CN108804870A - Key protein matter recognition methods based on Markov random walks - Google Patents

Key protein matter recognition methods based on Markov random walks Download PDF

Info

Publication number
CN108804870A
CN108804870A CN201810499870.6A CN201810499870A CN108804870A CN 108804870 A CN108804870 A CN 108804870A CN 201810499870 A CN201810499870 A CN 201810499870A CN 108804870 A CN108804870 A CN 108804870A
Authority
CN
China
Prior art keywords
protein
vertex
formula
attribute
score
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810499870.6A
Other languages
Chinese (zh)
Other versions
CN108804870B (en
Inventor
刘维
马良玉
陈昕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Yangzhou University
Original Assignee
Yangzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Yangzhou University filed Critical Yangzhou University
Priority to CN201810499870.6A priority Critical patent/CN108804870B/en
Publication of CN108804870A publication Critical patent/CN108804870A/en
Application granted granted Critical
Publication of CN108804870B publication Critical patent/CN108804870B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The key protein matter recognition methods based on Markov random walks that the object of the present invention is to provide a kind of, belongs to technical field of biological information.Key protein matter recognition methods based on Markov random walks:Use the thought of Markov random walks, the score for indicating its significance level is assigned to each vertex, the score on all vertex constitutes the vector of n row, provides the initial value of score, random walk and is modified in transmission in a network by score according to certain probability;The descending arrangement of score value is finally pressed, output score value is correspondingkA protein is final result.The present invention merges biological attribute and topological property improves the accuracy of identification key protein matter, while keeping prediction result more accurate, improves the efficiency of prediction.

