused by epivariant caused by mouse mutation, calculated as described in the paper. Among these 38 prediction scores, the scores from MutPred as well as the probability of protein damage caused by epivariant caused by mouse mutation and phastCons100way_vertebrate (conservative score) are significant. The damage score can be used as a quantitative prediction score to measure the probability that mouse mutation is deleterious.

[0080] If a missense mutation in the mouse is identical to a human mutation (e.g., both a nucleotide and an amino acid change), then the effects of the mutation in humans and mice are likely to be similar. Therefore, the human score can be used to predict the likelihood of damage in mice. A set of mouse ENU mutations with class labels (e.g., known damaging or neutral) can be retrieved from the Mutagenetix database. There are four sources of known mutation class labels: 1) Physically isolated mutations (linked to all other coding / splicing mutations in the pedigree) that are in important genes but can be transmitted from heterozygous G2 females and their heterozygous G1 males to homozygous G3 mice at a rate that does not significantly deviate from Mendelian expectations are considered neutral. 2) Conversely, isolated mutations in important genes that are not transmitted to homozygotes, i.e., the observed homozygous frequency is significantly lower than the Mendelian expected rate, are considered damaging. 3) Mutations that cause a qualitative (usually observable) phenotype are considered damaging mutations. 4) Mutations that have been shown to be significant in phenotypic screens of CRISPR replacement alleles are also considered damaging.

[0081] Mutations labeled as damaging or neutral can be transformed (lifted over) from the mouse genome to the human genome (translated into equivalent amino acids), and mutations that result in identical nucleotide and amino acid changes in both genomes are retained. Using the transformation tools used to convert genomic coordinates and annotation files between assemblies, approximately 4% of mouse mutations may not be mapped to corresponding human mutations. These mutations may not be included in the final dataset for model training and testing. Point-biserial correlation can be used to estimate the relationship between the mouse mutations labeled as damaging or neutral and the prediction scores of the most important human mutations. The correlation coefficient may be 0.525, 95% CI: 0.50 to 0.55. The corresponding human mutations can then be searched in the dbNSFP database to obtain scores for all available prediction methods. The retrieved scores, combined with the probability of phenotypically detectable damage for the mouse mutation, can be integrated with the input dataset and a logistic regression model can be trained and optimized using the train function of the R caret package and 10-fold cross validation. This process can be repeated multiple times (e.g., three times). Scaling of the data can be performed by the preprocessing function. The constructed model (classifier) ​​can then be used to calculate scores for a set of mutations with unknown class membership. The dataset used for prediction can be created in the same way as the dataset used for modeling. The score predicted by the model represents the probability that the mutation belongs to the damage class. The higher the score, the more likely the mutation is harmful.

[0082] The input dataset may contain mouse mutations (e.g., 3334 mutations), of which a portion (e.g., 1088) are deleterious and another portion (e.g., 2246) are neutral. In order to evaluate the performance of the constructed model in predicting membership in a new mutation class, the input dataset may be randomly divided into two groups: the first group consists of mutations (e.g., 2668 mutations, 80% of the original dataset, 871 deleterious mutations, and 1797 neutral mutations), which can be used to train and validate the logistic regression model; the second group consists of the remaining 666 mutations, which can be used to test the performance of the constructed model. The 80 / 20 split for training and testing may be performed randomly multiple times (e.g., 10 times). The AUC value of the ROC curve may be close to the average AUC value of 0.853±0.014.

[0083] The E-score can be used to estimate the likelihood of lethality in mice when a gene is knocked out. Essential and nonessential genes in mice can be distinguished by various independent features of the genes. Logistic regression was used to fit the features of known essential and nonessential genes in mice to obtain a training model for predicting the essentiality of unknown genes.

[0084] The model used the following gene features: 1) gene conservation in the Online Gene Essentiality Database (OGEE), connectivity in protein-protein interaction networks, expression stage during development, evolutionary age, GO terms, gene copy number, and gene product length, where the features are related to gene essentiality in many species including mice, 2) essentiality of human orthologous genes: genes that are essential for cell proliferation and survival in the tested cell lines under specific conditions can be defined as essential genes, and the frequency of being important in the tested human cell lines can be used as features in the model, 3) the probability of loss-of-function intolerance (pLI) score from the Exome Aggregation Consortium (ExAC): the closer the score is to 1, the more likely the gene is essential for human survival; 4) the minimum p-value of ENU-targeted mouse genes obtained from the lethal model by the linkage analyzer algorithm.

