A protein clustering method combining information granulation and evidential reasoning

Abstract

The application provides a protein clustering method combining information granulation and evidence reasoning, which comprises the following steps: granulating protein sequence data in cells, fusing information granules of the protein sequence data based on intersection relationship, fusing granule clusters of the protein sequence data based on density propagation, fusing granule groups of the protein sequence data based on distance, calculating evidence values of stable samples, calculating evidence values of unstable samples, assigning categories of the protein sequence samples according to the evidence values, and improving the accuracy of protein subcellular localization prediction, so as to facilitate biologists to judge or design specific proteins in each subcellular structure or to predict the function of the proteins.

Description

Technical Field

[0001] This invention relates to the field of proteomics in bioinformatics, specifically to a clustering analysis method that combines information particle fusion and evidence reasoning. Background Technology

[0002] Subcellular protein localization is a crucial area of ​​research in proteomics. Subcellular structures are more refined than the cell itself, typically observed using an electron microscope. They are characterized by their location within the cell, diverse functions, spatial isolation, yet coordinated efforts to maintain cellular integrity. In eukaryotic cells, for example, subcellular structures can be categorized into 11 types: cytoskeleton, cytoplasm, endoplasmic reticulum, endosomes, extracellular space, Golgi apparatus, mitochondria, nucleus, peroxisomes, cell membrane, and vacuoles. Each subcellular structure provides a relatively independent site for the specific proteins within it, enabling them to perform their functions. Subcellular protein localization refers to the specific location of a protein within the cell—that is, within which subcellular structure the protein resides. Since proteins can only function and be metabolized within their specific subcellular structures, any deviation in protein localization can negatively impact cellular function. Therefore, studying subcellular protein localization is of significant importance. Because proteins located in the same subcellular structure often have similar structures and functions, proteins with similar functions can be clustered together by analyzing their structures, thereby predicting the subcellular location of proteins. This provides technical support for subsequent identification or design of specific proteins in corresponding subcellular structures or for predicting the function of proteins.

[0003] Clustering analysis is an unsupervised machine learning technique that can effectively group massive amounts of data without prior knowledge, facilitating efficient information mining through classification or hierarchical analysis. The k-means algorithm, as the most commonly used clustering algorithm, is widely applied in proteomics research due to its ease of understanding and speed. However, it cannot handle non-convex data and is highly sensitive to outliers, thus performing poorly in protein subcellular localization prediction and resulting in poor evaluation. Furthermore, density-based clustering algorithms DBSCAN and DPC can handle non-convex data and have been repeatedly applied in proteomics research, but they cannot handle data with large density differences between clusters, thus also performing poorly in protein subcellular localization prediction and resulting in poor evaluation. Summary of the Invention

[0004] To address the aforementioned deficiencies in existing technologies, this invention proposes a protein subcellular localization prediction algorithm based on density clustering analysis combining information particle fusion and evidence reasoning. This algorithm is then used to perform clustering analysis on protein sequence data of a certain cell type to predict protein subcellular localization, thereby improving the accuracy of protein subcellular localization prediction. This facilitates bioinformaticians in identifying or designing specific proteins in various subcellular structures, or in predicting the functions of proteins.

[0005] According to a first aspect of the present invention, a protein clustering method combining information particle fusion and evidence reasoning is provided, comprising:

[0006] Step 1: Granulate the protein sequence data in the cell, based on the preset parameter k and the sparsity sd(x) of each protein sequence sample. i The protein sequence dataset U is granulated to generate a set of g information granules {G1, G2, ..., G...}. g}

[0007] Wherein, the protein sequence dataset U = {x1, x2, ..., x} n It consists of n protein sequence samples, each protein sequence sample x i Contains w attribute information {x i1 ,x i2 ,…,x iw}

[0008] k takes the value of This represents the rounding up of a non-integer value, where c is the given number of clusters. A cluster represents a subcellular structure type, and the number of clusters represents the number of subcellular structure types.

