Genetic data processing

Abstract

This invention provides methods, computer programs, computer-readable storage media, and systems that offer improved solutions for processing genetic data. [Solution] The processing method comprises the steps of: S10 obtaining a dataset containing microbial gene expression data of multiple patients; S20 clustering multiple patients into a set of clusters based on the microbial gene expression data; and S50 determining genes exhibiting expression that explains the clustering by selecting genes included in a set of metabolic pathways and / or identifying genes that show significantly different expression between clusters, thereby forming an improved solution for patient stratification based on metagenomic data.

Description

Technical Field

[0001] The present disclosure relates to the field of computer programs and systems, and more particularly, to methods, systems, and programs for gene data processing.

Background Art

[0002] Metabolic diseases including obesity impose a significant economic burden, which is estimated to be between $212.4 billion and $315.8 billion in 2010, showing a 48.7% increase. Factors contributing to this increase include the increased cost per obese individual, population growth, and the rising prevalence of obesity. It is notable that not only has obesity increased rapidly, but metabolic diseases such as atherosclerosis, cirrhosis, and type 2 diabetes have also increased, highlighting the urgency of research in this area as a public health priority.

[0003] One potential explanation for the increase in these diseases is the change in the living environment, which affects both an individual's health and their microbiome. In some studies, it has been recognized that the gut microbiota plays a central role in shaping the host's metabolic health by producing various reactive components that can affect the host's metabolism.

[0004] In parallel, advances in technologies such as metagenomics have enabled the scientific community to deepen its understanding of the relationships between different parts of the gut microbiota. Metagenomics is based on the analysis of genetic material, primarily DNA, extracted directly from environmental samples such as the fecal microbiota of the human body. This approach provides a holistic view of the entire microbial community. Through high-throughput sequencing technologies such as next-generation sequencing (NGS), metagenomic research generates vast amounts of sequencing data that can be analyzed using bioinformatics tools and software. Overall, this methodology generates enormous amounts of data, ranging from functional gene profiling (providing information on the functional potential of microbial communities by identifying and characterizing genes involved in various biological processes) to taxonomic composition (providing insights into the taxonomic composition of microbial communities by identifying the various types and relative abundances of microorganisms present in a sample).

[0005] These types of data each possess unique structures with different scales, variability, and distributions, hindering simple analyses for extracting meaningful information. Adding to this complexity, human patients exhibit significant variability, even within carefully selected cohorts, due to factors such as phenotypic variability (age, sex, and / or medical history), dietary habits, and differences in treatment response.

[0006] Current analyses generally address heterogeneity by direct consideration (e.g., dimensionality reduction) or by (often arbitrarily) aggregating information into clusters for inter-patient comparisons. Several methodologies have been employed, such as principal component analysis (PCA) and principal coordinate analysis (PCoA), which was developed for dimensionality reduction in metagenomic studies. PCoA centers its analysis on the distance matrix between samples rather than the covariance matrix of samples. PCoA chooses to decompose the distance matrix rather than directly analyzing the original observed data. Dimensionality reduction methods are useful for visualizing high-dimensional metagenomic data and assist in identifying clear patient clusters. Techniques such as k-means clustering, hierarchical clustering, and self-organizing maps (SOMs) are used to identify natural groupings within data without prior labeling. Alternatively, variable selection methods such as LASSO (Least Absolute Shrinkage and Selection Operator) have been employed in the analysis of metagenomic data. [Overview of the project] [Problems that the invention aims to solve]

[0007] These techniques effectively reduce the coefficients of less influential features to zero, thereby facilitating the selection of highly correlated traits. However, a significant limitation when applied to metagenomic data is their inability to adequately handle multicollinearity. Indeed, metagenomic datasets often consist of numerous features, including genes that exhibit high correlation due to functional redundancy or co-occurrence patterns.

[0008] More recently, methodologies like DeepMicro have emerged. This is a deep learning framework that uses various autoencoders to generate robust, low-dimensional representations from high-dimensional microbiome profiles to predict disease states. However, while these methods are primarily developed to predict patients' clinical states through hard clustering, they lack the ability to stratify patients into transitional states, thereby largely ignoring potential intermediate disease states in some individuals. Furthermore, these methodologies fail to uncover common characteristics among different patients within each group, demonstrating limitations in terms of information loss due to dimensionality reduction and multiple projections.

[0009] In this context, improved solutions for processing genetic data are still needed. [Means for solving the problem]

[0010] A computer implementation method for processing genetic data is provided. This method will be hereinafter referred to as the "processing method" or simply the "method." The processing method includes the step of obtaining a dataset containing microbial gene expression data from multiple patients. The method includes the step of clustering the multiple patients into a set of clusters based on the microbial gene expression data. The method includes the step of determining genes exhibiting expression that explain the clustering by selecting genes that are included in a set of metabolic pathways and / or by identifying genes that exhibit significantly different expression between clusters.

