Genetic and Environmental Factors in ME/CFS

Chronic fatigue syndrome (CFS) is a debilitating illness characterized by multiple unexplained symptoms including fatigue, cognitive impairment and pain. People with CFS have no characteristic physical signs or diagnostic laboratory abnormalities, and the etiology and pathophysiology remain unknown.

CFS represents a complex illness that includes alterations in homeostatic systems, involves multiple body systems and results from the combined action of many genes, environmental factors and risk-conferring behavior. In order to achieve understanding of complex illnesses, such as CFS, studies must collect relevant epidemiological, clinical and laboratory data and then integrate, analyze and interpret the information so as to obtain meaningful clinical and biological insight.

This issue of Pharmacogenomics represents such an approach to CFS. Data was collected during a 2-day in-hospital study of persons with CFS, other medically and psychiatrically unexplained fatiguing illnesses and nonfatigued controls identified from the general population of Wichita, KS, USA.

While in the hospital, the participants' psychiatric status, sleep characteristics and cognitive functioning was evaluated, and biological samples were collected to measure neuroendocrine status, autonomic nervous system function, systemic cytokines and peripheral blood gene expression. The data generated from these assessments was made available to a multidisciplinary group of 20 investigators from around the world who were challenged with revealing new insight and algorithms for integration of this complex, high-content data and, if possible, identifying molecular markers and elucidating pathophysiology of chronic fatigue.

The group was divided into four teams with representation from the disciplines of medicine, mathematics, biology, engineering and computer science. The papers in this issue are the culmination of this 6-month challenge, and demonstrate that data integration and multidisciplinary collaboration can indeed yield novel approaches for handling large, complex datasets, and reveal new insight and relevance to a complex illness such as CFS.

Objectives: To test the hypothesis that medically unexplained chronic fatigue and chronic fatigue syndrome (CFS) are heterogeneous conditions, and to define the different conditions using both symptom and laboratory data.

Methods: We studied 159 women from KS, USA. A total of 51 of these suffered from fatigue consistent with established criteria for CFS, 55 had chronic fatigue of insufficient symptoms/severity for a CFS diagnosis and 53 were healthy controls matched by age and body mass index (BMI) against those with CFS. We used principal components analyses to define factors that best described the variable space and to reduce the number of variables. The 38 most explanatory variables were then used in latent class analyses to define discrete subject groups.

Results: Principal components analyses defined six discrete factors that explained 40% of the variance. Latent class analyses provided several interpretable solutions with four, five and six classes. The four-class solution was statistically most convincing, but the six-class solution was more interpretable. Class 1 defined 41 (26%) subjects with obesity and relative sleep hypnoea. Class 2 were 38 (24%) healthy subjects. Class 3 captured 24 (15%) obese relatively hypnoeic subjects, but with low heart rate variability and cortisol. Class 4 were 23 (14%) sleep-disturbed and myalgic subjects without obesity or significant depression. The two remaining classes with 22 (14%) and 11 (7%) subjects consisted of the most symptomatic and depressed, but without obesity or hypnoea. Class 5 had normal sleep indices. Class 6 was characterized by disturbed sleep, with low sleep heart rate variability, cortisol, and sex hormones.

Conclusion: Chronic medically unexplained fatigue is heterogeneous. The putative syndromes were differentiated by obesity, sleep hypnoea, depression, physiological stress response, sleep disturbance, interoception and menopausal status. If these syndromes are externally validated and replicated, they may prove useful in determining the causes, pathophysiology and treatments of CFS.

Objectives: To validate a latent class structure derived empirically from a clinical data set obtained from persons with chronic medically unexplained fatigue.