[0009] Sparsity of each protein sequence sample It is an information granularity that simultaneously measures the global and local density of a protein sequence sample, where, To measure the global density of protein sequence samples, the radius of the neighborhood that maximizes the relative density of the protein sequence samples is used. Let |·| represent the relative density of each protein sequence sample under different neighborhood radii, |·| represent the cardinality of the set, and δ(x) represent the neighborhood of each protein sequence sample under different neighborhood radii. i ,d ij )={x z |x z ∈U,d(x i ,x z )≤d ij Protein sequence sample x i and x j The Euclidean distance d(x) between them i ,xj )=d ij =‖x i -x j ‖2, i,j=1,2,…,n and i≠j, The k-nearest neighbor radius of a protein sequence sample measures the local density of the protein sequence sample. Represents protein sequence sample x i The k-th nearest neighbor.

[0010] For each information particle G, there exists a protein sequence sample x. i , making And |G|=k, and, That is, the information particle G is a particle with x i Centered on A hypersphere with radius x contains k protein sequence samples, and the center x... i Its sparsity is minimal.

[0011] Step 2: Fusion of information particles from protein sequence data based on intersection relationships. If any two information particles intersect |G i ∩G j If |≥1, meaning the two information particles share a common protein sequence sample, then these two information particles are fused. After each pair of information particles is fused, a set of information particles {G1,G2,…,G...} is formed. g} Transform into a set of g * Individual clusters

[0012] Step 3: Perform cluster fusion of protein sequence data based on density propagation, targeting any cluster for GC. a and the nearest GC cluster b If sd * (GC a )≥sd * (GC b ), i.e., GC b The density is greater than that of GC a If the density is such that the two clusters are fused, and after each cluster is fully fused, a group of clusters is formed. Transformed into a set of g′ particle groups {GF1,GF2,…,GF g′}

[0013] Wherein, the distance d between any two clusters * (GC a GC b )=min{d ij |x i ∈GC a ,x j ∈GCb}, that is, the distance between the closest pair of protein sequence samples in two grain clusters is used as the distance between grain clusters, where a, b = 1, …, g * and a ≠ b, the sparsity of the grain cluster That is, the average value of the sparsity of protein sequence samples within a grain cluster represents the density of the grain cluster.

[0014] Step 4: Perform swarm fusion of protein sequence data based on distance. For a given number of clusters c, if c = g′ or c = g * then execute Step 4.1, if c < g′ then execute Step 4.2, and if c > g′ then execute Step 4.3.

[0015] Among them, the distance d between any two swarms * (GF a , GF b ) = min{d ij ∣x i ∈ GF a , x j ∈ GF b}}, that is, the distance between the closest pair of protein sequence samples in two swarms is used as the distance between the swarms.

[0016] Step 4.1: Take the g′ swarms {GF1, GF2, …, GF g′} or g * grain clusters directly as the initial clusters {Cl1, Cl2, …, Cl c}.

[0017] Step 4.2: Fuse the pair of swarms with the closest distance, update the number of swarms and the distance between swarms, and execute Step 4 again.

[0018] Step 4.3: Treat each swarm as a sub - data set, execute Steps 1 to Step 3 again for each sub - data set, combine all the swarms obtained into a new set of swarms, where all protein sequence samples not added to the swarms are regarded as a swarm, and execute Step 4 again.

[0019] Step 5: Calculate the evidence value of stable samples. Take the protein sequence samples within the initial cluster Ω = {Cl1, Cl2, …, Cl c} as stable samples, and calculate the evidence values of stable samples belonging to each single cluster and the cluster全集 (the whole set of clusters) where u = 1, 2, …, c, A ∈ 2 Ω \{Cl1, Cl2, …, Cl c , Ω}, initialize the set S, and add all stable samples to the set S.