[0011] The processing method may include one or more of the following: clustering is a process of performing multiple iterations of latent Dirichlet allocation (LDA), where each iteration assigns each patient to one of multiple topics; determining a graph including nodes representing multiple patients and edges connecting pairs of nodes, where the length of each edge connecting pairs of nodes is a function of the number of times the patients represented by the connected nodes appeared in the same topic over the multiple iterations of latent Dirichlet allocation (LDA); and performing graph-based clustering of the determined graph; the graph is determined based on the Fruchterman-Reingold force-oriented algorithm; The clustering of the database is performed based on the Louvain clustering algorithm; identifying genes that show significantly different expression between clusters includes the step of comparing the expression of each gene between each cluster; and, based on the results of the comparison, the step of retaining genes that have significantly different expression between clusters; the comparison of gene expression is performed based on the Kruskal-Wallis test, which calculates the respective p-value for each gene, and the retained genes optionally have a p-value of less than 1%; a set of metabolic pathways is selected by extracting the metabolic pathways that are most significantly enriched in terms of genes from an existing metabolic pathway database; the selection of the set of metabolic pathways is performed based on Fisher's exact test, which calculates the respective p-value for each metabolic pathway. The selected metabolic pathways in the set optionally have a p-value of less than 5%; and / or, genes whose relative expression exceeds a threshold are filtered from the gene database. The genes are determined from the genes remaining after filtering.

[0012] Furthermore, a computer implementation method for clustering is provided. This method is hereinafter referred to as the "Method of Use." The Method of Use includes the step of obtaining microbial gene expression data of a patient. The microbial gene expression data includes the expression of genes determined in the Processing Method. The Method of Use includes the step of determining the cluster to which the patient belongs based on the gene expression determined for that patient.

[0013] The method of use may include one or more of the following: determining the closest patient from among the patients in the determined cluster; and / or determining the closest patient using the Bray-Curtis dissimilarity index.

[0014] Furthermore, a computer program is provided that includes instructions for performing the processing method and / or usage method.

[0015] Furthermore, a computer-readable storage medium on which a computer program is recorded is provided.

[0016] Furthermore, a system is provided comprising a processor and memory, wherein a computer program is stored in the memory. The processor may optionally be coupled to a graphical user interface.

[0017] Furthermore, a device is provided that includes a data storage medium on which a computer program is recorded.

[0018] The device may form or serve as a non-temporary computer-readable medium, for example, in a SaaS (Software as a Service), other server, or cloud-based platform. Alternatively, the device may comprise a processor coupled to a data storage medium. Thus, the device may form all or part of a computer system (for example, if the device is a subsystem of the entire system). The system may further comprise a graphical user interface coupled to the processor. Non-limiting embodiments are described below with reference to the accompanying drawings. [Brief explanation of the drawing]

[0019] [Figure 1] A flowchart of an embodiment of this method is shown. [Figure 2] A flowchart of an embodiment of this method is shown. [Figure 3] An example of a determined graph is shown. [Figure 4] An example of distance-based cluster analysis is shown. [Figure 5] An example of comparing the results obtained using this processing method with those of a random forest is shown. [Figure 6] An example of this system is shown. [Modes for carrying out the invention]

[0020] Referring to the flowchart of FIG. 1, a computer-implemented method for gene data processing is proposed. This method is hereinafter referred to as the "processing method" or simply the "method". The processing method includes the step of obtaining a data set containing microbial gene expression data of a plurality of patients. This method includes the step of clustering a plurality of patients into a set of clusters based on the microbial gene expression data. This method includes the step of determining genes that exhibit expression explaining the clustering by selecting genes included in a set of metabolic pathways and / or by identifying genes that exhibit significantly different expression between clusters. This method forms an improved solution for gene data processing.

[0021] One of the main advantages worthy of note is that the method first enables clustering a plurality of patients based on their microbial gene expression data, thereby identifying groups of patients with similarities in microbial gene expression. Furthermore, the method can efficiently determine genes that exhibit expression explaining the clustering. In fact, the method can separate genes that are significantly expressed between groups and thereby confirm that these genes actually promote the stratification of patients within these clusters. This is done, for example, by comparing genes enriched for each metabolic pathway between previously defined clusters. Furthermore, the method enables rediscovering stratification or simply assigning new patients to individual clusters. In fact, the method determines genes that exhibit expression explaining the clustering and thus can find the cluster of a patient based only on the expression of these genes for that patient. Therefore, the method enables selecting intestinal microbial genes for stratifying patients into various homogeneous subgroups.

[0022] The method is implemented on a computer. This means that the steps (or substantially all steps) of the method are performed by at least one computer or any similar system. Thus, the steps of the method are performed by a computer, perhaps fully automatically or semi-automatically. In an example, the initiation of at least some of the steps of the method may be performed through interaction between the user and the computer. The required level of user-computer interaction may depend on the level of automation that is foreseen and set in balance with the need to fulfill the user's wishes. In an example, this level may be user-defined and / or pre-defined.

[0023] A typical example of the computer implementation of the method is to execute the method on a system adapted for this purpose. The system may comprise a processor and a processor coupled to a memory and a graphical user interface (GUI), and the memory has recorded thereon a computer program containing instructions for executing the method. The memory may also store a database. The memory is any hardware adapted for such storage and may perhaps include a plurality of physically separate parts (for example, for the program and perhaps for the database).