[0085] Phenotypic descriptions of 8032 genes in Mouse Genome Informatics (MGI) that can be knocked out in mice were carefully reviewed and manually selected to select a set of genes designated as "essential" or "non-essential" based on the following criteria: 1) If the homozygous knockout allele is clearly described as causing embryonic lethality, neonatal lethality, prenatal lethality, perinatal lethality, or preweaning lethality, then the gene can be considered essential for preweaning survival and can be classified as an essential gene, and the gene is assigned an E-score of 1, 2) If the homozygous knockout allele is compatible with survival, normal growth, no obvious phenotype, or has a certain phenotype but no obvious effect on survival, it can be classified as a non-essential gene, and the gene is assigned an E-score of 0. In addition, genes that cause significant lethality before weaning as validated by our CRISPR knockout experiments can be assigned an E-score of 1, and genes that cause normal Mendelian results in heterozygous mutant crosses as validated by our CRISPR knockout experiments can be assigned an E-score of 0.

[0086] A set of 7009 genes, of which 2587 can be labeled as essential genes and 4422 can be labeled as non-essential genes, can be integrated with the gene signature. The resulting dataset can be used to train and optimize a logistic regression model using the training function of the R caret package and 10-fold cross validation, which can be repeated multiple times (e.g., three times). Scaling of the data can be performed by the preProcess function. The preProcess function estimates the parameters required for each operation. The constructed model can then be used to predict the essentiality of the remaining mouse genes. The prediction score ranges between 0 and 1, and the closer the score is to 1, the more likely the gene is essential.

[0087] To evaluate the performance of the constructed model in predicting the necessity of unknown genes, the dataset used to construct the model can be randomly divided into two groups: one group includes 5608 genes (80% of the original dataset, 3538 non-essential genes and 2070 essential genes) for training and validating the logistic regression model; the remaining 1401 genes are used to test the performance of the constructed model in the training dataset. The 80 / 20 split method for training and testing can be randomly performed multiple times (e.g., 10 times); the AUC value generated by the ROC curve is close to the average AUC value of 0.891±0.0087.

[0088] Fig.11 is a flow chart illustrating exemplary operations 1100 of mutation processing according to certain aspects of the present inventive concepts. For example, operations 1100 may be performed by a CE system such as processor 103 and memory 115.

[0089] At block 1102, the CE system may receive one or more input features including phenotype data and mutation data. At block 1104, the CE system may generate a CE score indicating a probability of association between a phenotype and a mutation based on the one or more input features through a machine learning model.

[0090] In some aspects, the mutation data may also include a damage score that indicates the likelihood that the protein function associated with the mutation is impaired. For example, the CE system may generate a damage score by using another machine learning model trained with known deleterious mutations and neutral mutations.

[0091] In some aspects, one or more input features may also include a necessity score indicating the likelihood that a homozygous mouse of a robust knockout allele of a gene associated with a mutation will die before weaning age. The CE system may generate a necessity score by using another machine learning model trained with genes known to be non-essential for survival and genes known to be essential for survival. In some aspects, one or more input features also include features associated with an algorithm score indicating the likelihood that a mutation is causal.

[0092] In some aspects, the one or more input features also include linkage data generated using automated meiotic mapping (AMM) (e.g., performed by a linkage analyzer algorithm or program). In some aspects, when two or more mutations co-segregate, the CE system can determine which of the two or more mutations is a more robust causal candidate for the phenotype by omitting shared zygosity instances of the two or more mutations. In block 1104, a CE score can be generated based on the determination result.

[0093] In some aspects, one or more input features may include one or any combination of the following: the number of phenotypes for which the algorithm score of the mutation meets a threshold, the algorithm score indicating the likelihood that the mutation is causative; the average number of automated meiotic mapping (AMM) operations that result in a p-value that meets a threshold for each allele of a gene associated with the mutation; the algorithm score of the mutation or phenotype; the number of AMM operations that result in a p-value that meets a threshold for a gene associated with the mutation; the damage score of the mutation, which indicates the likelihood that the protein function associated with the mutation is impaired; the number of pedigrees in a superpedigree associated with the gene and whether the p-value resulting from the AMM operations on the superpedigree meets a threshold; the number of phenotypes for which the p-value of the superpedigree meets the threshold; the number of pedigrees that contribute to the p-value of the superpedigree that meets the threshold; the number of pedigrees in a superpedigree; the p-value meets the mutation Percentage of fluorescence-activated cell sorting (FACS) screens that meet the threshold; minimum p-value from AMM manipulation; percentage of variant allele (VAR) mice whose screen results overlap with those from B6 mice; whether superpedigree AMM manipulation results meet the threshold for null and missense alleles; whether superpedigree AMM manipulation results meet the threshold for null alleles; percentage of VAR mice whose screen results overlap with those from reference allele (REF) mice; difference between AMM manipulation results from heterozygous (HET) mice and VAR mice; number of female REF mice used for AMM manipulation; percentage of body weight screens with p-values ​​meeting the mutation threshold; number of female HET mice used for AMM manipulation; or difference between AMM manipulation results from REF mice and VAR mice.