[0020] Step 6: Calculate the evidence value for unstable samples, specifically the protein sequence sample x with the smallest sparsity that does not belong to set S. i Based on protein sequence samples x in set S j Calculate x i Evidence values ​​belonging to each individual cluster and the entire cluster set: in x represents i In set S, the evidence values ​​provided by the k nearest neighbors are merged using the Dempster merging rule formula. Where “·” represents any Cl u Or Ω, x i Add to set S, and repeat step 6 until S = U.

[0021] Step 7: Assign the protein sequence sample to a category based on the evidence value. For each protein sequence sample x i Assign it to cluster The final clusters are the clustering results. If, for a given outlier threshold τ, Then x i These are outlier samples.

[0022] Furthermore, the protein clustering method combining information particle fusion and evidence reasoning provided by the present invention is characterized in that the outlier threshold τ is in the range of [0.99, 1].

[0023] According to a second aspect of the present invention, a computer device is provided, characterized in that it comprises:

[0024] Memory, used to store instructions; and

[0025] A processor is configured to invoke instructions stored in the memory to execute a protein clustering method combining information particle fusion and evidence reasoning, which is a first aspect.

[0026] According to a third aspect of the present invention, a computer-readable storage medium is provided, characterized in that it stores instructions that, when executed by a processor, perform a protein clustering method combining information particle fusion and evidence reasoning of the first aspect.

[0027] Compared with the prior art, the above-mentioned technical solution conceived in this invention has at least the following beneficial effects:

[0028] (1) The newly proposed sparsity metric formula can simultaneously measure the global density and local density of a sample.

[0029] (2) The sparsity-based information granulation method can generate information granules not only in high-density regions but also in low-density regions. Therefore, it can handle data with large density differences between clusters.

[0030] (3) Based on three newly designed information particle fusion strategies, by fusing information particles and forming an initial cluster structure, the situation where the preset cluster center or density peak has a decisive influence on the clustering results can be avoided, the problem of finding the cluster center is difficult and the rationality is low can be overcome, and the limitations brought about by the traditional information particle convex structure can be broken. Therefore, it can process data with no geometrically meaningful cluster center or containing clusters of different shapes.

[0031] (4) Using the improved evidence-based reasoning method to allocate samples can make the sample allocation results more reasonable and can effectively identify outliers.

[0032] (5) It further improves the effectiveness of cluster analysis results.

[0033] (6) A density clustering analysis algorithm combining information particle fusion and evidence reasoning was used to cluster the protein sequence data of cells, which enabled the prediction of protein subcellular localization and obtained accurate protein subcellular localization results, thus providing technical support for judging or designing specific proteins in corresponding subcellular structures or predicting the function of proteins.

[0034] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0035] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0036] Figure 1 This is a schematic flowchart illustrating a protein clustering method that combines information particle fusion and evidence reasoning, according to an exemplary embodiment. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0038] Using yeast protein sequence data as an example, the protein clustering method of this invention, which combines information particle fusion and evidence reasoning, is used for analysis, such as... Figure 1As shown, subcellular localization prediction of proteins in yeast cells can be performed, that is, after cluster analysis, it can be determined which subcellular structure each protein belongs to.

[0039] This dataset contains 1484 protein sequence samples, each with 8 attribute information points, and 10 types of given subcellular structures. The dataset is shown in the table below.

[0040]

[0041]

[0042] The specific implementation steps of the embodiment are as follows:

[0043] Step 1: Granulate the protein sequence data of yeast cells. Based on the preset parameter k, and the sparsity sd(x) of each protein sequence sample... i The protein sequence dataset U is granulated to generate a set of g information granules {G1, G2, ..., G...}. g}

[0044] Wherein, the protein sequence dataset U = {x1, x2, ..., x} n It consists of n protein sequence samples, each protein sequence sample x i Contains w attribute information {x i1 ,x i2 ,…,x iw In this embodiment, n = 1484, w = 8;

[0045] k takes the value of This represents the rounding up of non-integer values, where c is the given number of clusters. Each cluster represents a subcellular structure type, and the number of clusters represents the number of subcellular structure types. In this example, the sample size is n = 1484, and the number of clusters is c = 10.