[0024] A dataset containing microbial gene expression data, the resulting set of clusters, and / or determined genes exhibiting expression that explain the clustering may be stored in a database. “Database” means any collection of data (i.e., information) organized for searching and retrieval (e.g., a relational database, e.g., one based on a given structured language, e.g., SQL). Once stored in memory, a database enables rapid searching and retrieval by a computer. Databases are, in fact, structured to facilitate saving, retrieving, modifying, and deleting data in conjunction with various data processing operations. A database may consist of a file or a set of files that can be broken down into records consisting of one or more fields. A field is the basic unit of data storage. Users can retrieve data primarily through queries. Using keywords and sort commands, users can quickly search, rearrange, group, and select fields within many records according to the rules of the database management system being used to obtain or create reports on a particular set of data.

[0025] This method is for processing genetic data, including the processing of microbial gene expression data. This includes the analysis and interpretation of information found in genetic data, such as DNA sequences, mutations, genetic variations, and other elements related to genes and genomes. Genetic data processing can be used in genomic research, personalized medicine, medical genetics, and various other scientific and medical fields. For example, the method may be included in a medical process, which may include using the determined patient clusters after the method has been performed to help healthcare professionals, such as physicians and nurses, treat and / or cure patients' diseases and / or infections. Specifically, the method may include comparing patient information with data from other patients in the same cluster.

[0026] For example, the method of use may involve determining the closest patient (the patient in the cluster that is most similar to the patient under study from a genetic standpoint) from among all other patients in the determined cluster, and then estimating information about the patient under study based on the similarity to the determined closest patient. By analyzing the similarity to the closest patient, the method of use may involve inferring additional information about the patient under study, such as possible disease progression, response to treatment, or potential risk factors. This comparative approach enables more individualized and informed treatment planning by leveraging the medical history and outcomes of similar patients. For example, if the closest patient in the cluster responded well to a particular treatment, the same treatment may be considered for the patient under study. The method of use makes it possible to find such information and suggest it to healthcare professionals to support their practice.

[0027] The distance between patients within the same cluster can be calculated in various ways. For example, it can be determined based on each patient's metagenomic data. Alternatively, if clinical data is available, it may be calculated from clinical data. Generally, Euclidean distance can be used, but Minkowski distance may also be used. This method may also use the graph and the resulting clustering to calculate geodesic distance (the shortest distance expressed as the number of edges between two nodes).

[0028] For example, the closest patient can be determined using the Bray-Curtis dissimilarity index. The Bray-Curtis dissimilarity index can be calculated using the following formula:

number

[0029] The method of use may then include outputting information about the nearest determined patient. Each other patient in the cluster may have the same disease and / or infection, and the output information may show the stage of the disease and / or infection, the treatment given to them, and / or their response to that treatment for one or more nearest patients. The output information can be used to infer similar information about that patient (e.g., to predict the stage of the disease and / or infection, and / or their response to a given treatment). By allowing patients to be compared only with a subset of patients in the database (i.e., patients in the same cluster), the method improves the inference of patient information from the database. In particular, since the clusters are based on microbial gene expression data, the method allows patients to be compared with patients who share the same microbial gene expression, thereby allowing this to be taken into account during inference.

[0030] Furthermore, this method improves the estimation of patient information from patient databases, thereby improving how healthcare professionals assist in treating and / or curing patients, and ultimately improving the treatment of patients' diseases and / or infections themselves. In particular, the method allows for targeting genes that exhibit expression that influences the outcome. In fact, microbial gene expression data obtained for patients may include only the expression of genes identified as showing significantly different expression between clusters. Thus, the method also allows for a reduction in the number and complexity of genes that need to be searched within patient data.

[0031] Next, we will explain the acquisition of the dataset (S10).

[0032] The dataset includes microbial gene expression data from multiple patients. For example, the microbial gene expression data may contain information about the expression of microbial genes in each of several patients. Gene expression can refer to information about the process by which the information contained in a gene is used to produce functional products (usually proteins, but sometimes non-coding RNA). This process can be fundamental to the function of a cell and the expression of its genetic traits. Microbial gene expression data can be extracted from environmental samples, such as the fecal microbiota of the human body. Microbial gene expression data from multiple patients in feces can characterize the gut microbiota of those patients, i.e., all the microorganisms inhabiting the intestines of multiple patients. Microbial gene expression data makes it possible to identify, for each patient, the bacterial species present in that patient's gut microbiota, their relative abundance, and the functions they can perform.

[0033] Acquisition S10 may include determining a dataset containing microbial gene expression data for multiple patients. The determination of a dataset containing microbial gene expression for multiple patients can be performed in any way. For example, determining the dataset may include extracting microbial gene expression data for multiple patients from an actual environmental sample (e.g., a human fecal microbiota). Alternatively, the microbial gene expression data may have already been extracted at the time the method is performed. For example, the microbial gene expression data may be recorded in memory. In that case, acquisition S10 may include extracting the recorded microbial gene expression data for multiple patients from this memory. Yet another method is that the memory may store microbial gene expression data for only some of the multiple patients. In that case, acquisition S10 may include extracting the microbial gene expression data for these some patients from memory and extracting the microbial gene expression data for the remaining patients from an actual environmental sample.