Description

Key protein matter recognition methods based on Markov random walks
Technical field
The invention belongs to technical field of biological information, random by Markov mainly in protein-protein interaction network The technology of the algorithm identification key protein matter of migration, more particularly to network topological information and protein bio attribute in PPI networks The method for identifying key protein matter.
Background technology
Protein is the indispensable substance of institute in vital movement, almost takes part in all periods of vital movement, and is closed Key protein is even more to have played the effect that do not replace in this course, the missing of key protein matter may cause life entity without Method is survived.Therefore, identify that the key protein matter in PPI networks not only facilitates the adjusting and controlling growth process for understanding cell, and Help can be provided to the research of biological evolution mechanism.In addition, in biomedical sector, the identification of key protein matter is controlled in disease It treats and the design etc. of drug target cell has great importance.
Before the present invention proposes, the identification field of key protein matter most begins by the topological characteristic of network to know Not, for example, degree centrality (DC), betweenness center (BC), local average degree of communication (LAC), Li et al. people fusion PPI and gene table Centrality Measurement Method PeC, Zhang et al. fusion PPI network topology characteristics are proposed up to data and gene co-expressing information carries CoEWC methods are gone out, but the shortcomings that these methods identification key protein matter is:(1) it only considered possessed by network itself Topological characteristic, and have ignored the intrinsic biological characteristic of protein.(2) the PPI networks obtained by Bioexperiment, which exist, makes an uproar Sound so that there are false positives for protein interaction data.
Invention content
The purpose of the present invention, which is that, overcomes drawbacks described above, develops the key protein matter identification based on Markov random walks Method.Key protein matter recognition methods based on Markov random walks uses the thought of Markov random walks, to each Vertex assigns the score for indicating its significance level, and the score on all vertex constitutes the vector of n row, provides the initial of score Value random walk and is modified according to certain probability by score in transmission in a network.It is finally descending by score value Arrangement, the corresponding k protein of output score value is final result.
Key protein matter recognition methods based on Markov random walks, is mainly characterized by following steps:
(1) PPI networks and biological information are inputted;
(2) according to the attribute value on protein vertex and side right value, the weight q between protein vertex is calculated, builds weight Matrix;
(3) all attribute values are normalized, build attribute matrix;
(4) according to the interaction relationship between protein vertex, transfer matrix is built;
(5) according to PageRank algorithms, iteration obtains score vector r, determines to return to probability P by the attribute on vertex;
(6) it obtains object function and object function is optimized, to initial value r, q declines formula using gradient and changes Generation update;
(7) r after acquisition iteration(t)=(r1,r2,…,rn) the descending sequence of value, after sequence it is maximum k value for close Key protein.
The step (2) calculates the weight q between protein vertex according to the attribute value and side right value on protein vertex, Build weight matrix:By step (1) according to PPI networks, the weight between protein passes through common neighbours' phase between them It is acquired like degree, based on expression similarity, GO semantic similarities.
All attribute values are normalized the step (3), build attribute matrix:Pass through Z-Score or normalization Method makes attribute value all be included in (0,1) range, and all vertex attribute vectors constitute an attribute matrix.
Advantages of the present invention and effect are that this method not only considers the topological characteristic of protein-protein interaction network, together When have also contemplated the biological attribute of protein, and then overcome data noise it is high caused by negative effect.Merge biological attribute and Topological property improves the accuracy of identification key protein matter, while keeping prediction result more accurate, improves the efficiency of prediction. Extend the technology biological information field application range and practicability.
Description of the drawings
Fig. 1 --- the present invention is based on the flow diagrams of the key protein matter recognition methods of Markov random walks.
Fig. 2 --- present invention figure compared with the quantity for the key protein matter that other methods identify.
Fig. 2 a be the present invention preceding 100 protein in key protein matter number comparison figure;
Fig. 2 b be the present invention preceding 200 protein in key protein matter number comparison figure;
Fig. 2 c be the present invention preceding 300 protein in key protein matter number comparison figure;
Fig. 2 d be the present invention preceding 400 protein in key protein matter number comparison figure;
Fig. 2 e be the present invention preceding 500 protein in key protein matter number comparison figure;
Fig. 2 f be the present invention preceding 600 protein in key protein matter number comparison figure;
The statistical indicator results contrast figure of Fig. 3 --- the present invention and other methods.
Specific implementation mode
