CN101307359A - A method for identifying human gene promoters - Google Patents

Abstract

The invention discloses a human gene promoter identification method, which can be used for the determination of the human gene promoter region and the interpretation of its structure and function, and can be used to discover new unknown genes, including the following steps: a) Based on the principal component analysis method, Establish a base generalized property score characterization system; b) use the base generalized property score to characterize the structure of human gene promoters and non-promoters; c) use the self-cross covariance method to characterize each promoter and non-promoter Variables were normalized; d) Using radial basis kernel support vector machine to establish a human gene promoter recognition model.

Description

A method for identifying human gene promoters

technical field

The invention relates to a human gene recognition method, in particular to a human gene promoter recognition method.

Background technique

The drawing of the human gene draft has successfully accelerated the analysis of the entire gene. The promoter is an important regulatory region for the transcriptional activity of each gene. The identification of promoter regions and their structure-function interpretation are the basis for understanding gene expression patterns, gene regulatory networks, cell differentiation and development. Promoter prediction plays a vital role in discovering new unknown genes and improving expression vectors or gene delivery systems in gene therapy methods. Promoter prediction has attracted widespread attention, and its prediction procedures are based on different concepts. The fundamental principle is that the characteristics of the promoter region are different from the DNA characteristics of other genes. These concepts include signal-based and content-based. Computer prediction and identification of biological promoters is a challenging task. The diversity of promoters and the limitations of the understanding of transcriptional regulation mechanisms have brought great difficulties to related research work. Homology alignment algorithms have been used for nucleotide sequence homology alignment, but promoter prediction is still in its infancy. Although alignment algorithms can be used to cluster homologous promoters, in most cases, the same The sequence conservation of the promoter element of the source gene is much lower than that of its coding sequence, therefore, the similarity search no longer provides useful clues for its functional identification (Duret et al., Curr. Opin. Struct. Biol., 1997, 7 :399). In addition, many promoters are regulated by multiple signaling pathways, and the functional requirements of specifically responding to different stimuli make the organizational structure of promoters more complex and diverse. Sometimes even promoters regulated by the same signaling pathway may have no sequence homology at all (Kirchhamer, et al., Proc. Natl. Acad. Sci. U.S.A., 1996, 93: 9322). In addition, there are many sequence structural features like transcription factor binding sites in promoters, and these characteristic structures are not unique to promoters, they are scattered throughout the genome, how to filter out these numerous noise signals has also become a Difficulties faced by computer prediction of promoters in large genomes (Sap, et al., Nature, 1989, 340: 242; Bohjanen, et al., Nucleic Acids Res., 1997, 25: 4481; Wang, et al. , Proc. Natl. Acad. Sci. U.S.A., 1998, 95:492). There are some programs that describe the sequence characteristics of promoters based on the transcription factor binding properties obtained from experiments, and use them as the basis for promoter prediction in turn, but the actual effect is not very ideal, and the omissions and false positives are serious.

Contents of the invention

In view of this, in order to solve the problems existing in the above-mentioned promoter prediction, the present invention provides a human gene promoter recognition method, which can be used for the determination of the human gene promoter region and the interpretation of its structure and function, and can be used to discover new unknowns. Gene.

The object of the present invention is achieved like this: a kind of human gene promoter recognition method comprises the steps:

a) Based on the principal component analysis method, a base generalized property score characterization system is established;

b) Characterize the structures of human gene promoters and non-promoters by applying the base generalized property score;

c) Normalize the characterization variables of each human gene promoter and non-promoter with the self-cross covariance method;

d) Establishment of human gene promoter recognition model with radial basis kernel support vector machine.

