# On the Use of Multivariate Methods for Analysis of Data from Biological Networks

^{1}

^{2}

^{3}

^{4}

^{5}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Preliminary Information

#### 2.1. Univariate Statistical Analysis

#### 2.2. Multivariate Statistical Analysis

**X**, where each sample

**x**is a vector containing a fixed number of measurements. With a two-class problem (again consider the placebo versus treatment example), a subset of these samples

**X**belongs to one class while the remaining subset of samples

_{1}**X**belongs to the other class. The purpose of FDA is to calculate the projection vector

_{2}**w**, which transforms each

**x**to a single score variable t, that best separates the samples in

**X**and

_{1}**X**. Separability is quantified by J, the ratio of the between-class scatter to the within-class scatter, and

_{2}**w**is chosen to maximize this quantity [4]. Figure 1a summarizes this linear transformation performed in FDA as applied to individual samples.

**w**to best separate

**X**and

_{1}**X**, KFDA first applies a nonlinear transformation to each

_{2}**x**, expressed as

**f**= ϕ(

**x**), to map each to a higher-dimensional variable space

**f**. Since the explicit mapping of ϕ(

**x**) is not known, an implicit mapping can be defined such that the inner product between any two ϕ(

**x**) is a Mercer kernel [5]. In a two-class problem, all

**f**belonging to one class make up

**F**while the

_{1}**f**in the other class comprise

**F**. The vector

_{2}**w**that best separates

**F**and

_{1}**F**is then determined, with the linear projection t =

_{2}**w**∙

**f**capturing nonlinear relationships in the original variable space of

**x**. Like FDA, nonlinear KFDA also aims to maximize the value of J. A schematic of the operations involved in KFDA is provided in Figure 1b. It should also be noted that the radial basis function, a commonly-used kernel, will be used in this work.

## 3. Advantages of Multivariate Approaches for Biological Network Analysis

#### 3.1. Advantages of Using Multiple Correlated Measurements for Diagnosis: A General Case

_{in}. Methionine is converted to SAM at a rate v

_{1}by methionine adenosyltransferase enzymes. SAM is then converted to SAH by methyltransferase enzymes at the rate v

_{2}, or is depleted by other reactions and excreted at a rate described by v

_{deplete}. Finally, SAH is converted to other FOCM products at a rate v

_{out}.

_{in}, which can be due to a number of reasons. All modeled metabolite concentrations (methionine, SAM, SAH) will then increase with time, along with their associated reaction rates. Clinical measurement of SAH sometime afterwards will indicate an elevated concentration of SAH, and following scenario (2) or scenario (3) the unwary clinician might conclude that the patient has a decreased SAM/SAH ratio. However, with an additional measurement of SAM it would be discovered that the SAM concentration is also elevated and the SAM/SAH ratio is relatively unchanged. The only way to verify this is to incorporate multiple measurements into the diagnosis and obtain a bigger picture of the network being studied.

#### 3.2. Advantages of Using Multivariate Approaches over Univariate Approaches: Application to ASD Classification Using Clinical Measurements of FOCM/TS Metabolites

#### 3.3. Advantages of Nonlinear Approaches over Linear Approaches: Application to ASD Classification Using Clinical Measurements of Urine Toxic Metals