The present invention technical thought be:Biological attribute and topological property are combined, the think of of Markov random walks is used Think, the score for indicating its significance level is assigned to each vertex, and the score on all vertex constitutes the vector of n row, provides The initial value of score random walk and is modified according to certain probability by score in transmission in a network.That is root first The weight between protein is obtained according to common neighbours' similarity, based on expression similarity, GO semantic similarities, weight square can be obtained Battle array constitutes an attribute matrix according to all vertex attribute vectors.Secondly, vertex is obtained by protein interaction relationship Transition probability between, thus to obtain transfer matrix.It is final to obtain object function, and object function is optimized, finally Identify key protein matter.Fusion biological attribute and topological property help to understand the function of agnoprotein matter, special for explaining Determine the molecular mechanism important in inhibiting of function, while important theoretical foundation can be provided medicine target cell design etc..Institute Naturally enough it is suitable for the detection of key protein matter with the key protein matter recognition methods based on Markov random walks.
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
As shown in Figure 1, the key protein matter recognition methods based on Markov random walks, includes the following steps:
Step 1:Input PPI networks and biological information
Step 2:According to the attribute value on protein vertex and side right value, the weight q between protein vertex, structure power are calculated Weight matrix
Common neighbours' similarity (NTE):The topological characteristic of protein-protein interaction network is in the identification of key protein matter There is irreplaceable status, according to " center-lethal " rule, key protein matter is more likely that cluster exists in a network , therefore, we use common neighbours' similarity (NTE) as one of index to weigh the key of protein.In non-directed graph G In (V, E), common neighbours' similarity between protein u and v is expressed as:
NTE (u, v)=| Cu∩Cv|+1 formula (1)
Wherein, Cu(or Cv) indicate PPI nodes u (or v) neighbours set;|Cu∩Cv| indicate node u and The number of the common neighbours of v, the i.e. number of triangle belonging to side.By adding " 1 " come so that result is all higher than behind result 0, to avoid going wrong in standardising process.
Gene expression similarity (GES):Since gene expression data acquisition is easier and in the identification of key protein matter There is extensive use in field, while the gene co-expressed more likely becomes key protein matter, so we use gene table Up to similarity (GES) as an index for weighing key protein matter.The gene calculated between protein u and v that we use The formula for expressing similarity is as follows:
Wherein, s is the quantity of sample in gene expression data, and U and V are the gene code of corresponding protein u, v, UiWith ViExpressions of the gene U and V in corresponding sample i is indicated respectively,WithIt is being averaged for the expression of gene U and V Value, then o (U) and o (v) indicates the standard deviation of the expression of gene U, V respectively.
GO semantic similarities (GOS):Gene ontology (gene ontology, GO) is the molecule work(of gene (gene outcome) The relevant information of energy, biological process and cell components provides a specification, accurate terminology.To the semanteme of GO terms Similitude carries out the importance that measurement is GO applications, and GO semantic similarities are the biological properties based on gene to disclose The functional similarity of gene, and two connected key protein matter more likely participate in the same bioprocess.Permitted in recent years More scholars are proposed the measure of GO Semantic Similarities, we calculate GO semantic similarities, the party using the method for Lin The characteristics of method, is:First, two standardization that compare the sum of information content of concept;Secondly, it is assumed that be compared two are general Thought is independent.GO semantic similarities between protein u, v are defined using following formula:
Wherein, the protein u and v, c of gene U, V codings interaction1、c2The respectively GO terms of gene U, V, S (c1, c2) it is node c1、c2Nearest public ancestor node set, the example probabilistic of variable c is P (c), PmsBe they it is public most The probability that nearly ancestors occur.
In PPI networks, the weight (w between proteinij) can be acquired by the similarity between them, it is specific to calculate Formula is as follows:
wij=a1NTE(i,j)+a2GES(i,j)+a3GOS (i, j) formula (5)
Wherein, parameter a1、a2、a3In (0,1) range, and and it is 1.
Matrix W=[wij] be PPI networks weight matrix, wijFor side (vi, vj) on weight:
Step 3:All attribute values are normalized, attribute matrix is built
All properties value is standardized (can be such that attribute value is all included in (0,1) with Z-Score or method for normalizing In range), all vertex attribute vectors constitute an attribute matrix B=[bij]nxm
Step 4:According to the interaction relationship between protein vertex, transfer matrix is built;
Provide constant k<N finds out the maximum k protein of importance, i.e. Top-k, referred to as key protein matter in G.I Using Markov random walks thought, to each vertex viAssign a score r for indicating its significance leveli (0), own The score value on vertex constitutes a score vectorFor the column vector of n × 1, the initial value of r is provided, according to Certain probability migration and is modified in transmission in a network by score.From viIt is transmitted to vjDefinition of probability be:
In this way, the transition probability between all-pair constitutes the transfer matrix P=[p of n × nij]。
Step 5:According to PageRank algorithms, iteration obtains score vector r, determines to return to probability P by the attribute on vertex
In traditional PageRank algorithms based on random walk, score vector r is updated with following iteration:
r(k+1)=α PΤr(k)+(1-α)P0Formula (8)
Wherein α is constant, α ∈ (0,1), P0∈ (0,1) is constant, is that the particle of migration returns to the probability of Original Departure Point. In the algorithm that this chapter is proposed, we use the attribute b on vertexiTo determine to return to probability P0If
