Difference between revisions of "Team:Jiangnan China/Model"

    Before model, you should have a strain with corresponding capability or function. In our project, we got the anti-acid strain L.lactis by the following steps:

    We firstly mutant our parent strain L.lactis NZ9000, and 3 key mutant strains were screened from 35,000 mutant strains under high throughput screening, namely L.lactis WH101, WH102, and WH103.
    And then we did acid stress analysis on these three mutant strains.

    Figure 1 The survival rate of 4 strains, L.lactis NZ9000, L.lactis WH101, L.lactis WH102, L.lactis WH103. On the left it’s colony distribution of parent strain and acid-tolerant strains under a pH of 4.0 stress gradient of 10-3, and on the right it’s the survival rate of parent strain and acid tolerant strain (pH 4.0).

    Figure 2 The growth curve of 4 strains, L.lactis NZ9000, L.lactis WH101, L.lactis WH102, L.lactis WH103. A: pH 7.0, B: pH 5.0, C: pH 4.5.

    From figure 1 and figure 2, we screened out L.lactis WH101, which has remarkably 16000-fold higher survival rate than the parent strain at pH4.0 for 5h, which is the highest among the reported survival rates at the same condition.
    Then we start our model to find key anti-acid component.

    We then perform PCA on the data of the 61 common deferential genes.
    Statistically, PCA is one of the most widely used data compression algorithms. In PCA, data is converted from the original coordinate system to the new coordinate system, which is determined by the data itself. When converting the coordinate system, the direction with the largest variance is taken as the direction of the coordinate axis, because the maximum variance of the data gives the most important information of the data. The first new axis selects the method with the largest variance in the original data, and the second new axis selects the direction orthogonal to the first new coordinate axis and the second largest variance. This process is repeated and the number of repetitions is the feature dimension of the original data.
    The specific code that converts the data into feature spaces that retain only the first N principal components is as follows:

from numpy import *

def loadDataSet(filename,delim='\t')
        fr=open(filename)
        stringArr=[line,strip().split(delim) for line in fr.readlines()]
        datArr=[map(float.line)for line in stringArr]
        return mat(datArr)

def pca(dataMat,topNfeat=4096):
        meanVals=mean(dataMat,axis=0)
        meanRemoved=dataMat-meanVals
        covMat=cov(meanRemoved,rowvar=0)
        eigVals,eigVects=linalg.eig(mat(conMat))
        eigValInd=argsort(eigVals)
        eigValInd=eigValInd[:-(topNfeat+1):-1]
        redEigVects=eigVects[:,eigValInd]
        lowDDataMat=meanRemoved*redEigVects
        reconMat=(lowDDataMat*redEigVects.T)+meanVals
        return lowDDataMat,reconMat

    Click here for more details about PCA:    https://en.wikipedia.org/wiki/Principal_component_analysis

    Figure 4 The proportion of each data in the dimension under the two principal components. The data here are only taken from the top 10 of 61 common differential genes.

    PCA sorted the data after analysis, and the top 10 are showed here because 61 data are too much. The first five genes with the proportion greater than 0.2 are experimentally verified. The experimental results show that LLNZ_RS02280 (msmK) has the best anti-acid capability. This is almost identical to the direct experimental validation from 61 common differential genes.
    With these two principal components, we improved the accuracy of the result to 0.98. Although LLNA_RS02280 (msmK) is not the first in the model, it is in the top five, which shows that the model still has great reference.