Methods: The strategies utilized in this validation study included: recalculating latent class analysis (LCA) results varying random seeds and the number of initial random starting sets; recalculating LCA results by substituting alternate variables to demonstrate a robust solution; determining the statistical significance of between-class differences on disability, fatigue and demographic measures omitted from the data set used for LCA; cross-classifying class membership using established Centers for Disease Control and Prevention (CDC) research criteria for chronic fatigue syndrome (CFS) to compare the relative proportions of subjects designated CFS, chronic fatigue (not CFS) or healthy controls captured by the latent classes.

Results: Recalculation of results and substitution of variables for low-loading variables demonstrated a robust LCA result. Highly significant between-class differences were confirmed between Class 2 (well) and those interpreted as ill/fatigued. Analysis of between-class differences for the fatigue groups revealed significant differences for all disability and fatigue variables, but with equivalent levels of reported activity and reduction in motivation. Cross-classification against established CDC criteria demonstrated that 89% of subjects constituting Class 2 (well) were indeed nonfatigued controls. A general tendency for grouping CFS cases in the multiple symptomatic classes was noted.

Conclusion: This study established reasonably good validity for an empirically-derived latent class solution reflecting considerable heterogeneity among subjects with medically unexplained chronic fatigue. This work strengthens the growing understanding of CFS as a heterogeneous entity comprised of several conditions with different underlying pathophysiological mechanisms.

Gene expression profile of empirically delineated classes of unexplained chronic fatigue.

Journal: Pharmacogenomics. 2006 Apr;7(3):375-386.

Authors: Liran Carmel¹, Sol Efroni², Peter D White³, Eric Aslakson⁴, Uté Vollmer-Conna⁵ & Mangalathu S Rajeevan^4,*

Affiliations: [1] National Center for Biotechnology Information, National Library of Medicine, National Institutes of Health, Bethesda, Maryland, USA
[2] National Cancer Institute Center for Bioinformatics, National Institutes of Health, Bethesda, Maryland, USA
[3] University of London, Department of Psychological Medicine, Barts, London and Queen Mary School of Medicine and Dentistry, London, UK
[4] Centers for Disease Control and Prevention, 1600 Clifton Road, MSG 41, Atlanta, GA 30333, USA. mor4@cdc.gov
[5] University of New South Wales, School of Psychiatry, Sydney, Australia
[*] Author for correspondence

NLM Citation: PMID: 16610948

Objectives: To identify the underlying gene expression profiles of unexplained chronic fatigue subjects classified into five or six class solutions by principal component (PCA) and latent class analyses (LCA).

Methods: Microarray expression data were available for 15,315 genes and 111 female subjects enrolled from a population-based study on chronic fatigue syndrome. Algorithms were developed to assign gene scores and threshold values that signified the contribution of each gene to discriminate the multiclasses in each LCA solution. Unsupervised dimensionality reduction was first used to remove noise or otherwise uninformative gene combinations, followed by supervised dimensionality reduction to isolate gene combinations that best separate the classes.

Results: The authors' gene score and threshold algorithms identified 32 and 26 genes capable of discriminating the five and six multiclass solutions, respectively. Pair-wise comparisons suggested that some genes (zinc finger protein 350 [ZNF350], solute carrier family 1, member 6 [SLC1A6], F-box protein 7 [FBX07] and vacuole 14 protein homolog [VAC14]) distinguished most classes of fatigued subjects from healthy subjects, whereas others (patched homolog 2 [PTCH2] and T-cell leukemia/lymphoma [TCL1A]) differentiated specific fatigue classes.

Conclusion: A computational approach was developed for general use to identify discriminatory genes in any multiclass problem. Using this approach, differences in gene expression were found to discriminate some classes of unexplained chronic fatigue, particularly one termed interoception.

Polymorphisms in genes regulating the HPA axis associated with empirically delineated classes of unexplained chronic fatigue.

Journal: Pharmacogenomics. 2006 Apr;7(3):387-394.