[0034] Next, we will explain clustering S20 for multiple patients.

[0035] Clustering of multiple patients S20 can be performed in any way. For example, clustering of multiple patients S20 can be performed based on any known algorithm. Clustering S20 may include determining a set of clusters and assigning each patient to one of the clusters in that set. Clustering S20 may consider the patients' microbial gene expression data to form clusters and group patients with similar microbial gene expression data into the same cluster. The number of clusters may be predetermined or may be inferred during clustering S20. This inference may aim to optimize one or more criteria, such as maximizing the similarity of microbial gene expression data within the same cluster, maximizing the distance between patients in different clusters, and / or minimizing the total number of resulting clusters.

[0036] In the embodiment, the clustering process S20 may include three steps: S21, which involves performing multiple iterations of the Latent Dirichlet Allocation (LDA); S22, which involves determining a graph; and S23, which involves performing graph-based clustering. Performing clustering S20 based on these three steps improves patient clustering by enabling the maximization of similarity of microbial gene expression data within the same cluster, the maximization of distance between patients within different clusters, and the minimization of the total number of resulting clusters. Each of these three steps, S21, S22, and S23, is described in more detail in the following paragraphs.

[0037] The first step S21 consists of performing multiple iterations of the Latent Dirichlet Allocation (LDA) (each iteration yields its own Latent Dirichlet Allocation (LDA)). The Latent Dirichlet Allocation (LDA) is a process that assigns each patient to one of a given set of topics. Each iteration of the Latent Dirichlet Allocation (LDA) provides a patient's respective assignment to a given set of topics. Multiple iterations can provide different assignments for multiple patients (in terms of the number of topics considered in each iteration and the resulting distribution of patients within the topics considered in each iteration). The number of topics may be a hyperparameter optimized using, for example, a grid search, to enable obtaining the desired best performance. For example, the method may consider a number of topics ranging from 3 (the minimum number of latent topics) to n latent topics (where n is a positive integer).

[0038] Next, the second step S22 consists of determining a graph representing different assignments obtained from multiple iterations of the Latent Dirichlet Allocation (LDA). Determining the graph may include, for each pair of patients, counting the number of times the two patients in that pair were found in the same topic during different iterations of the Latent Dirichlet Allocation (LDA) (each time the two patients are found in the same topic, there are two appearances of the two patients in the same topic). The counts obtained for each pair of patients may be represented in a matrix. Next, determining the graph may include arranging the nodes representing the patients such that the distance between each node pair (i.e., the length of the edge between the two nodes) is approximately proportional to the number of appearances counted for those two nodes.

[0039] Count-based graph determination can be performed using known algorithms such as the Fruchterman-Reingold force-oriented algorithm. For example, graph determination may involve an initial step of randomly projecting all nodes into space so that they do not overlap. The algorithm may treat nodes as repelling objects, while edges may be considered springs that keep the nodes connected. Graph determination can then perform an iterative process, which involves calculating the attractive and repulsive forces between each node pair in each iteration and adjusting their positions accordingly until convergence is achieved (the cost of weighting deviations from proportionality is minimized), meaning the nodes no longer move in space. For graphical purposes, when the graph is displayed, the thickness of each edge may also depend on its frequency of appearance (the shorter and thicker the line, the more frequently it appears).

[0040] Next, the third step S23 consists of performing graph-based clustering of the determined graph. Graph-based clustering may be performed based on an algorithm configured to divide the graph into subcommittees of nodes (i.e., subcommittees corresponding to the resulting clusters) in order to maximize modularity. The algorithm may take as input a computed matrix representing the counts obtained for each pair of patients. Modularity may be quantified by the edge density within a community and the edge density between different communities. The algorithm may be, for example, the Louvain clustering algorithm or the Leiden clustering algorithm. The algorithm may perform the following steps: The first step may include randomly assigning each node to a community. The second step may include moving nodes within each community to optimize modularity (if the modularity increases after moving a node, the algorithm may assign that node to a new community). The third step may include merging neighboring communities (to create larger, more homogeneous communities). The fourth step may be a refining step. The refining process may involve moving nodes between newly merged communities and calculating modularity (to optimize it again).

[0041] The resulting clustering can be used in many ways. For example, one might be to stratify patients from metagenomic data. However, it can also be used to stratify patients from other types of data that follow probability distributions such as the Dirichlet distribution.

[0042] Next, we will explain S50, the determination of genes whose expression explains clustering. Genes whose expression explains clustering will be referred to as "explanatory genes" below.