HereFor the column vectors of m × 1, qjFor the weight of j-th of attribute, such formula is:
r(k+1)=α PΤr(k)+(1-α)P0=α PΤr(k)+(1-α)B·q(k)Formula (10)
If function is (10) formula and r(k+1)Square error:
We will find out r, q so that J (r, q) is minimum, that is, solves following optimization problem:
Constraints r>0, q>0 refers to r, and all score values in q are all positive number.
Step 6:Obtain object function and object function optimized, to initial value r, q using gradient decline formula into Row iteration updates
After obtaining object function, we start to optimize object function.Ask J for r, the local derviation of q first:From formula (11):
J (r, q)=(α PΤ·r+(1-α)B·q-r)Τ·(αPΤR+ (1- α) Bq-r) formula (13)
2rΤPPΤr+2α(1-α)rΤP·B·q-2αrΤP·r+(1-α)2qΤBΤB·q-2(1-α)rΤB·q+rΤr
It can be obtained by formula (13):
According to above-mentioned gradient, for initial value r(0), q(0), we decline formula using gradient and are updated with regard to row iteration:
Wherein, ρ is iteration sum.
Step 7:R after acquisition iteration(t)=(r1,r2,…,rn) the descending sequence of value, maximum k value is after sequence Key protein matter.
Embodiment:
In order to verify the performance for the algorithm EPM that this chapter is proposed, we are by the quantity of the key protein matter of identification and other Five kinds of methods (DC, BC, LAC, PeC and CoEWC) are compared.To each method we select top100, top200, The protein identification result of top300, top400, top500 and top600 are as Candidate Set, to the protein in each Candidate Set It seeks common ground again with the key protein matter set of standard, to obtain the quantity of true key protein matter in Candidate Set, experiment knot Fruit is as shown in Figure 2.
It can be seen that from Fig. 2 a, 2b, 2c, 2d, 2e, 2f, in yeast PPI networks, the algorithm EPM that we are proposed is knowing Effect that will be good than other methods can be obtained in other key protein matter, take top100, top200, top300, top400, When the key protein matter of top500 and top600 is as Candidate Set, the protein amounts that algorithm that this chapter is proposed identifies are apparent Higher than other methods.Wherein, compared with PeC methods, before taking top100, top200, top300, top400, top500 and When top600 protein, 16.4%, 18.8%, 19.5%, 19.4%, 20.5% and has been respectively increased in the accuracy rate of EPM 22.6%.
In order to further show advantages of the EPM in prediction key protein matter, we attempt in a smaller data set The difference of upper (taking 200 protein of top) analysis EPM and other methods.We find out Chong Die with EPM in this 200 protein Protein, and key analysis is carried out to remaining protein, as shown in table 1.
1 key protein matter volume comparison analysis of table
Table 1 analyzes the key protein matter identified in the data set of top200 by EPM and other 5 kinds of methods and non-pass The quantity of key protein compares.Wherein MiIndicate other 5 kinds of centrality methods for comparison, | EPM ∩ Mi| it is EPM and other The quantity of the key protein matter overlapping of method identification, | Mi- EPM | expression passes through MiIdentify and pass that EPM could not be identified The quantity of key protein, it is similar, | EPM-Mi| indicate that EPM is identified and MiFail the number of key protein matter identified Amount.It should be clear that coming from table, by EPM rather than the number of key protein matter that identifies of other methods Amount is significantly more than the quantity for the key protein matter that other methods rather than EPM are identified, and the key identified by EPM methods The quantity of non-key protein is also significantly less than other methods in protein.These results show EPM algorithms consider topology and More biological informations can effectively improve the prediction result of key protein matter.
In order to further evaluate performance of the EPM methods in terms of key protein matter prediction, we by its in other five kinds Disposition method is compared, we introduce statistics performance estimating method, including 6 evaluation indexes, sensibility sensitivity (SN), specific specificity (SP), positive ident value positive predictive value (PPV), feminine gender mark Value negative predictive value (NPV), F- assess F-measure (F) and precision accuracy (ACC), these The definition difference of statistical indicator is as follows:
SN indicates the ratio that key protein matter is predicted correctly.
SP indicates the ratio that non-key protein is correctly excluded.
PPV indicates the ratio correctly identified in the key protein matter identified.
NPV indicates to be predicted correctly as the ratio of non-key protein in the protein excluded.
F indicates the harmonic-mean of susceptibility and positive predictive value.
ACC indicates the ratio of correct result in all recognition results.
Wherein, it refers to that key protein matter is correctly identified as key protein that TP (True Positives), which represents true positives, The quantity of matter;It refers to that the non-key egg of key protein matter is mistakenly identified as by algorithm that FP (False positives), which represents false positive, The quantity of white matter;It refers to that non-key protein is identified as non-key protein that TN (True Negatives), which represents true negative, Quantity;It refers to that key protein matter is mistakenly identified as non-key protein that FN (False Negatives), which represents false negative, Quantity.Above six kinds of fingers target value is bigger, illustrates that the recognition performance of algorithm is better, the result of calculation of the statistical indicator of each method is such as Shown in Fig. 2.
From figure 3, it can be seen that 6 indexs of EPM estimate obviously higher than other five kinds of centrality in either method, Compared with DC, BC and LAC method based on network topology, the accuracy rate of EPM is significantly higher, and with incorporated gene expression number According to PeC methods compare, the algorithm of this chapter can still obtain higher accuracy rate.