## 4. Conclusions

## Acknowledgments

## Author Contributions

## Conflicts of Interest

## References

- Frye, R.E.; James, S.J. Metabolic pathology of autism in relation to redox metabolism. Biomark. Med.
**2014**, 8, 321–330. [Google Scholar] [CrossRef] [PubMed] - Morgan, D.B.; Carver, M.E.; Payne, R.B. Plasma creatinine and urea: Creatinine ratio in patients with raised plasma urea. Br. Med. J.
**1977**, 2, 929–932. [Google Scholar] [CrossRef] [PubMed] - Lemieux, I.; Lamarche, B.; Couillard, C.; Pascot, A.; Cantin, B.; Bergeron, J.; Dagenais, G.R.; Després, J.-P. Total cholesterol/HDL cholesterol ratio vs LDL cholesterol/HDL cholesterol ratio as indices of ischemic heart disease risk in men: The Quebec Cardiovascular Study. Arch. Intern. Med.
**2001**, 161, 2685–2692. [Google Scholar] [CrossRef] [PubMed] - Fisher, R.A. The use of multiple measurements in taxonomic problems. Ann. Eugen.
**1936**, 7, 179–188. [Google Scholar] [CrossRef] - Mika, S.; Ratsch, G.; Weston, J.; Scholkopf, B.; Mullers, K.R. Fisher discriminant analysis with kernels. In Proceedings of the 1999 IEEE Neural Networks for Signal Processing IX Workshop, Madison, WI, USA, 23–25 August 1999; pp. 41–48. [Google Scholar]
- Ruxton, G.D. The unequal variance t-test is an underused alternative to Student’s t-test and the Mann–Whitney U test. Behav. Ecol.
**2006**, 17, 688–690. [Google Scholar] [CrossRef] - Mann, H.B.; Whitney, D.R. On a test of whether one of two random variables is stochastically larger than the other. Ann. Math. Stat.
**1947**, 18, 50–60. [Google Scholar] [CrossRef] - Scheffé, H. The Analysis of Variance; John Wiley & Sons: New York, NY, USA, 1999. [Google Scholar]
- Cortes, C.; Vapnik, V. Support-vector networks. Mach. Learn.
**1995**, 20, 273–297. [Google Scholar] [CrossRef] - Johnson, S.C. Hierarchical clustering schemes. Psychometrika
**1967**, 32, 241–254. [Google Scholar] [CrossRef] [PubMed] - Wold, S.; Sjöström, M.; Eriksson, L. PLS-regression: A basic tool of chemometrics. Chemom. Intell. Lab. Syst.
**2001**, 58, 109–130. [Google Scholar] [CrossRef] - Hastie, T.; Tibshirani, R.; Friedman, J. The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Second Edition, 2nd ed.; Springer: New York, NY, USA, 2011. [Google Scholar]
- Appling, D.R. Compartmentation of folate-mediated one-carbon metabolism in eukaryotes. FASEB J.
**1991**, 5, 2645–2651. [Google Scholar] [PubMed] - Anderson, O.S.; Sant, K.E.; Dolinoy, D.C. Nutrition and epigenetics: An interplay of dietary methyl donors, one-carbon metabolism and DNA methylation. J. Nutr. Biochem.
**2012**, 23, 853–859. [Google Scholar] [CrossRef] [PubMed] - Finkelstein, J.D.; Martin, J.J. Homocysteine. Int. J. Biochem. Cell Biol.
**2000**, 32, 385–389. [Google Scholar] [CrossRef] - Vitvitsky, V.; Thomas, M.; Ghorpade, A.; Gendelman, H.E.; Banerjee, R. A functional transsulfuration pathway in the brain links to glutathione homeostasis. J. Biol. Chem.
**2006**, 281, 35785–35793. [Google Scholar] [CrossRef] [PubMed] - Deth, R.; Muratore, C.; Benzecry, J.; Power-Charnitsky, V.