[0043] The descriptive gene determination S50 includes the identification S52 of genes that show significantly different expression between clusters. The genes identified in this step S52 are the determined descriptive genes. In step S52, the genes may be identified from a database that references genes, genomes, and / or biological pathways, such as a known database like the KEGG (Kyoto Encyclopedia of Genes and Genomes) database. As an example, the descriptive gene determination S50 may also include the selection S51 of genes included in a set of metabolic pathways prior to this identification step S52. In that case, the genes are identified in step S52 from (i.e., only from) the genes selected in step S51. The genes may be identified in step S52 from all genes referenced in the aforementioned (e.g., known) database. These two steps S51 and S52 are described in more detail in the following paragraphs.

[0044] In step S51, the gene may be selected from all genes referenced in the (e.g., known) database. For example, the database may list all metabolic pathways (including those within a set of metabolic pathways) and include information for each gene indicating which metabolic pathway it belongs to. The selection of genes included in the set of metabolic pathways S51 is performed based on this information and may include extracting genes that the information indicates are included in at least one of the metabolic pathways in the set. For example, selection S51 may include scanning all genes in the database and retaining only those genes that are indicated to be involved in at least one of the metabolic pathways in the set.

[0045] Identifying genes that show significantly different expression between clusters (S52) may involve two steps. In the first step, identification (S52) may involve comparing the expression of each gene between each cluster. This comparison of gene expression may be performed based on the Kruskal-Wallis test. This is a nonparametric test based on rank assignment that identifies significant differences in terms of gene expression between clusters. The Kruskal-Wallis test is particularly suitable for comparing gene expression between clusters because the data do not follow a known distribution (such as a Gaussian distribution). The Kruskal-Wallis test may involve the following steps: The first step involves joining the patient's metagenomic data and sorting them in ascending order. The second step involves assigning ranks to the data. The third step involves calculating the Kruskal-Wallis statistic. The fourth step involves comparing the calculated Kruskal-Wallis statistic to a chi-squared distribution to determine the respective p-values. These steps may be repeated for each gene in the dataset, thereby calculating the respective p-value for each gene.

[0046] In the second step, identification S52 may include retaining genes that have significantly different expression levels between clusters based on the comparison results. For example, the comparison of gene expression may involve calculating the respective p-values ​​for each gene, and the method may retain only the genes whose respective p-values ​​are below a threshold (e.g., 1%). The identified explanatory genes may be those retained during this step.

[0047] A metabolic pathway is a series of interconnected chemical reactions that occur within a cell. The set of metabolic pathways considered in this method may represent only a portion of all known metabolic pathways (for example, those referenced in the aforementioned database).

[0048] In the embodiments, the method may include a step S40 of selecting a set of metabolic pathways. The selection of the set of metabolic pathways S40 may depend on the disease or infection under study. Selection S40 is performed by extracting the metabolic pathways that are most significantly enriched in terms of genes from an existing database of metabolic pathways. Metabolic pathways may be referenced in known databases (e.g., KEGG). In such databases, each metabolic pathway may be associated with a list of genes or a list of metabolites. However, one gene may also be associated with multiple different pathways. A metabolic pathway is said to be enriched if it contains multiple genes.

[0049] The selection of a set of metabolic pathways S40 may be performed by conducting a metabolic pathway enrichment analysis to detect various enriched pathways from among all metabolic pathways referenced in the database, for example. The selected set of metabolic pathways may include the pathways that are most significantly enriched in terms of genes, according to this enrichment analysis. For example, the selected set may include a predetermined number of most enriched pathways, or it may include metabolic pathways with enrichment levels above a threshold. The selection of a set of metabolic pathways S40 may be performed based on the enrichment p-value results of Fisher's exact test, which can be used to calculate the respective p-values ​​for each metabolic pathway. The metabolic pathways selected for the set may have p-values ​​below a threshold (e.g., 5%).

[0050] In the embodiments, the method may include gene filtering S30. The filtering S30 step may be performed prior to the selection of a set of metabolic pathways (step S40) and the determination of explanatory genes (step S50). In this case, the explanatory genes are determined from the genes remaining after filtering. For example, filtering S30 may include assigning a value to each gene first referenced in the database, and filtering S30 may include retaining only genes with a value higher than a threshold. This value may represent the relative expression (i.e., its normalized expression) of the gene. The relative expression (or relative abundance) of a gene may be calculated as described in the 2017 paper "The gut microbiome in atherosclerotic cardiovascular disease" by Jie et al., Nat Commun 8,845, https: / / doi.org / 10.1038 / s41467-017-00900-1. In the first step, the data is sequenced, and the resulting reads can then be aligned with a reference catalog containing gene sequences (gene counting). One of the identification criteria here may include alignments greater than 35% and scores greater than 60. In the second step, the relative abundance of a gene can be calculated by dividing the raw abundance of that gene by the total abundance of all reads in the sample. In the third step, the relative abundance of a knockout (KO) can be calculated by dividing the raw abundance of that KO by the total abundance of all genes in the sample. This value is a hyperparameter and may have been measured during the extraction of genes from the environmental sample. In this case, the threshold used for filtering may be less than 1E-3 and / or greater than 1E-5, for example, substantially equal to 5E-4. Filtering S30 improves the efficiency and accuracy of data processing and the detection of important genes. In fact, filtering S30 allows the algorithm to focus only on a subset of relevant genes from the database.