[0094] At block 1106, the processing system can output the CE score. For example, in some aspects, the processing system can generate a candidate status (e.g., an excellent, good, potential, or poor candidate) for the association between the phenotype and the mutation based on the CE score and an algorithmic score indicating the likelihood that the mutation is causal.

[0095] These and other various arrangements will be described more fully herein. It will be understood by those skilled in the art when reading the following disclosure that the various aspects described herein may be a method, a computer system or a computer program product. Therefore, these aspects may take the form of a complete hardware implementation, a complete software implementation, or at least one implementation combining software and hardware aspects. In addition, these aspects may also take the form of a computer program product stored by one or more computer-readable storage media (e.g., a non-transient computer-readable medium), wherein or on the storage medium, a computer-readable program code or instruction is contained. Any suitable computer-readable storage medium may be used, including a hard disk, a CD-ROM, an optical storage device, a magnetic storage device and / or any combination thereof. In addition, the various signals representing data or events described herein may be transmitted between a source and a destination in the form of electromagnetic waves through a signal-conducting medium, and the signal-conducting medium is, for example, a metal wire, an optical fiber and / or a wireless transmission medium (e.g., air and / or space).

[0096] Embodiments of the present inventive concept include various steps, which are described in this specification. These steps can be performed by hardware components or embodied in machine executable instructions, which can be used to perform these steps by a general or special processor programmed with these instructions. In addition, these steps can also be performed by a combination of hardware, software and / or firmware.

[0097] Although specific embodiments are discussed, it should be understood that this is for illustrative purposes only. Those skilled in the relevant art will recognize that other components and configurations may also be used without departing from the spirit and scope of the present invention. Therefore, the following description and drawings are illustrative only and are not to be construed as limiting. Many specific details are described herein to provide a thorough understanding of the present invention. However, in some cases, in order to avoid ambiguous descriptions, well-known or conventional details are not described. Reference to an embodiment or an embodiment in the present invention may refer to the same embodiment or any embodiment; such reference refers to at least one embodiment.

[0098] Reference to an "embodiment" or "one embodiment" means that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the inventive concept. The phrase "in one embodiment" appearing in different places in the specification does not necessarily refer to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. In addition, various features may be present in some embodiments and not in others.

[0099] The terms used in this specification generally have their ordinary meaning in the art, within the context of the inventive concept and within the specific context in which each term is used. Any one or more of the terms discussed herein may use alternative language and synonyms, and no special meaning should be given to whether a term is elaborated or discussed in detail herein. In some cases, synonyms for certain terms are provided. The description of one or more synonyms does not exclude the use of other synonyms. The examples used anywhere in this specification (including examples of any term discussed herein) are merely exemplary and are not intended to further limit the scope and meaning of the inventive concept or any example term. Similarly, the inventive concept is not limited to the various embodiments given in this specification.

[0100] Without intending to limit the scope of the inventive concept, examples of instruments, devices, methods and related results thereof implemented according to the inventive concept are given below. Please note that titles or subtitles may be used in the examples for the convenience of the reader, but this should in no way limit the scope of the inventive concept. Unless otherwise defined, the technical and scientific terms used herein have the meanings commonly understood by those of ordinary skill in the art to which the inventive concept belongs. In the event of a conflict, this document shall prevail, including definitions.

[0101] Other features and advantages of the inventive concept will be described in the following description, and some features and advantages will be apparent from the description, or can be understood by practicing the principles disclosed herein. The features and advantages of the inventive concept can be realized and obtained by the instruments and combinations specifically pointed out in the appended claims. These and other features of the inventive concept will become more apparent from the following description and the appended claims, or can be understood by practicing the principles set forth herein.

Claims

1. A mutation treatment method, comprising: receiving one or more input features including phenotypic data and mutation data; generating, by a machine learning model, a candidate explorer (CE) score indicating a probability of an association between a phenotype and a mutation based on the one or more input features; as well as Based on the CE score, an indication of the association between the phenotype and the mutation is output.

2. The method of claim 1, wherein the indication of the association comprises a candidate state for the association based on the CE score and an algorithmic score indicating the likelihood that a mutation is causative.

3. The method of claim 1, wherein the mutation data comprises a damage score indicating the likelihood that protein function associated with the mutation is impaired.