-A.; Waly, M. How environmental and genetic factors combine to cause autism: A redox/methylation hypothesis. NeuroToxicology
**2008**, 29, 190–201. [Google Scholar] [CrossRef] [PubMed] - James, S.J.; Melnyk, S.; Jernigan, S.; Cleves, M.A.; Halsted, C.H.; Wong, D.H.; Cutler, P.; Bock, K.; Boris, M.; Bradstreet, J.J.; et al. Metabolic endophenotype and related genotypes are associated with oxidative stress in children with autism. Am. J. Med. Genet. B Neuropsychiatr. Genet.
**2006**, 141, 947–956. [Google Scholar] [CrossRef] [PubMed] - Adams, J.B.; Audhya, T.; McDonough-Means, S.; Rubin, R.A.; Quig, D.; Geis, E.; Gehn, E.; Loresto, M.; Mitchell, J.; Atwood, S.; et al. Nutritional and metabolic status of children with autism vs. neurotypical children, and the association with autism severity. Nutr. Metab.
**2011**, 8, 34. [Google Scholar] [CrossRef] [PubMed] - Melnyk, S.; Fuchs, G.J.; Schulz, E.; Lopez, M.; Kahler, S.G.; Fussell, J.J.; Bellando, J.; Pavliv, O.; Rose, S.; Seidel, L.; et al. Metabolic imbalance associated with methylation dysregulation and oxidative damage in children with autism. J. Autism Dev. Disord.
**2012**, 42, 367–377. [Google Scholar] [CrossRef] [PubMed] - Yi, P.; Melnyk, S.; Pogribna, M.; Pogribny, I.P.; Hine, R.J.; James, S.J. Increase in plasma homocysteine associated with parallel increases in plasma S-adenosylhomocysteine and lymphocyte DNA hypomethylation. J. Biol. Chem.
**2000**, 275, 29318–29323. [Google Scholar] [CrossRef] [PubMed] - Jones, D.P. Redox potential of GSH/GSSG couple: Assay and biological significance. Methods Enzymol.
**2002**, 348, 93–112. [Google Scholar] [PubMed] - Vargason, T.; Howsmon, D.P.; Melnyk, S.; James, S.J.; Hahn, J. Mathematical modeling of the methionine cycle and transsulfuration pathway in individuals with autism spectrum disorder. J. Theor. Biol.
**2017**, 416, 28–37. [Google Scholar] [CrossRef] [PubMed] - Howsmon, D.P.; Kruger, U.; Melnyk, S.; James, S.J.; Hahn, J. Classification and adaptive behavior prediction of children with autism spectrum disorder based upon multivariate data analysis of markers of oxidative stress and DNA methylation. PLoS Comput. Biol.
**2017**, 13, e1005385. [Google Scholar] [CrossRef] [PubMed] - Kohavi, R. A study of cross-validation and bootstrap for accuracy estimation and model selection. In Proceedings of the 14th International Joint Conference on Artificial Intelligence, Montreal, QC, Canada, 20–25 August 1995; Morgan Kaufmann Publishers Inc.: San Francisco, CA, USA, 1995; Volume 2, pp. 1137–1143. [Google Scholar]
- Adams, J.B.; Howsmon, D.P.; Kruger, U.; Geis, E.; Gehn, E.; Fimbres, V.; Pollard, E.; Mitchell, J.; Ingram, J.; Hellmers, R.; et al. Significant association of urinary toxic metals and autism-related symptoms—A nonlinear statistical analysis with cross validation. PLoS ONE
**2017**, 12, e0169526. [Google Scholar] [CrossRef] [PubMed] - Adams, J.B.; Audhya, T.; McDonough-Means, S.; Rubin, R.A.; Quig, D.; Geis, E.; Gehn, E.; Loresto, M.; Mitchell, J.; Atwood, S.; et al. Toxicological status of children with autism vs. neurotypical children and the association with autism severity. Biol. Trace Elem. Res.
**2012**, 151, 171–180. [Google Scholar] [CrossRef] [PubMed] - Rossignol, D.A.; Genuis, S.J.; Frye, R.E. Environmental toxicants and autism spectrum disorders: A systematic review. Transl. Psychiatry
**2014**, 4, e360. [Google Scholar] [CrossRef] [PubMed]