Authors: Alicia K Smith¹, Peter D White², Eric Aslakson¹, Uté Vollmer-Conna³ & Mangalathu S Rajeevan^1,*

Affiliations:
[1] Centers for Disease Control and Prevention, Division of Viral and Rickettsial Diseases, National Center for Infectious Diseases, 1600 Clifton Road, MSG41, Atlanta, GA 30333, USA. mor4@cdc.gov
[2] University of London, Department of Psychological Medicine, Barts, London and Queen Mary School of Medicine and Dentistry, London, UK
[3] University of New South Wales, School of Psychiatry, Sydney, Australia
[*] Author for correspondence

NLM Citation: PMID: 16610949

Chronic fatigue syndrome (CFS) is characterized by persistent or relapsing fatigue that is not alleviated by rest, causes substantial reduction in activities and is accompanied by a variety of symptoms. Its unknown etiology may reflect that CFS is heterogeneous.

Latent class analyses of symptoms and physiological systems were used to delineate subgroups within a population-based sample of fatigued and nonfatigued subjects [1] . This study examined whether genetic differences underlie the individual subgroups of the latent class solution.

Polymorphisms in 11 candidate genes related to both hypothalamic-pituitary-adrenal (HPA) axis function and mood-related neurotransmitter systems were evaluated by comparing each of the five ill classes (Class 1, n = 33; Class 3, n = 22; Class 4, n = 22; Class 5, n = 17; Class 6, n = 11) of fatigued subjects with subjects defined as well (Class 2, n = 35).

Of the five classes of subjects with unexplained fatigue, three classes were distinguished by gene polymorphsims involved in either HPA axis function or neurotransmitter systems, including proopiomelanocortin (POMC), nuclear receptor subfamily 3, group C, member 1 (NR3C1), monoamine oxidase A (MAOA), monoamine oxidase B (MAOB), and tryptophan hydroxylase 2 (TPH2).

These data support the hypothesis that medically unexplained chronic fatigue is heterogeneous and presents preliminary evidence of the genetic mechanisms underlying some of the putative conditions.

Gene expression correlates of unexplained fatigue.

Journal: Pharmacogenomics . 2006 Apr;7(3):395-405.

Authors: Toni Whistler^1,*, Renee Taylor², R Cameron Craddock¹, Gordon Broderick³, Nancy Klimas⁴ & Elizabeth R Unger¹

Affiliations:
[1] Centers for Disease Control and Prevention, Viral Exanthems and Herpesvirus Branch, Atlanta, GA, 30333, USA. taw6@cdc.gov
[2] University of Illinois at Chicago, Department of Occupational Therapy, Chicago, IL, 60612, USA. rtaylor@uic.edu
[3] University of Alberta, Institute for Biomolecular Design, Edmonton, Alberta, T6G 2H7, Canada. gordon.broderick@ualberta.ca
[4] Miami Veterans Affairs Medical Center, Miami, FL, 33125, USA. Nancy.klimas@va.gov
[*] Author for correspondence

NLM Citation: PMID: 16610950

Quantitative trait analysis (QTA) can be used to test whether the expression of a particular gene significantly correlates with some ordinal variable. To limit the number of false discoveries in the gene list, a multivariate permutation test can also be performed.

The purpose of this study is to identify peripheral blood gene expression correlates of fatigue using quantitative trait analysis on gene expression data from 20,000 genes and fatigue traits measured using the multidimensional fatigue inventory (MFI).

A total of 839 genes were statistically associated with fatigue measures. These mapped to biological pathways such as oxidative phosphorylation, gluconeogenesis, lipid metabolism, and several signal transduction pathways.

However, more than 50% are not functionally annotated or associated with identified pathways. There is some overlap with genes implicated in other studies using differential gene expression.

However, QTA allows detection of alterations that may not reach statistical significance in class comparison analyses, but which could contribute to disease pathophysiology.

This study supports the use of phenotypic measures of chronic fatigue syndrome (CFS) and QTA as important for additional studies of this complex illness. Gene expression correlates of other phenotypic measures in the CFS Computational Challenge (C3) data set could be useful. Future studies of CFS should include as many precise measures of disease phenotype as is practical.

