Team:Worldshaper-Shanghai/Model

Model

To establish the modeling, we got 66 sample of red fluorescence protein by adding different concentration of uric acid to the cellulose acetate membrane that was reformed by gene pHucO and protein HucR-mCherry. we prepared before. The gradient is from 0 to 1000 ordinally. The initial plan was to distinguish the difference between color by eye, but we found that the color change is too weak for human eye to see. So we tried to take photo under florescence capture device and got the following pictures. Next step is to distinguish the color difference of pictures, not by eye which is neither precise nor a good stander for our detection, by ImageJ, an professional image processing program(http://imagej.nih.gov/ij). We got 66 values measured by ImageJ represent difference between the color roughly. Here I said roughly because the color in each picture is microscopical different and one digital data cannot describe the whole image.

Here are the data:

        	y       x1
        1	30.500	0
        2	31.236	0
        3	34.310	0
        4	27.518	0
        5	26.920	0
        6	29.869	0
        7	28.012	100
        8	28.101	100
        9	27.680	100
        10	20.794	100
        11	24.336	100
        12	23.726	100
        13	29.198	200
        14	29.628	200
        15	29.358	200
        16	29.045	200
        17	27.957	200
        18	29.402	200
        19	28.649	300
        20	27.674	300
        21	28.695	300
        22	37.169	300
        23	31.894	300
        24	33.117	300
        25	27.122	400
        26	26.896	400
        27	27.227	400
        28	25.331	400
        29	26.588	400
        30	24.775	400
        31	28.189	450
        32	27.598	450
        33	27.058	450
        34	29.104	450
        35	29.706	450
        36	28.957	450
        37	26.281	500
        38	26.199	500
        39	27.194	500
        40	29.398	500
        41	29.347	500
        42	28.689	500
        43	29.446	600
        44	28.042	600
        45	28.529	600
        46	28.893	600
        47	27.133	600
        48	27.104	600
        49	30.038	700
        50	29.912	700
        51	29.301	700
        52	29.278	700
        53	27.765	800
        54	27.311	800
        55	29.422	800
        56	28.589	800
        57	27.449	800
        58	27.966	800
        59	21.332	900
        60	20.389	900
        61	21.551	900
        62	19.390	900
        63	26.186	900
        64	25.612	1000
        65	25.314	1000
        66	24.015	1000
    

During the final process, we need to use R program to establish the above table. So, under the direction of professor, we successfully learned how to use R program to establish the table here is the code we used.

        setwd('~/Desktop/IGEM/')
        df=read.table('test.txt',head=T)
        head(df)
        df
        library('ggpubr')
            
        plot(df$y,df$x1)
        head(df)
        library(ggpubr)
        ggscatter(df, x = "x1", y = "y",
            color = "#00AFBB", shape = 19, size = 3,
            # Points color, shape and size
            add = "reg.line",  # Add regressin line
            add.params = list(color = "black", fill = "lightgray"), # Customize reg. line
            conf.int = TRUE, # Add confidence interval
            cor.coef = TRUE, # Add correlation coefficient. see ?stat_cor
            ylab='Indensity',xlab='Dilution ratio',
            cor.coeff.args = list(method = "pearson", label.x = 300, label.sep = "\n")
        )
    

But there is a problem: Although fluorescence value is significantly correlated with uric acid concentration(p=0.0016), the correlation coefficient is relatively low, only about -0.38.(showed in picture)

Figure 1

We assumed that maybe some of the data’s deviation is still caused by the vary in color in different position of the picture. For example, in this picture, the light part is relatively nonuniform, lighter up and darker down, resulting the mistake calculating.

Figure 4

Two strategies were used to enhance the correlation coefficient.

Strategy1: Cut the initial rectangle picture into square to find the uniform part. However, the result isn’t significantly improved, we even got smaller correlation coefficient than before.(showed in picture below)

Figure 2

Strategy 2: In order to further eliminate the uneven part, we cut the picture in circle and let the program read only the color inside the circle, we hoped the data may be more uniform. However, the correlation coefficient we got is -0.37 this time, pretty similar to what we got at first time.

Figure 5

Consequently, we analyzed the similar three results and concluded that the deviation is caused by the small sample size.