**Figure 1.**Schematics of the transformations used in Fisher Discriminant Analysis (FDA) and Kernel Fisher Discriminant Analysis (KFDA): (

**a**) In FDA, the dot product of vector

**w**with data sample

**x**is calculated to obtain the projected value t; (

**b**) KFDA first maps each sample

**x**to a higher-dimensional space

**f**according to the nonlinear transformation ϕ(

**x**). The dot product of

**w**with

**f**(rather than with

**x**) is then calculated to obtain the projection t.

**Figure 2.**Diagram of major metabolites and reactions involved in the folate-dependent one-carbon metabolism (FOCM) and transsulfuration (TS) pathways. DNA methylation plays an important role in epigenetics and glutathione (GSH) is responsible for the clearance of environmental toxins.

**Figure 3.**A simplified representation of a subset of reactions in FOCM responsible for DNA methylation.

**Figure 4.**Probability density functions (PDFs) of five measurements for ASD and NT cohorts from the IMAGE study: (

**a**) % DNA methylation; (

**b**) 8-hydroxyguanosine; (

**c**) glutamylcysteine; (

**d**) free cystine/free cysteine; (

**e**) % oxidized glutathione. These PDFs are based on the standardized values of each measurement (i.e., all samples for a measurement are scaled such that the mean value is 0 and the standard deviation is 1).

**Figure 5.**Multivariate analysis with FDA using five measurements from the IMAGE study (% DNA methylation, 8-hydroxyguanosine, glutamylcysteine, free cystine/free cysteine, and % oxidized glutathione). The scores are the projected values obtained by leave-one-out cross-validation with FDA, while the PDFs were obtained by fitting to the scores. The shown threshold corresponds to a Type I error of 4.8% and a Type II error of 5%.

**Figure 6.**Results of classification using linear FDA with three urine toxic metal measurements (aluminum, cesium, tungsten) as inputs. FDA scores were from leave-one-out cross-validation and the PDFs were obtained by fitting to the scores. The Type I and Type II errors are both 50%.

**Figure 7.**Results of classification using nonlinear KFDA with three urine toxic metal measurements (aluminum, cesium, tungsten) as inputs. KFDA scores were from leave-one-out cross-validation and the PDFs were obtained by fitting to the scores. The corresponding Type I and Type II errors are 29% and 28%, respectively.

**Table 1.**Means and standard deviations of five FOCM/TS measurements for the autism spectrum disorder (ASD) and neurotypical (NT) cohorts from the Integrated Metabolic and Genomic Endeavor (IMAGE) study. Reported p-values were obtained from the two-tailed Welch’s t-test.

Measurement | ASD Mean ± SD | NT Mean ± SD | p-Value |
---|---|---|---|

n = 83 | n = 76 | ||

% DNA methylation | 3.37 ± 0.87 | 4.26 ± 0.90 | <0.001 |

8-hydroxyguanosine (pmol/mg DNA) | 89.2 ± 27.9 | 56.7 ± 17.9 | <0.001 |

glutamylcysteine (µM) | 1.87 ± 0.46 | 2.37 ± 0.59 | <0.001 |

free cystine/free cysteine | 1.51 ± 0.58 | 1.06 ± 0.35 | <0.001 |

% oxidized glutathione | 0.22 ± 0.07 | 0.12 ± 0.04 | <0.001 |

**Table 2.**Means and standard deviations of levels of three urine toxic metals in the ASD and NT cohorts from the Comprehensive Nutritional and Dietary Intervention Study. Metal levels are in units of µg/g of creatinine. Reported p-values were obtained from the two-tailed Welch’s t-test.

Measurement | ASD Mean ± SD | NT Mean ± SD | p-Value |
---|---|---|---|

n = 67 | n = 50 | ||

Aluminum | 9.03 ± 6.55 | 8.55 ± 11.15 | n.s. |

Cesium | 4.03 ± 1.92 | 3.74 ± 1.75 | n.s. |

Tungsten | 0.29 ± 0.25 | 0.29 ± 0.21 | n.s. |

© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Vargason, T.; Howsmon, D.P.; McGuinness, D.L.; Hahn, J.
On the Use of Multivariate Methods for Analysis of Data from Biological Networks. *Processes* **2017**, *5*, 36.
https://doi.org/10.3390/pr5030036

**AMA Style**

Vargason T, Howsmon DP, McGuinness DL, Hahn J.
On the Use of Multivariate Methods for Analysis of Data from Biological Networks. *Processes*. 2017; 5(3):36.
https://doi.org/10.3390/pr5030036

**Chicago/Turabian Style**

Vargason, Troy, Daniel P. Howsmon, Deborah L. McGuinness, and Juergen Hahn.
2017. "On the Use of Multivariate Methods for Analysis of Data from Biological Networks" *Processes* 5, no. 3: 36.
https://doi.org/10.3390/pr5030036