[0051] If the method includes a filtering step S30, genes can be identified in step S52 from the genes remaining after filtering (i.e., genes with values ​​higher than the threshold). If the method includes both a filtering step S30 and a selection step S51, genes can be selected in step S51 from the genes remaining after filtering (i.e., genes with values ​​higher than the threshold).

[0052] Next, an example of implementing this method will be presented with reference to Figures 2 to 6.

[0053] Cardiac metabolic diseases represent a significant global health challenge, necessitating a deeper understanding of disease progression and patient trajectories. The emergence of high-throughput technologies has generated vast amounts of heterogeneous and complex data reflecting the unique characteristics of each patient. To effectively extract meaningful insights from this vast treasure trove of information, this method can be employed to stratify them. Patient stratification is the process of grouping patient populations into clinically or molecularly defined subgroups based on identifiable characteristics related to disease course, prognosis, or treatment response.

[0054] One of the main challenges addressed by this method is effectively stratifying patients in the context of metabolic diseases using high-dimensional datasets (metagenomics), and then reducing dimensionality to extract valuable insights. Specifically, this method implements a computational framework that first stratifies individuals into homogeneous subpopulations. Subsequently, it identifies microbial genes that show significantly different expression levels among these groups and hypothesizes that these metagenomic genes can function as biomarkers. To test this hypothesis, this method may involve applying classification-based algorithms to the data to successfully reclassify individuals into their respective groups. Thus, the previously identified genes can be used as biomarkers for stratification.

[0055] The implementation of this method focuses on the analysis of metagenomics-related data from cardiovascular disease patients. It demonstrates the ability to stratify patients into distinct groups based on their gut microbiota sequencing data, cluster them, and identify specific bacterial genes that distinguish them and can function as biomarkers for stratification. Following an investigation of metabolic pathway enrichment across the entire dataset, the method may include performing a statistical assessment of metagenomic gene expression for each enriched pathway within each patient cluster. These genes can then be used in a patient classification task, where a significantly higher ability may be observed in accurately classifying patients into previously defined clusters compared to classification by clinical phenotype. Thus, the method can be used to effectively stratify individuals and yield a set of genes suitable for use as biomarkers for potential therapies and disease classification states.

[0056] "Omics" refers to the process of capturing the whole picture of various biomolecules (DNA, RNA, proteins, etc.) within a cell, organism, or population. These are large datasets that present various technical challenges, the primary challenge being the extraction of information that has both mathematical relevance and biological significance. One of the main technical problems that this method will solve is stratifying patients into biologically homogeneous groups and identifying clusters of patients who share similarities within the data, which may reflect the patient's state more accurately than clinical status defined based on clinical biomarkers. Subsequently, it is necessary to understand the differences between patients who share the same "omics" that lead to this stratification and identify potential distinguishing biomarkers between groups, which may be potentially viable and targetable in therapeutic strategies.

[0057] This method solves the problem by implementing a framework that combines patient stratification, identification of enriched signaling pathways (pathway analysis), and methodologies for selecting discriminative bacterial genes between patient groups established by stratification. Furthermore, this method may further include demonstrating that these genes can be used to classify specific patients, enabling them to be classified into previously established subgroups by stratification with greater precision than clinical groups.

[0058] One of the main advantages of this method is that it first allows patients to be stratified into a network using existing methodologies. Then, by applying clustering to this network, the method allows for the identification of groups of patients with similarities. Finally, by comparing enriched genes for each pathway across previously defined clusters, the method allows for the isolation of genes that are significantly expressed across groups and confirms that these genes actually facilitate patient stratification within these clusters.

[0059] Therefore, by using only this gene pool, it is possible to rediscover stratification or simply assign new patients to individual clusters. In conclusion, this method enables the selection of gut microbiota genes for stratifying patients into various homogeneous subgroups.

[0060] Here are some terms to explain. KEGG (Kyoto Encyclopedia of Genes and Genomes) is a collection of databases that store information about genes, genomes, and biological pathways. KO (KEGG Ortholog) is a group of genes considered functionally equivalent across different organisms. This means that genes within a KO group may have different DNA sequences, but all perform the same basic function.

[0061] This method identifies specific biomarkers that enable patient stratification. A biomarker is defined as a measurable indicator of a biological state or condition and is used for diagnosis, monitoring of health status, or evaluation of response to treatment. In the first step, using data available from a database referencing a first set of genes (e.g., a set of 7125 KEGG orthologue genes), the method includes a step S30 in which genes whose relative expression exceeds a threshold (e.g., 5E-4). The filtering step S30 may include applying a metabolic scope defined as the threshold (values ​​below this threshold may be due to noise caused by experimental conditions or detection limits). This threshold is set to 5E-4, so that all genes with relative expression lower than this value can be excluded. Thus, filtering S30 makes it possible to obtain a second set of genes (e.g., a subset of 922 genes), which is a subset of the first set of genes, and which is fewer in number than the first set, and can be used and analyzed in the stratification shown in Figure 2.