Further, step a) specifically includes the following steps:

a1) Select 1209 kinds of 0D-3D property parameters of 5 kinds of bases;

a2) Correlation analysis was performed on 1209 property parameters, and 41 property parameters were selected;

a3) Process the obtained base property parameters with the principal component analysis method to obtain 4 principal components;

a4) Calculate the scores of each principal component, and define the score vector as the base generalized property score;

Further, step b) specifically includes: using the four principal components involved in the base generalized property score vector to characterize the human gene promoter and non-promoter sequences along the 5'→3' direction, where each base is represented by 4 base generalized property score vector representation;

Further, step c) specifically includes the following steps: process the characterization variables of each promoter and non-promoter sequence obtained by self-cross covariance, set the step size 1 to 6, make the number of characterization variables of each sequence consistent, and Variables obtained through self-cross covariance processing are used as independent variables of the promoter recognition model;

Further, step d) specifically includes the following steps: first, define two indicator variables, respectively use "1" to represent a promoter sample, and use "-1" to represent a non-promoter sample, and use this indicator variable as the dependent variable of the promoter recognition model , Modeling Human Gene Promoter Recognition Using Radial Basis Kernel Support Vector Machines.

A human gene promoter identification method of the present invention, wherein the selected base generalized property score contains a large amount of information, clear physical and chemical meaning, strong characterization ability, easy interpretation of results, good expansion performance and simple operation; The variance method normalizes the characterization variables of each promoter and non-promoter, this method can minimize the loss of original variable information, and at the same time fully consider the interaction and mutual influence between adjacent bases ; and the radial basis kernel support vector machine can well correlate the relationship between the sequence representation variable transformed by self-cross covariance and the observed classification value through the kernel function technology, which can effectively prevent the model from overfitting. At the same time, The built model has good generalization performance.

Other advantages, objects and features of the present invention will be set forth in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the investigation and research below, or can be obtained from Taught in the practice of the present invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

Description of drawings

In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with the accompanying drawings, wherein:

Fig. 1 is a schematic diagram of the subject operating characteristic analysis of the identification result of the support vector machine model of the present invention.

Detailed ways

With reference to the accompanying drawings, the method for using the method of the present invention for human gene promoter recognition will be described in detail as an example, including the following steps:

a) Based on the principal component analysis method, a base generalized property score characterization system is established;

Collect 1209 property parameters of 5 kinds of bases (A, C, G, T and U), including: composition characteristics, number of functional groups, atomic center fragments and molecular characteristics, molecular electric distance vector (MEDV), molecular holographic distance vector ( MHDV), topology, number of rotations and paths, connectivity index, information index, autocorrelation, edge adjacency index, Burden eigenvalue, topological charge index, eigenvalue index, Randic molecular profile, geometry, radial basis based on different atomic distances Descriptor function (RDF), descriptor obtained by molecular structure characterization based on electric diffraction method (MoRSE), descriptor of weighted integral invariant molecule (WHIM) descriptor and set of geometry, topology and atomic weight (GETAWAY) descriptor, etc.; In addition, other relevant properties are included: property parameters such as highest occupied orbital (HOMO) energy, dipole moment and Wiener index.

Principal component analysis is used to compress the number of descriptors. In order to avoid serious multiple correlations between variables from harming principal components, correlation analysis is first performed on 1209 original variables. For each group of variables with a correlation coefficient greater than or equal to 0.90, according to its For the load size in the original variable matrix, one of them is retained, and the others are deleted. Finally, there are 41 remaining variables, which mainly reflect the following information of the base: average molecular weight, number of multiple bonds, average aromatic polarization, average Electrical topological state, total electron energy, thermodynamic properties, Moriguchi octane-partition coefficient (logP), number of urea derivatives, number of hydrogen bond acceptor atoms (N, O, F), E-state topological parameters, Kier flexibility index, Highest occupied orbital (HOMO) energy, molecular holographic distance vector, dipole moment, torsion energy and spatial structure, etc. After 41 variables are transformed by principal component analysis, the first 4 principal components accumulatively explain 99.99% of the variance of the original data matrix (5×41). The converted principal component scores are shown in Table 1. Therefore, these 4 principal components can be used The score matrix (5×4) replaces the original variable matrix (5×41).