Identifying illness parameters in fatiguing syndromes using classical projection methods.

Journal: Pharmacogenomics. 2006 Apr;7(3):407-419.

Authors: Gordon Broderick^1,*, R Cameron Craddock ², Toni Whistler ², Renee Taylor³, Nancy Klimas⁴ & Elizabeth R Unger²

Affiliations:
[1] University of Alberta, Institute for Biomolecular Design, Edmonton, Alberta, T6G 2H7, Canada. gordon.broderick@ualberta.ca
[2] Centers for Disease Control and Prevention, Viral Exanthems and Herpesvirus Branch, Atlanta, GA, 30333, USA
[3] University of Illinois at Chicago, Department of Occupational Therapy, Chicago, IL, 60612, USA
[4] University of Miami, Miami Veterans Affairs Medical Center, Miami, FL, 33125, USA
[*] Author for correspondence

NLM Citation: PMID: 16610951

Objectives: To examine the potential of multivariate projection methods in identifying common patterns of change in clinical and gene expression data that capture the illness state of subjects with unexplained fatigue and nonfatigued control participants.

Methods: Data for 111 female subjects was examined. A total of 59 indicators, including multidimensional fatigue inventory (MFI), medical outcome Short Form 36 (SF-36), Centers for Disease Control and Prevention (CDC) symptom inventory and cognitive response described illness. Partial least squares (PLS) was used to construct two feature spaces: one describing the symptom space from gene expression in peripheral blood mononuclear cells (PBMC) and one based on 117 clinical variables. Multiplicative scatter correction followed by quantile normalization was applied for trend removal and range adjustment of microarray data. Microarray quality was assessed using mean Pearson correlation between samples. Benjamini-Hochberg multiple testing criteria served to identify significantly expressed probes.

Results: A single common trend in 59 symptom constructs isolates of nonfatigued subjects from the overall group. This segregation is supported by two co-regulation patterns representing 10% of the overall microarray variation. Of the 39 principal contributors, the 17 probes annotated related to basic cellular processes involved in cell signaling, ion transport and immune system function. The single most influential gene was sestrin 1 (SESN1), supporting recent evidence of oxidative stress involvement in chronic fatigue syndrome (CFS). Dominant variables in the clinical feature space described heart rate variability (HRV) during sleep. Potassium and free thyroxine (T4) also figure prominently.

Conclusion: Combining multiple symptom, gene or clinical variables into composite features provides better discrimination of the illness state than even the most influential variable used alone. Although the exact mechanism is unclear, results suggest a common link between oxidative stress, immune system dysfunction and potassium imbalance in CFS patients leading to impaired sympatho-vagal balance strongly reflected in abnormal HRV.

Exploration of statistical dependence between illness parameters using the entropy correlation coefficient.

Journal: Pharmacogenomics. 2006 Apr;7(3):421-428.

Authors: R Cameron Craddock^1,*, Renee Taylor², Gordon Broderick³, Toni Whistler¹, Nancy Klimas⁴ & Elizabeth R Unger¹

Affiliations:
[1] Centers for Disease Control and Prevention, Viral Exanthems and Herpesvirus Branch, Atlanta, GA, 30333, USA. cmi5@cdc.gov
[2] University of Illinois at Chicago, Department of Occupational Therapy, Chicago, IL, 60612, USA. rtaylor@uic.edu
[3] University of Alberta, Institute for Biomolecular Design, Edmonton, Alberta, T6G 2H7, Canada. gordon.broderick@ualberta.ca
[4] University of Miami/Miami Veterans Affairs Medical Center, Miami, FL, 33125, USA. Nancy.klimas@va.gov
[*] Author for correspondence

NLM Citation: PMID: 16610952

The entropy correlation coefficient (ECC) is a useful tool for measuring statistical dependence between variables. We employed this tool to search for pairs of variables that correlated in the chronic fatigue syndrome (CFS) Computational Challenge dataset. Highly related variables are candidates for data reduction, and novel relationships could lead to hypotheses regarding the pathogenesis of CFS.