[0062] Next, in step S20, the method includes a step of applying Latent Dirichlet Allocation (LDA) S21, which makes it possible to simultaneously assign a probability of belonging to a topic to each patient and similarly to a probability of belonging to a topic to each gene. At the end of various allocation runs, the method includes a step S21' in which both patients and genes are assigned to the topic with the highest probability. This process is repeated multiple times, thereby calculating the number of times each patient is associated with the same topic as other patients S22, and subsequently, it becomes possible to represent the patients in graph form (using a representation based on the Fruchterman-Reingold force-oriented algorithm).

[0063] Figure 3 shows two graphical representations of the determined graph of patients stratified based on metagenomics: a first representation 101 in which patients are highlighted based on their clinical status, and a second representation 103 in which patients are highlighted based on their clustering assignment (Louvain cluster). The resulting matrix is ​​then clustered using an algorithm called Louvain clustering based on community detection S23. Thus, patients are clustered into homogeneous subgroups in terms of metagenomic gene expression (see example of the resulting graph in Figure 3).

[0064] In addition to this, in parallel with a second set of 922 gene knockouts, the method includes a step of performing metabolic pathway enrichment analysis S40 to detect various enriched pathways. Pathway enrichment is a computational process that identifies biological functions that are statistically overexpressed in a set of genes more than would be expected by chance, as described in the paper Chicco, D. and Agapito, G., 2022, “Nine quick tips for pathway enrichment analysis”, PLoS Comput Biol 18, e1010348, https: / / doi.org / 10.1371 / journal.pcbi.1010348.

[0065] Of the 922 genes, 515 could be recognized as being included in existing metabolic pathways in the KEGG database (see Figure 2), resulting in a third set of 515 genes. This method involves extracting these metabolic pathways and selecting only those that are most significantly enriched from a genetic standpoint S40 (adjusted p-value < 0.05, where p-value quantifies the strength of evidence for the null hypothesis by reflecting the probability of observing such an extreme result if the null hypothesis were true). In this example, this yields a set of 49 significantly enriched metabolic pathways (see Table 1 below). [Table 1]

[0066] For each of these enrichment pathways, the method may include the step of extracting all genes contained therein (S51) and comparing the expression of each gene among the previously established clusters. The method then retains only the genes that have significantly different expression among the clusters (adjusted p-value < 0.01, using the Kruskal-Wallis test).

[0067] In this implementation example, the method then includes the step of calculating the Bray-Curtis dissimilarity for all patients to evaluate the hypothesis used. There is a clustering tendency among patients within each cluster. Notably, when patients are sorted according to their cluster assignment, a consistent pattern emerges. Figure 4 shows this distance-based analysis using the Bray-Curtis metric for a set of 129 identified and significantly expressed genes. Individuals within the same cluster have small distances from each other (indicating proximity) and therefore show similarity in their data. Thus, patients sharing the same cluster show similarity, at least in their metagenomic characteristics, regardless of their clinical phenotype.

[0068] The utility of identified genes in interpersonal identification is demonstrated by using a random forest approach that utilizes a previously identified set of genes as input. Predictive outputs include either patient-assigned clusters determined by prior stratification, or individual clinical classifications.

[0069] The dataset is split into training and test subsets, with 80% allocated to training and 20% to testing. Subsequently, the parameters of the random forest model are fine-tuned. The accuracy of classifying patients within each cluster is at a high level (average accuracy over 20 iterations: 0.86). A visual representation of the classification performance in the form of a confusion matrix 201 between predicted patient clusters and actual patient clusters is presented in Figure 5.

[0070] Simultaneously, the accuracy of classifying patients into different clinical phenotypic groups decreases (average precision over 20 iterations: 0.43). Another confusion matrix of predicted clinical phenotypes versus true phenotypes, using a random forest for 129 genes,203, illustrates the classification results of patients' clinical conditions and is shown in Figure 5.

[0071] These results demonstrate that, using the previously established stratification and gene selection framework of this method (approximately 129 identified genes) involving over 900 genes, it is possible to reclassify patients into homogeneous genomic subpopulations. Therefore, these marker genes can be defined as biomarkers for patient stratification.

[0072] This method can be applied across a variety of contexts, including diseases related to the gut microbiota, such as inflammatory bowel disease (IBD), ulcerative colitis, Clostridium difficile infection, and colorectal cancer, for which relevant data are available.

[0073] Figure 6 shows an example of a system, where the system is a client computer system, such as a user's workstation.