Claims (6)

1. the key protein matter recognition methods based on Markov random walks, which is characterized in that the recognition methods includes as follows Step:
(1) PPI networks and biological information are inputted;
(2) according to the attribute value on protein vertex and side right value, the weight q between protein vertex is calculated, builds weight matrix;
(3) all attribute values are normalized, build attribute matrix;
(4) according to the interaction relationship between protein vertex, transfer matrix is built;
(5) according to PageRank algorithms, iteration obtains score vector r, determines to return to probability P by the attribute on vertex;
(6) it obtains object function and object function is optimized, to initial value r, q declines formula using gradient and is iterated more Newly;
(7) r after acquisition iteration(t)=(r1,r2,…,rn) the descending sequence of value, maximum k value is crucial egg after sequence White matter.
2. the key protein matter recognition methods according to claim 1 based on Markov random walks, which is characterized in that institute Attribute value and side right value of the step (2) according to protein vertex are stated, the weight q between protein vertex is calculated, builds weight square Battle array:By step (1) according to PPI networks, the weight between protein passes through common neighbours' similarity between them, gene table It is acquired up to similarity, GO semantic similarities;
Common neighbours' similarity is expressed as:
NTE (u, v)=| Cu∩Cv|+1 formula (1)
Wherein, CuIndicate the set in the neighbours of PPI nodes u, CvIndicate the collection in the neighbours of PPI nodes v It closes;|Cu∩Cv| indicate the number of the common neighbours of node u and v, the i.e. number of triangle belonging to side;
The formula for calculating the gene expression similarity between protein u and v is as follows:
Wherein, s is the quantity of sample in gene expression data, and U and V are the gene code of corresponding protein u, v, UiAnd ViPoint Not Biao Shi expressions of the gene U and V in corresponding sample i,WithIt is the average value of the expression of gene U and V, thenWithThe standard deviation of the expression of gene U, V is indicated respectively;
GO semantic similarities are calculated using the method for Lin:
Wherein, the protein u and v, c of gene U, V codings interaction1、c2The respectively GO terms of gene U, V, S (c1,c2) be Node c1、c2Nearest public ancestor node set, the example probabilistic of variable c is P (c), PmsIt is node c1、c2It is public nearest The probability that ancestors occur;
In PPI networks, the weight w between proteinijIt is acquired by the similarity between two protein, specific formula for calculation It is as follows:
wij=a1NTE(i,j)+a2GES(i,j)+a3GOS (i, j) formula (5)
Wherein, parameter a1、a2、a3In (0,1) range, and a1、a2、a3The sum of be 1;
Matrix W=[wij] be PPI networks weight matrix, wijFor side (vi, vj) on weight:
3. the key protein matter recognition methods according to claim 1 based on Markov random walks, which is characterized in that institute It states step (2) all attribute values are normalized, the method for building attribute matrix is:Pass through Z-Score or normalization side Method makes attribute value all be included in (0,1) range, and all vertex attribute vectors constitute an attribute matrix.
4. the key protein matter recognition methods according to claim 1 based on Markov random walks, which is characterized in that institute Step (4) is stated according to the interaction relationship between protein vertex, the method for building transfer matrix is:
Provide constant k<N finds out the maximum k protein of importance, i.e. Top-k, referred to as key protein matter in G;Using The thought of Markov random walks, to each vertex viAssign a score r for indicating its significance leveli (0), all vertex Score value constitutes a score vectorFor the column vector of n × 1, the initial value of r is provided, according to certain Probability migration and is modified in transmission in a network by score;From viIt is transmitted to vjDefinition of probability be:
In this way, the transition probability between all-pair constitutes the transfer matrix P=[p of n × nij]。
5. the key protein matter recognition methods according to claim 1 based on Markov random walks, which is characterized in that institute Step (5) is stated according to PageRank algorithms, iteration obtains score vector r, and the specific side for returning to probability P is determined by the attribute on vertex Method is:
In traditional PageRank algorithms based on random walk, score vector r is updated with following iteration:
r(k+1)=α PTr(k)+(1-α)P0Formula (8)
Wherein α is constant, α ∈ (0,1), P0∈ (0,1) is constant, is that the particle of migration returns to the probability of Original Departure Point;Use vertex Attribute biTo determine to return to probability P0If
HereFor the column vectors of m × 1, qjFor the weight of j-th of attribute, such formula is:
r(k+1)=α PTr(k)+(1-α)P0=α PTr(k)+(1-α)B·q(k)Formula (10)
If function is (10) formula and r(k+1)Square error:
Find out r, q so that J (r, q) is minimum, that is, solves following optimization problem:
Constraints r>0, q>0 refers to r, and all score values in q are all positive number.
6. the key protein matter recognition methods according to claim 1 based on Markov random walks, which is characterized in that institute It states step (6) to obtain object function and optimize object function, to initial value r, q declines formula using gradient and is iterated Newer method is:
After obtaining object function, object function is optimized:
Ask J for r, the local derviation of q first:From formula (11):
J (r, q)=(α PT·r+(1-α)B·q-r)T·(αPTR+ (1- α) Bq-r) formula (13)
2rTPPTr+2α(1-α)rTP·B·q-2αrTP·r+(1-α)2qTBTB·q-2(1-α)rTB·q+rTr
It can be obtained by formula (13):
According to above-mentioned gradient, for initial value r(0), q(0), decline formula using gradient and row iteration updated:
Wherein, ρ is iteration sum.
CN201810499870.6A 2018-05-23 2018-05-23 Markov random walk-based key protein identification method Active CN108804870B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810499870.6A CN108804870B (en) 2018-05-23 2018-05-23 Markov random walk-based key protein identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810499870.6A CN108804870B (en) 2018-05-23 2018-05-23 Markov random walk-based key protein identification method