Methods: Data for 130 female participants in the Wichita (KS, USA) clinical study [1] was coded into numerical values. Metric data was grouped using Gaussian mixture models; the number of groups was chosen using Bayesian information content. The pair-wise correlation between all variables was computed using the ECC. Significance was estimated from 1000 iterations of a permutation test and a threshold of 0.01 was used to identify significantly correlated variables.

Results: The five dimensions of multidimensional fatigue inventory (MFI) were all highly correlated with each other. Seven Short Form (SF)-36 measures, four CFS case-defining symptoms and the Zung self-rating depression scale all correlated with all MFI dimensions. No physiological variables correlate with more than one MFI dimension. MFI, SF-36, CDC symptom inventory, the Zung self-rating depression scale and three Cambridge Neuropsychological Test Automated Battery (CANTAB) measures are highly correlated with CFS disease status.

Discussion: Correlations between the five dimensions of MFI are expected since they are measured from the same instrument. The relationship between MFI and Zung depression index has been previously reported. MFI, SF-36, and Centers for Disease Control and Prevention (CDC) symptom inventory are used to classify CFS; it is not surprising that they are correlated with disease status. Only one of the three CANTAB measures that correlate with disease status has been previously found, indicating the ECC identifies relationships not found with other statistical tools.

Conclusion: The ECC is a useful tool for measuring statistical dependence between variables in clinical and laboratory datasets. The ECC needs to be further studied to gain a better understanding of its meaning for clinical data.

Gene expression profile exploration of a large dataset on chronic fatigue syndrome.

Journal: Pharmacogenomics. 2006 Apr;7(3):429-440.

Authors: Hong Fang¹, Qian Xie¹, Roumiana Boneva², Jennifer Fostel³, Roger Perkins¹ & Weida Tong^4,*

Affiliations:
[1] Z-Tech Corporation at NCTR, Division of Bioinformatics, 3900 NCTR Road, Jefferson, AR 72079, USA
[2] Centers for Disease Control and Prevention, Atlanta, GA 30333, USA
[3] Alpha-Gamma Technologies Incorporation, Suite 350, 4700 Falls of Neuse Road, Raleigh, NC, 27609, USA
[4] National Center for Toxicological Research (NCTR), Center for Toxicoinformatics, Food and Drug Administration, HFT-020, 3900 NCTR Road, Jefferson, Arkansas 72079, USA. wtong@nctr.fda.gov
[*] Author for correspondence

NLM Citation: PMID: 16610953

Objective: To gain understanding of the molecular basis of chronic fatigue syndrome (CFS) through gene expression analysis using a large microarray data set in conjunction with clinically administrated questionnaires.

Method: Data from the Wichita (KS, USA) CFS Surveillance Study was used, comprising 167 participants with two self-report questionnaires (multidimensional fatigue inventory [MFI] and Zung depression scale [Zung]), microarray data, empiric classification, and others. Microarray data was analyzed using bioinformatics tools from ArrayTrack.

Results: Correspondence analysis was applied to the MFI questionnaire to select the 23 samples having either the most or the least fatigue, and to the Zung questionnaire to select the 26 samples having either the most or least depression; ten samples were common, resulting in a total of 39 samples. The MFI and Zung-based CFS/non-CFS (NF) classifications on the 39 samples were consistent with the empiric classification. Two differentially-expressed gene lists were determined, 188 fatigue-related genes and 164 depression-related genes, which shared 24 common genes and involved 11 common pathways. Principal component analysis based on 24 genes clearly separates 39 samples with respect to their likelihood to be CFS. Most of the 24 genes are not previously reported for CFS, yet their functions are consistent with the prevailing model of CFS, such as immune response, apoptosis, ion channel activity, signal transduction, cell-cell signaling, regulation of cell growth and neuronal activity. Hierarchical cluster analysis was performed based on 24 genes to classify 128 (=167-39) unassigned samples. Several of the 11 identified common pathways are supported by earlier findings for CFS, such as cytokine-cytokine receptor interaction and neuroactive ligand-receptor interaction. Importantly, most of the 11 common pathways are interrelated, suggesting complex biological mechanisms associated with CFS.

Conclusion: Bioinformatics is critical in this study to select definitive sample groups, analyze gene expression data and gain insight into biological mechanisms. The 24 identified common genes and 11 common pathways could be important in future studies of CFS at the molecular level.

Objective: To identify biomarkers of chronic fatigue syndrome (CFS) and related disorders through analysis of microarray data, pathology test results and self-report symptom profiles.

Method: To empirically derive the symptom domains of the illnesses, factor analysis was performed on responses to self-report questionnaires (multidimensional fatigue inventory, Centers for Disease Control and Prevention (CDC) symptom inventory and Zung depression scale) before validation with independent datasets. Gene expression patterns that distinguished subjects across each factor dimension were then sought.

Results: A four-factor solution was favored, featuring 'fatigue' and 'mood disturbance' factors. Scores on these factors correlated with measures of disability on the Short Form (SF)-36. A total of 57 genes that distinguished subjects along each factor dimension were identified, although the separation was significant only for subjects beyond the extreme (15(th) and 85(th)) percentiles of severity. Clustering of laboratory parameters with expression of these genes revealed associations with serum measurements of pH, electrolytes, glucose, urea, creatinine, and liver enzymes (aspartate amino transferase [AST] and alanine amino transferase [AST]); as well as hematocrit and white cell count.

Conclusion: CFS is a complex syndrome that cannot simply be associated with changes in individual laboratory tests or expression levels of individual genes. No clear association with gene expression and individual symptom domains was found. However, analysis of such multifacetted datasets is likely to be an important means to elucidate the pathogenesis of CFS.

Objectives: To provide a mathematical introduction to the Wichita (KS, USA) clinical dataset, which is all of the nongenetic data (no microarray or single nucleotide polymorphism data) from the 2-day clinical evaluation, and show the preliminary findings and limitations, of popular, matrix algebra-based data mining techniques.

Methods: An initial matrix of 440 variables by 227 human subjects was reduced to 183 variables by 164 subjects. Variables were excluded that strongly correlated with chronic fatigue syndrome (CFS) case classification by design (for example, the multidimensional fatigue inventory [MFI] data), that were otherwise self reporting in nature and also tended to correlate strongly with CFS classification, or were sparse or nonvarying between case and control. Subjects were excluded if they did not clearly fall into well-defined CFS classifications, had comorbid depression with melancholic features, or other medical or psychiatric exclusions. The popular data mining techniques, principle components analysis (PCA) and linear discriminant analysis (LDA), were used to determine how well the data separated into groups. Two different feature selection methods helped identify the most discriminating parameters.

Results: Although purely biological features (variables) were found to separate CFS cases from controls, including many allostatic load and sleep-related variables, most parameters were not statistically significant individually. However, biological correlates of CFS, such as heart rate and heart rate variability, require further investigation.

Conclusions: Feature selection of a limited number of variables from the purely biological dataset produced better separation between groups than a PCA of the entire dataset. Feature selection highlighted the importance of many of the allostatic load variables studied in more detail by Maloney and colleagues in this issue [1] , as well as some sleep-related variables. Nonetheless, matrix linear algebra-based data mining approaches appeared to be of limited utility when compared with more sophisticated nonlinear analyses on richer data types, such as those found in Maloney and colleagues [1] and Goertzel and colleagues [2] in this issue.