[0074] The client computer in this example comprises a central processing unit (CPU) 1010 connected to an internal communication bus 1000, and random access memory (RAM) 1070 also connected to the bus. The client computer further comprises a graphics processing unit (GPU) 1110 associated with video random access memory 1100 connected to the bus. The video RAM 1100 is also known in the art as a frame buffer. A mass storage controller 1020 manages access to mass storage devices such as a hard drive 1030. Mass storage devices suitable for tangibly executing computer program instructions and data include, for example, semiconductor memory devices such as EPROMs, EEPROMs, and flash memory devices; magnetic disks such as internal hard disks and removable disks; and all forms of non-volatile memory, including magneto-optical disks. Any of the above may be supplemented or incorporated by specially designed application-specific integrated circuits (ASICs). A network adapter 1050 manages access to the network 1060. The client computer may also include haptic devices 1090 such as a cursor control device and a keyboard. A cursor control device is used in a client computer to allow the user to selectively position the cursor at any desired location on the display 1080. In addition, the cursor control device allows the user to select various commands and input control signals. The cursor control device includes a number of signal generating devices for input control signals to the system. Typically, the cursor control device may be a mouse, with the mouse buttons used to generate signals. Alternatively or additionally, the client computer system may be equipped with a pressure-sensitive pad and / or a pressure-sensitive screen.

[0075] A computer program may comprise instructions executable by a computer, the instructions including means for causing the system to execute the Method. The program may be recordable on any data storage medium, including the system's memory. The program may be implemented, for example, in digital electronic circuits, or in computer hardware, firmware, software, or a combination thereof. The program may be implemented as a device, for example, as a product tangibly implemented in a machine-readable storage device for execution by a programmable processor. The steps of the Method may be executed by a programmable processor that executes a program of instructions for performing the functions of the Method by manipulating input data and generating outputs. The processor may therefore be programmable and coupled to receive data and instructions from a data storage system, at least one input device, and at least one output device, and to transmit data and instructions to them. The application program may be implemented in a high-level procedural or object-oriented programming language, or, if desired, in assembly language or machine language. In either case, the language may be a compiled or interpreted language. The program may be a complete installation program or an update program. In any case, the application of the program on the system results in instructions for executing the Method. Alternatively, the computer program may be stored and executed on a server in a cloud computing environment, which communicates with one or more clients over a network. In such a case, the processing unit executes the instructions contained in the program, thereby executing this method on the cloud computing environment.

Claims

1. A computer implementation method for processing genetic data, The process involves obtaining a dataset containing microbial gene expression data from multiple patients (S10); The process of clustering the multiple patients into a set of clusters based on the microbial gene expression data (S20); The steps include: selecting genes included in a set of metabolic pathways (S51) and / or identifying genes that show significantly different expression among the clusters (S52) to determine genes that exhibit expression explaining the clustering (S50); A method of having.

2. The aforementioned clustering (S20) is The process involves performing multiple iterations of the Latent Dirichlet Allocation Method (LDA) (S21), wherein each iteration assigns each patient to one of several topics; A step (S22) of determining a graph including nodes representing the plurality of patients and edges connecting pairs of nodes, wherein the length of each edge connecting pairs of nodes is a function of the number of times the patients represented by the connected nodes appeared in the same topic over multiple iterations of the latent Dirichlet allocation method (LDA); The steps include: (S23) performing graph-based clustering of the determined graph; The method according to claim 1, comprising:

3. The method according to claim 2, wherein the graph is determined based on the Fruchterman-Reingold force-oriented algorithm.

4. The method according to claim 2 or 3, wherein the graph-based clustering is performed based on the Louvain clustering algorithm.

5. The method according to any one of claims 1 to 4, wherein the identification of genes showing significantly different expression among the clusters (S52) comprises the steps of: comparing the expression of each gene among the clusters; and, based on the results of the comparison, retaining the genes that have significantly different expression among the clusters.

6. The method according to claim 5, wherein the comparison of gene expression is performed based on the Kruskal-Wallis test, which calculates the p-value for each gene, and the retained gene optionally has a p-value of less than 1%.

7. The method according to any one of claims 1 to 6, further comprising the step of selecting a set of metabolic pathways by extracting the metabolic pathway that is most significantly enriched from a database of existing metabolic pathways in terms of genes (S40).

8. The method according to claim 7, wherein the selection of the set of metabolic pathways (S40) is performed based on Fisher's exact test, which calculates the respective p-value for each metabolic pathway, and the selected set of metabolic pathways optionally has a p-value of less than 5%.

9. The method according to any one of claims 1 to 8, further comprising the step of filtering out genes whose relative expression exceeds a threshold from a gene database (S30), wherein the gene is determined from the genes remaining after the filtering.

10. A method for implementing the use of clustering described in claim 1 on a computer, A step of obtaining microbial gene expression data of a patient, wherein the microbial gene expression data includes the expression of a gene determined by the method of claim 1; A step of determining the cluster to which the patient belongs based on the gene expression determined for the patient. A method of having.

11. The method according to claim 10, further comprising the step of determining the closest patient from among the patients of the determined cluster.

12. The method according to claim 11, wherein the nearest patient is determined using the Bray Curtis dissimilarity index.

13. A computer program having instructions, when executed by a computer, to cause the computer to perform the method according to any one of claims 1 to 9 and / or the method according to any one of claims 10 to 12.

14. A computer-readable storage medium on which the computer program described in claim 13 is recorded.

15. A system comprising a processor and a memory coupled to the processor, wherein the computer program described in claim 13 is recorded in the memory.

Images