4. The method according to claim 3, further comprising: The damage score is generated by another machine learning model trained using known deleterious and neutral mutations.

5. The method of claim 1, wherein the one or more input features further comprise an essentiality score indicating the likelihood that mice homozygous for a robust knockout allele of a gene associated with the mutation will be lethal before weaning age.

6. The method according to claim 5, further comprising: The essentiality score is generated by another machine learning model trained using genes known to be non-essential for survival and genes known to be essential for survival.

7. The method of claim 1, wherein the one or more input features further comprise a feature associated with an algorithm score indicating a likelihood that the mutation is causative.

8. The method of claim 1, wherein the one or more input features further comprise linkage data generated using automated meiotic mapping (AMM).

9. The method according to claim 1, further comprising: When two or more mutations are co-segregated, determining which of the two or more mutations is a more robust causative candidate for the phenotype by omitting shared zygosity instances of the two or more mutations, wherein the CE score is generated based on the determination.

10. The method of claim 1, wherein the one or more input features include at least one of the following: the number of phenotypes for which the algorithm score of the mutation meets a threshold, the algorithm score indicating the likelihood that the mutation is causative; the average number of AMM operations that resulted in a p-value meeting the threshold for each allele of the gene associated with the mutation; Algorithmic score for mutation or phenotype; the number of AMM operations that resulted in a p-value that met the threshold for genes associated with the mutation; a damage score for the mutation, the damage score indicating the likelihood that protein function associated with the mutation is impaired; the number of pedigrees in the superpedigree associated with the gene, and whether the p-value resulting from the AMM operation of the superpedigree meets the threshold; The number of phenotypes whose p-values ​​of the superpedigree meet the threshold; the number of pedigrees that contribute to the p-value of the superpedigree that meets the threshold; the number of pedigrees in the superpedigree; Percentage of fluorescence-activated cell sorting (FACS) screens with p values ​​meeting mutation thresholds; Minimum value of p-value from AMM operations; The percentage of variant allele (VAR) mice whose screening results overlap with B6 mice; Whether the AMM operation results of the superpedigree meet the thresholds for null and missense alleles; Whether the AMM operation results of the superpedigree meet the threshold of null alleles; The percentage of VAR mice whose screening results overlapped with the reference allele (REF) mice; The difference between the AMM manipulation results of heterozygous (HET) and VAR mice; Number of female REF mice used for AMM manipulation; Percentage of bodyweight screens with p-values ​​meeting the mutation threshold; The number of female HET mice used for the AMM procedure; or Difference between the AMM manipulation results of REF and VAR mice.

11. A mutation processing device, comprising: Memory; as well as One or more processors coupled to the memory, configured to: receiving one or more input features including phenotypic data and mutation data; generating, by the machine learning model, a candidate explorer (CE) score indicating a probability of an association between the phenotype and the mutation based on the one or more input features; and Based on the CE score, an indication of the association between the phenotype and the mutation is output.

12. The apparatus of claim 11, wherein the indication of association comprises an association candidate state based on the CE score and an algorithmic score indicating a likelihood that a mutation is causal.

13. The apparatus of claim 11, wherein the mutation data comprises a damage score indicating the likelihood that protein function associated with the mutation is impaired.

14. The apparatus of claim 13, wherein the one or more processors are further configured to generate a damage score by another machine learning model trained using known deleterious mutations and neutral mutations.

15. The apparatus of claim 11, wherein the one or more input features further comprise a necessity score indicating the likelihood that mice homozygous for a robust knockout allele of a gene associated with the mutation will be lethal before weaning age.

16. The apparatus of claim 15, wherein the one or more processors are further configured to generate the essentiality score by using another machine learning model trained using genes known to be non-essential for survival and genes known to be essential for survival.

17. The apparatus of claim 11, wherein the one or more input features further comprise a feature associated with an algorithm score indicating a likelihood that the mutation is causal.

18. The apparatus of claim 11, wherein the one or more input features further comprise linkage data generated using automated meiotic mapping (AMM).

19. The device according to claim 11, wherein: When two or more mutations co-segregate, the one or more processors are also configured to determine which of the two or more mutations is a more robust causal candidate for the phenotype by omitting shared zygosity instances of the two or more mutations, wherein the one or more processors are configured to generate the CE score based on the determination.

20. A non-transitory computer-readable medium having stored therein instructions which, when executed by one or more processors, cause the one or more processors to: receiving one or more input features including phenotypic data and mutation data; Based on the one or more input features, generating a candidate explorer (CE) score indicating a probability of association between the phenotype and the mutation by a machine learning model; and Based on the CE score, an indication of the association between the phenotype and the mutation is output.

Images