[0046] Sparsity of each protein sequence sample It is an information granularity that simultaneously measures the global and local density of a protein sequence sample, where, To measure the global density of protein sequence samples, the radius of the neighborhood that maximizes the relative density of the protein sequence samples is used. Let |·| represent the relative density of each protein sequence sample under different neighborhood radii, |·| represent the cardinality of the set, and δ(x) represent the neighborhood of each protein sequence sample under different neighborhood radii. i ,d ij )={x z |x z ∈U,d(x i ,x z)≤d ij Protein sequence sample x i and x j The Euclidean distance d(x) between them i ,x j )=d ij =‖x i -x j ‖2, i,j=1,2,…,n and i≠j, The k-nearest neighbor radius of a protein sequence sample measures the local density of the protein sequence sample. Represents protein sequence sample x i The k-th nearest neighbor.

[0047] For each information particle G, there exists a protein sequence sample x. i , making And |G|=k, and, That is, the information particle G is a particle with x i Centered on A hypersphere with radius x contains k protein sequence samples, and the center x... i Its sparsity is minimal.

[0048] Step 2: Fusion of information particles from protein sequence data based on intersection relationships. If any two information particles intersect |G i ∩G j If |≥1, meaning the two information particles share a common protein sequence sample, then these two information particles are fused. After each pair of information particles is fused, a set of information particles {G1,G2,…,G...} is formed. g} Transform into a set of g * Individual clusters

[0049] Step 3: Perform cluster fusion of protein sequence data based on density propagation, targeting any cluster for GC. a and the nearest GC cluster b If sd * (GC a )≥sd * (GC b ), i.e., GC b The density is greater than that of GC a If the density is such that the two clusters are fused, and after each cluster is fully fused, a group of clusters is formed. Transformed into a set of g′ particle groups {GF1,GF2,…,GF g′}

[0050] Wherein, the distance d between any two clusters * (GC a GCb ) = min{d ij ||x i ∈ GC a , x j ∈ GC b}, that is, the distance between the closest pair of protein sequence samples in two grain clusters is used as the distance between grain clusters, where a, b = 1, …, g * and a ≠ b, the sparsity of the grain cluster That is, the average value of the sparsity of protein sequence samples within the grain cluster represents the density of the grain cluster.

[0051] Step 4: Perform grain swarm fusion on protein sequence data based on distance. For a given number of clusters c, if c = g′ or c = g * then execute Step 4.1, if c < g′ then execute Step 4.2, if c > g′ then execute Step 4.3.

[0052] Among them, the distance d between any two grain swarms * (GF a , GF b ) = min{d ij |x i ∈ GF a , x j ∈ GF b}, that is, the distance between the closest pair of protein sequence samples in two grain swarms is used as the distance between grain swarms.

[0053] Step 4.1: Directly use the g′ grain swarms {GF1, GF2, …, GF g′} or g * grain clusters as the initial clusters {Cl1, Cl2, …, Cl c}.

[0054] Step 4.2: Fuse the pair of grain swarms with the closest distance, update the number of grain swarms and the distance between grain swarms, and execute Step 4 again.

[0055] Step 4.3: Treat each grain swarm as a sub - data set, execute Steps 1 to Step 3 again for each sub - data set, combine all the obtained grain swarms into a new set of grain swarms, where all protein sequence samples not included in the grain swarms are regarded as a grain swarm, and execute Step 4 again.

[0056] Step 5: Calculate the evidence value of stable samples. Use the protein sequence samples within the initial cluster Ω = {Cl1, Cl2, …, Cl c} as stable samples, and calculate the evidence values of stable samples belonging to each single cluster and the cluster全集 (the whole set of clusters). Among them, u = 1, 2, …, c, A∈2 Ω {Cl1,Cl2,…,Cl} c ,Ω}, initialize set S, and add all stable samples to set S.