Publications (2)

Publication Number Publication Date
CN108804870A true CN108804870A (en) 2018-11-13
CN108804870B CN108804870B (en) 2021-11-19

Family

ID=64091419

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810499870.6A Active CN108804870B (en) 2018-05-23 2018-05-23 Markov random walk-based key protein identification method

Country Status (1)

Country Link
CN (1) CN108804870B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109637579A (en) * 2018-12-18 2019-04-16 长沙学院 A kind of key protein matter recognition methods based on tensor random walk
CN110660448A (en) * 2019-09-20 2020-01-07 长沙学院 Key protein identification method based on topological and functional characteristics of protein
CN110910952A (en) * 2019-11-21 2020-03-24 衡阳师范学院 Method for predicting basic protein by using chemical reaction strategy
CN113436729A (en) * 2021-07-08 2021-09-24 湖南大学 Synthetic lethal interaction prediction method based on heterogeneous graph convolution neural network
CN113936743A (en) * 2021-11-12 2022-01-14 大连海事大学 Protein complex identification method based on heterogeneous PPI network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868582A (en) * 2016-03-25 2016-08-17 陕西师范大学 A method of identifying protein compounds by using a fruit fly optimization method
CN106355044A (en) * 2016-08-15 2017-01-25 上海电机学院 Protein composite identification method based on random walking model