Table 1 4 principal component scores of 41 property parameters of 5 bases

The analysis of the 4 principal component loads found that the WHIM index in the symmetrical direction of the third component weighted by the atomic mass is the largest positive contribution to the first principal component. The WHIM descriptor belongs to the 3D geometric descriptor and is the weight of the atomic coordinates. The PCA of the matrix covariance matrix is obtained, followed by the descriptor based on the structural information content. Both types of descriptors can be regarded as stereo (Steric) characteristic descriptors. The negative contributions are variables such as the Moran autocorrelation descriptor weighted by atomic polarizability and torsional energy. The positive contribution to the second principal component is the unweighted 3D-MoRSE descriptor component and electron energy and other variable information obtained by characterizing the molecular 3D structure based on the electron diffraction method. The larger negative contributions are variables such as the sum of topological distances between nitrogen atoms (N) and oxygen atoms (O). In the third principal component, the variables with larger positive loadings are the 2-channel Kier modified α shape index and Kier flexibility index, both of which belong to the topological class of descriptors. Information such as the average atomic polarizability (for carbon atoms) and the average molecular weight, which have a relatively large load, belong to the molecular composition class descriptors. The seventh component of the molecular holographic distance vector proposed by our research group is positively correlated with the loading of the fourth principal component. Molecular holographic distance vector is a descriptor based on molecular 2D topological structure obtained by dividing atoms into 13 atomic types and further defining atomic properties and relative bond lengths. The seventh component represents the atomic environment C- and >N-, >P -The holographic distance between ("-", ">", "<" respectively indicate that 1, 2, 2 non-hydrogen atoms or chemical bonds are connected with it). The non-weighted 3D-MoRSE descriptor component and the Moran autocorrelation descriptor weighted by the atomic polarizability are variable information that present a large negative correlation. For convenience, these 4 principal component score vectors are called base generalized property scores, because these 4 score vectors synthesize most of the information of the 1209 property parameters of bases from multiple perspectives, so it can be considered to try to use them for Nucleic acid sequence characterization.

b) Characterize the structures of human gene promoters and non-promoters by applying the base generalized property score;

Select 565 human gene promoter sequences, 3819 non-promoter sequences (890 exons and 2929 introns), and use the 4 principal components involved in the base generalized property score vector to pair the selected sequences along the 5' → Characterize in the 3' direction, and each base in the sequence is characterized by a vector of 4 base generalized property scores. Each sequence is characterized by n×4 variables according to the number of bases it contains (defined as n).

c) Normalize the characterization variables of each human gene promoter and non-promoter with the self-cross covariance method;

The characteristic variables of each promoter and non-promoter sequence are obtained by autocross covariance (ACC). This method takes into account all the interaction effects between the base parameters of different positions in the sequence. Therefore, it can be maximized during the data transformation process. reduce information loss. Assuming that the shortest sequence length in the studied sample set is l+1, for any sequence containing n bases, the ACC process is as follows:

${\mathrm{<span\; class="notranslate">\; ACC</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; l</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; a</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; Z</span>}}_{\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; l</span>}}}{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1,2,3</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; )</span>$

In the formula: l is the step length; i and i+l are the positions of the bases in the sequence; a and b are the corresponding descriptor component numbers of the i-th and i+l bases respectively, and for the base generalized property score vector, Its a, b=1,2,3,4. It can be seen that when all possible step sizes are calculated (l=1, 2, 3, ..., l), the number of descriptors of sequences of different lengths in the sample set after ACC processing is finally 4 ² ×l, Here, the step size l is selected as 6, so that each sequence can be represented by 4 ² ×6=96 variables, and the variables obtained through self-cross covariance processing are used as independent variables of the promoter recognition model.

d) Build a human gene promoter recognition model with radial basis kernel support vector machine;

First define two indicator variables, use "1" to represent the promoter sample, use "-1" to represent the non-promoter sample (exon and intron), and use this indicator variable as the dependent variable of the promoter recognition model, A human gene promoter recognition model was established with radial basis kernel support vector machine, and its parameters were set as: C=200.0, K(x, x _i )=exp(-0.125||xx _i || ² ). If A _cc is defined as the percentage of the total number of samples in which the number of correctly predicted samples is calculated, S _p is the percentage of the number of correctly predicted promoter samples, S _n is the percentage of the number of correctly predicted non-promoter samples, and MCC is Matthews Statistical parameters such as correlation coefficients were cross-validated by the leave-one-out method, and the support vector machine model identified _Acc = 83.8, S _n = 67.1, S _p = 86.3 and MCC = 565 promoters and 3819 non-promoters in the training set. 0.442, and further cross-validated by leaving 1/5 method to get A _cc = 81.7, S _n = 66.9, S _p = 83.8 and MCC = 0.406, which shows that based on the generalized base property score characterization, self-cross covariance normalization processing, The model built by the radial basis kernel support vector machine modeling process can better identify human gene promoters. The number of support vectors obtained by the leave-one-out method and the leave-one-fifth method accounted for 62.1% and 68.3% of the total samples, respectively, that is, 37.9% and 31.7% of the samples can be safely deleted without affecting their prediction of new samples The effect further shows that the support vector classification machine has good generalization performance.

Further take (1-S _p ) as the abscissa (X-axis), sensitivity (S _n ) as the ordinate (Y-axis), draw the receiver operating characteristic curve, see Figure 1, it can be seen that the retention of the built model The areas corresponding to the one method and the left 1/5 method are 0.835 and 0.819 respectively.

In order to further verify the prediction effect of the invented method on human gene promoters, 100 promoters and 100 containing genes were selected from the EPD database (http://www.epd.isb-sib.ch/) different from the training The subsequences are predicted, and the results predicted by the radial basis kernel support vector machine model are listed in Table 2. At the same time, 7 prediction servers are selected to compare the prediction results of 200 sequences. After comparison, it is found that the S _n obtained by the method of the present invention is and MCC are the highest, indicating that it has obvious advantages for the prediction of human gene promoters.

Table 2 Comparison of human gene promoter prediction results

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims (5)

1. a process for recognising human gene promoter is characterized in that comprising the steps:

A), make up base broad sense character score representation system based on principal component analytical method;

B) application base broad sense character score characterizes the structure of Human genome promotor and non-promotor;

C) with the sign variable of everyone genoid promotor and non-promotor being done normalized from intersecting covariance method;

D) set up Human genome Promoter Recognition model with radially basic nuclear SVMs.

2. according to a kind of process for recognising human gene promoter of claim 1, it is characterized in that step a) specifically comprises the steps:

A1) 1209 kinds of 0D-3D nature parameters of 5 kinds of bases of selection;

A2) 1209 kinds of nature parameters are done correlation analysis, selectedly obtain 41 nature parameters;

A3) handle the base nature parameters that obtains with PCA, obtain 4 principal constituents;

A4) calculate each principal component scores, will get resolute and be defined as base broad sense character score.

3. according to a kind of process for recognising human gene promoter of claim 2, it is characterized in that step b) specifically comprises: characterize with the sequence of related 4 principal constituents of base broad sense character score vector to Human genome promotor and non-promotor, each base in the sequence is with 4 base broad sense character score characterization vectors.

4. according to a kind of process for recognising human gene promoter of claim 3, it is characterized in that step c) specifically comprises the steps: with handling each promotor obtain and the sign variable of non-promoter sequence from intersecting covariance, it is 6 that step-length l is set, make the sign variable number unanimity of each sequence, and will handle the variable obtain independent variable(s) through intersecting covariance certainly as the Promoter Recognition model.

5. according to each a kind of process for recognising human gene promoter in the claim 1 to 4, it is characterized in that step d) specifically comprises the steps: at first to define two indieating variables, use " 1 " expression promotor sample respectively, with the non-promotor sample of " 1 " expression, with the dependent variable of this indieating variable, set up Human genome Promoter Recognition model with radially basic nuclear SVMs as the Promoter Recognition model.

Images