[0057] Step 6: Calculate the evidence value for unstable samples, specifically the protein sequence sample x with the smallest sparsity that does not belong to set S. i Based on protein sequence samples x in set S j Calculate x i Evidence values ​​belonging to each individual cluster and the entire cluster set: in x represents i In set S, the evidence values ​​provided by the k nearest neighbors are merged using the Dempster merging rule formula. Where “·” represents any Cl u Or Ω, x i Add to set S, and repeat step 6 until S = U.

[0058] Dempster merge rule formulas include the following three general formulas:

[0059] Formula 1:

[0060] Formula 2:

[0061] Formula 3:

[0062] Where A, B, and C can be any Cl u or Ω

[0063] Formula in step 6 This refers to using Equation 1 to pair Merge until all Combined into one value

[0064] Step 7: Assign the protein sequence sample to a category based on the evidence value. For each protein sequence sample x i Assign it to cluster The final clusters are the clustering results. If, for a given outlier threshold τ, Then x i These are outlier samples.

[0065] For particle swarm samples, i.e., protein sequence samples x i It may be a protein with a unique structure, possibly possessing distinctive functions, and thus deserves special attention.

[0066] In some embodiments, the outlier threshold τ ranges from [0.99, 1].

[0067] In this embodiment, the dataset does not contain outliers, therefore no outlier threshold τ is set, and the data is directly based on... By allocating samples, we can eventually obtain the clustering results of the dataset, which corresponds to the subcellular localization of each protein sequence sample.

[0068] To demonstrate the effectiveness of this invention, clustering experiments were conducted on yeast protein sequence data using k-means clustering, DBSCAN clustering, DPC clustering, and the algorithm of this invention. The number of clusters in the k-means and DPC clustering methods was set as the actual number of clusters in the dataset. The effectiveness of each method was evaluated using the ARI (Adjusted Land Coefficient), AMI (Adjusted Mutual Information), and FMI (Fowlkes-Mallows Index). Higher values ​​for each index indicate greater effectiveness. Since the experimental results of the k-means clustering method are random, the average of the index values ​​from 10 experiments was used as the final index value. The clustering results are shown in the table below.

[0069]

[0070] In comparative experiments based on yeast protein sequence data, the algorithm of this invention showed higher values ​​than other clustering methods under all evaluation indicators. This indicates that the algorithm of this invention can perform protein subcellular localization prediction very well, that is, it can cluster proteins very well, and its performance is better than some existing clustering methods.

[0071] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the following claims.