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105868582A (en) * 2016-03-25 2016-08-17 陕西师范大学 A method of identifying protein compounds by using a fruit fly optimization method
CN106355044A (en) * 2016-08-15 2017-01-25 上海电机学院 Protein composite identification method based on random walking model

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
张小驰 等: "一种基于随机游走的迭代加权子图查询算法", 《计算机研究与发展》 *
洪海燕 等: "基于PPI网络的关键蛋白质的高效预测算法", 《计算机科学》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109637579A (en) * 2018-12-18 2019-04-16 长沙学院 A kind of key protein matter recognition methods based on tensor random walk
CN109637579B (en) * 2018-12-18 2022-04-15 长沙学院 Tensor random walk-based key protein identification method
CN110660448A (en) * 2019-09-20 2020-01-07 长沙学院 Key protein identification method based on topological and functional characteristics of protein
CN110660448B (en) * 2019-09-20 2022-02-01 长沙学院 Key protein identification method based on topological and functional characteristics of protein
CN110910952A (en) * 2019-11-21 2020-03-24 衡阳师范学院 Method for predicting basic protein by using chemical reaction strategy
CN113436729A (en) * 2021-07-08 2021-09-24 湖南大学 Synthetic lethal interaction prediction method based on heterogeneous graph convolution neural network
CN113936743A (en) * 2021-11-12 2022-01-14 大连海事大学 Protein complex identification method based on heterogeneous PPI network
CN113936743B (en) * 2021-11-12 2024-04-26 大连海事大学 Protein complex identification method based on heterogeneous PPI network

Also Published As

Publication number Publication date
CN108804870B (en) 2021-11-19

Similar Documents

Publication Publication Date Title
CN108804870A (en) Key protein matter recognition methods based on Markov random walks
CN108733976B (en) Key protein identification method based on fusion biology and topological characteristics
Pathan et al. Identifying stroke indicators using rough sets
CN110647904B (en) Cross-modal retrieval method and system based on unmarked data migration
CN110837602A (en) User recommendation method based on representation learning and multi-mode convolutional neural network
Lakshmi et al. Data clustering using k-means based on crow search algorithm
Naik et al. Efficient feature selection using one-pass generalized classifier neural network and binary bat algorithm with a novel fitness function
Kuo et al. An application of sine cosine algorithm-based fuzzy possibilistic c-ordered means algorithm to cluster analysis
CN113806582B (en) Image retrieval method, image retrieval device, electronic equipment and storage medium
CN109686402A (en) Based on key protein matter recognition methods in dynamic weighting interactive network
Kumar et al. Future of machine learning (ML) and deep learning (DL) in healthcare monitoring system
Hussain et al. Clustering uncertain graphs using ant colony optimization (ACO)
Ngueilbaye et al. SDLER: stacked dedupe learning for entity resolution in big data era
CN108256030B (en) Ontology-based density adaptive concept semantic similarity calculation method
Patil et al. Approaches for network analysis in protein interaction network
CN112149556B (en) Face attribute identification method based on deep mutual learning and knowledge transfer
CN109857928A (en) User preference prediction technique based on polynary credit evaluation
Saha et al. Missing value imputation in DNA microarray gene expression data: a comparative study of an improved collaborative filtering method with decision tree based approach
Chen et al. Research on Disease Prediction Method Based on R‐Lookahead‐LSTM
Ye et al. Boosted sine cosine algorithm with application to medical diagnosis
Srinivas et al. Hybrid Approach for Prediction of Cardiovascular Disease Using Class Association Rules and MLP.
Wang et al. Fuzzy k-means with variable weighting in high dimensional data analysis
Vinutha et al. EPCA—enhanced principal component analysis for medical data dimensionality reduction
CN117079821B (en) Patient hospitalization event prediction method
CN116110594B (en) Knowledge evaluation method and system of medical knowledge graph based on associated literature

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant