This article is the entry-level gene regulatory networks article introduces some basic concepts and common GRN models.
Concept: gene regulatory network (Gene Regulatory Network, GRN), referred to as regulatory network, refers to the network within a cell or an interaction between genes and the genes within the genome is formed, especially gene regulation (gene regulation) lead between genes effect.
GRN is a mechanism to control gene expression in vivo, the main process of gene expression are transcription translation +
GRN construction method:
Most of the methods used to analyze the gene static data network, such as the gene expression of matrix, which is a time of gene expression. In fact, we need to consider dynamic network, so as to approximate real GRN.
Some network model:
1, Boolean network
Boolean network is one of the most simple model. Boolean network, the status of each gene is only "on" and "off", "on" indicates that the gene expression, "off" indicates that the gene is not expressed. Gene interactions between Boolean expression is represented by: and, or, not, such as A and not B -> C.
The network oversimplification, there are limitations.
2, the linear model is
a linear model is a model of continuous GRN. Linear model, a gene expression level is represented by a weighted number of other genes and the expression level, the right is the quantitative relationship between genes: genes expressed excitation positive weights, negative weight expressed gene suppression, 0 indicates two genes right It does not matter.
X- I (T + [Delta] t) = [sum] w ij of X- J (T) + [eta]
of the network is a simple mathematical model, only processing gene expression data having a linear relationship, a small range of applications.
Related Model: weighting matrix model
3, Markov model
Markov chain is a stochastic process, suitable for the analysis of time series gene expression data. In Markov Model, a Markov chain is assumed that the level of gene expression at a time determines the timing of the next level of gene expression, the following formula:
C (T) = JC (T-. 1)
Construction of GRN process, based on Markov Markov model of gene expression profile of feature extraction and clustering showed good adaptability.
If you want to improve the accuracy of the model can be improved order Markov model.
4, differential equation model
differential equation model of a gene assume a variable, a network of n genes may be represented by an n-dimensional differential equation as follows:
DX I (T) / dt = F I (X . 1 , X 2 , ..., X n- ), X i (T) is the i-th level of gene expression, n is the number of genes represented in the network.
According to different requirements and modeling biological data, determined according to the actual function fi. There are generally two forms:
F I (X J , [Theta] ij of , [alpha]) = X J [alpha] / (X J + [theta] ij of[alpha])
Fi (XJ, [alpha]) = E -αx J / (. 1 + E -αx J )
Construction of Differential GRN advantages: powerful, flexible, complex relationships described beneficial gene networks.
5, a Bayesian network model
to Bayes' theorem, the theoretical basis and assumptions, showing the relationship between the probability of a random variable to form acyclic graph (DAG) used has, in each gene is a network node, the relationship between each of the regulatory It is an edge.
The model can handle random events, control of noise, can be obtained causal relationships between variables, in GRN models, Bayesian networks have an advantage over other models .
Related Model: neural network model, illustrated Gaussian model
6, the correlation model mutual information
mutual information between genes associated correlation model entropy and mutual information is described.
The entropy of a gene expression pattern A 
, P (X I ) is present in the gene expression values xi frequency, n is the number of interval expression levels. The greater the entropy, approaching the level of gene expression randomly distributed.
Mutual information MI (A, B) = H (A) + H (B) -H (A, B), if the MI (A, B) = 0 , the two genes are not related to the gene expression pattern between the two , the closer the relationship between the biological If larger, the two non-random gene-related MI (a, B),. 7, random equation model
......
Some may refer to the experience of building a GRN model: GRN is sparse, that is, not every interaction between all of the genes [3]; most variable gene regulatory networks is continuous, coarse discretization lose information fine discretization too many parameters, preferably directly continuous variable [4];
Ref:
[1]. https://baike.baidu.com/item/%E8%B0%83%E6%8E%A7%E7%BD%91%E7%BB%9C/5844691
[2]. https://wenku.baidu.com/view/34dff5ef19e8b8f67c1cb958.html
[3]. Zhang X, Chong K H, Zheng J, et al. A Monte Carlo method for in silico modeling and visualization of Waddington's epigenetic landscape with intermediate details[J]. bioRxiv, 2018.
. [4] Lei Yao Shan, Shiding Hua, Wang Yifei bioinformatics studies of gene regulatory networks [J] Journal of Nature, 2004 (01): 7-12.