[0072] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A protein clustering method combining information granulation and evidential reasoning, characterized in that, Comprising: Step 1: granulate the information of protein sequence data in cells, according to preset parameter k, based on the sparsity sd(x i ) of each protein sequence sample, granulate the protein sequence data set U to generate a group of g information granules {G1, G2,..., G g} Wherein, the protein sequence data set U = {x1, x2,..., xn} is composed of n protein sequence samples, each protein sequence sample x n contains w attribute information {x1, x2,..., xw}. i i1 i2 iw};​​​ k takes values denotes the upward rounding of non-integers, c is the given number of clusters, one cluster represents one subcellular structure class, and the number of clusters represents the number of subcellular structure classes; Sparsity of each protein sequence sample It is an information granularity that simultaneously measures the global and local density of a protein sequence sample, where, To measure the global density of protein sequence samples, the radius of the neighborhood that maximizes the relative density of the protein sequence samples is used. Let |·| represent the relative density of each protein sequence sample under different neighborhood radii, |·| represent the cardinality of the set, and δ(x) represent the neighborhood of each protein sequence sample under different neighborhood radii. i d ij )={x z |x z ∈U, d(x i x z )≤d ij Protein sequence sample x i and x j The Euclidean distance d(x) between them i x j )=d ij =||x i -x j ||2, i, j = 1, 2, ..., n and i ≠ j, The k-nearest neighbor radius of a protein sequence sample measures the local density of the protein sequence sample. Represents protein sequence sample x i The kth nearest neighbor; For each information grain G, there exists a protein sequence sample x i such that and |G| = k, and that is, information grain G is a hypersphere with x i as the center and as the radius, containing k protein sequence samples, and the sparsity of the center x i is the smallest; Step 2: Fusion of information particles from protein sequence data based on intersection relationships. If any two information particles intersect |G i ∩G j If |≥1, meaning the two information particles share a common protein sequence sample, then these two information particles are fused. After each pair of information particles is fused, a set of information particles {G1, G2, ..., G...} is formed. g } Transform into a set of g * Individual clusters Step 3: Cluster fusion of protein sequence data based on density propagation, for any cluster GC a and the cluster GC b closest to it, if sd * (GC a ) ≥ sd * (GC b ), i.e. the density of GC b is greater than the density of GC a , then fuse the two clusters, after each cluster is fused, a set of clusters is converted into a set of g' clusters {GF1, GF2,..., GF g′} wherein the distance d between any two granules * (GC a , GC b ) = min{d ij | x i ∈ GC a , x j ∈ GC b}, i.e. the distance between the closest pair of protein sequence samples in two granules as the distance between granules, wherein a, b = 1,..., g * and a≠b, the sparsity of a granule i.e. the average of the sparsity of protein sequence samples within a granule represents the density of the granule; Step 4: Swarm fusion of protein sequence data based on distance, for a given number of clusters c, if c = g' or c = g * Step 4.1 is performed if c < g', Step 4.2 is performed if c > g', Step 4.3 is performed if c > g'. wherein the distance d between any two populations is * (GF a , GF b ) = min{d ij | x i ∈ GF a , x j ∈ GF b}, i.e. the distance between the pair of protein sequence samples that are closest to each other in the two populations is taken as the distance between the populations; Step 4.1: Divide the g′ particle groups {GF1, GF2, ..., GF...} into g′ groups. g′ } or g * Individual clusters Directly used as the initial cluster {Cl1, Cl2, ..., Cl c }; Step 4.2: merging the nearest pair of particle groups, updating the number of particle groups and the distance of particle groups, and performing step 4 again; Step 4.3: regarding each particle group as a sub-data set, performing steps 1-3 again for each sub-data set, combining all the obtained particle groups to form a new set of particle groups, wherein all protein sequence samples not included in the particle groups are regarded as a particle group, and performing step 4 again; Step 5: Calculate the evidence value of stable samples, take the protein sequence samples in the initial cluster Ω = {Cl1, Cl2,..., Cl c} as stable samples, and calculate the evidence value of stable samples belonging to each single cluster and cluster set Where u = 1, 2,..., c, A∈2 Ω \{Cl1, Cl2,..., Cl c , Ω}, initialize the set S, and add all stable samples to the set S; Step 6: Calculate the evidence value for unstable samples, specifically the protein sequence sample x with the smallest sparsity that does not belong to set S. i Based on protein sequence samples x in set S j Calculate x i Evidence values ​​belonging to each individual cluster and the entire cluster set: in x represents i In set S, the evidence values ​​provided by the k nearest neighbors are merged using the Dempster merging rule formula. Where "·" represents any Cl u Or Ω, x i Add to set S, and repeat step 6 until S = U; Step 7: Assigning protein sequence samples classes according to the evidence values, for each protein sequence sample x i assign it to a cluster The final clusters are the clustering results, if for a given outlier threshold τ, then x i is an outlier sample.

2. The protein clustering method of claim 1, wherein, Further comprising: The value range of the outlier threshold τ is [0.99, 1].

3. A computer device, characterized by Comprising: A memory for storing instructions; And A processor for calling the instructions stored in the memory to perform the protein clustering method combining information particle fusion and evidence reasoning according to any one of claims 1-2.

4. A computer-readable storage medium, characterized in that, The instructions stored in the memory are executed by the processor to perform the protein clustering method combining information particle fusion and evidence reasoning according to any one of claims 1-2.

Images