Study population: We examined the relationship between chronic fatigue syndrome (CFS) and allostatic load in a population-based, case-control study of 43 CFS patients and 60 nonfatigued, healthy controls from Wichita, KS, USA.

Methods: An allostatic load index was computed for all study participants using available laboratory and clinical data, according to a standard algorithm for allostatic load. Logistic regression analysis was used to compute odds ratios (ORs) as estimates of relative risk in models that included adjustment for matching factors and education; 95% confidence intervals (CIs) were computed to estimate the precision of the ORs.

Results: CFS patients were 1.9-times more likely to have a high allostatic load index than controls (95% CI = 0.75, 4.75) after adjusting for education level, in addition to matching factors. The strength of this association increased in a linear trend across categories of low, medium and high levels of allostatic load (p = 0.06).

Conclusion: CFS was associated with a high level of allostatic load. The three allostatic load components that best discriminated cases from controls were waist:hip ratio, aldosterone and urinary cortisol.

Objective: This paper asks whether the presence of chronic fatigue syndrome (CFS) can be more accurately predicted from single nucleotide polymorphism (SNP) profiles than would occur by chance.

Methods: Specifically, given SNP profiles for 43 CFS patients, together with 58 controls, we used an enumerative search to identify an ensemble of conjunctive rules that predict whether a patient has CFS.

Results: The accuracy of the rules reached 76.3%, with the highest accuracy rules yielding 49 true negatives, 15 false negatives, 28 true positives and nine false positives (odds ratio [OR] 8.94, p < 0.0001). Analysis of the SNPs used most frequently in the overall ensemble of rules gave rise to a list of 'most important SNPs', which was not identical to the list of 'most differentiating SNPs' that one would calculate via studying each SNP independently. The top three genes containing the SNPs accounting for the highest accumulated importances were neuronal tryptophan hydroxylase (TPH2), catechol-O-methyltransferase (COMT) and nuclear receptor subfamily 3, group C, member 1 glucocorticoid receptor (NR3C1).

Conclusion: The fact that only 28 out of several million possible SNPs predict whether a person has CFS with 76% accuracy indicates that CFS has a genetic component that may help to explain some aspects of the illness.

Objectives: To further explore the relationship between chronic fatigue syndrome (CFS) and allostatic load (AL), we conducted a computational analysis involving 43 patients with CFS and 60 nonfatigued, healthy controls (NF) enrolled in a population-based case-control study in Wichita (KS, USA). We used traditional biostatistical methods to measure the association of high AL to standardized measures of physical and mental functioning, disability, fatigue and general symptom severity. We also used nonlinear regression technology embedded in machine learning algorithms to learn equations predicting various CFS symptoms based on the individual components of the allostatic load index (ALI).

Methods: An ALI was computed for all study participants using available laboratory and clinical data on metabolic, cardiovascular and hypothalamic-pituitary-adrenal (HPA) axis factors. Physical and mental functioning/impairment was measured using the Medical Outcomes Study 36-item Short Form Health Survey (SF-36); current fatigue was measured using the 20-item multidimensional fatigue inventory (MFI); frequency and intensity of symptoms was measured using the 19-item symptom inventory (SI). Genetic programming, a nonlinear regression technique, was used to learn an ensemble of different predictive equations rather just than a single one. Statistical analysis was based on the calculation of the percentage of equations in the ensemble that utilized each input variable, producing a measure of the 'utility' of the variable for the predictive problem at hand. Traditional biostatistics methods include the median and Wilcoxon tests for comparing the median levels of subscale scores obtained on the SF-36, the MFI and the SI summary score.

Results: Among CFS patients, but not controls, a high level of AL was significantly associated with lower median values (indicating worse health) of bodily pain, physical functioning and general symptom frequency/intensity. Using genetic programming, the ALI was determined to be a better predictor of these three health measures than any subcombination of ALI components among cases, but not controls.

Comprehensive, systematic and integrated data-centric statistical approaches to disease modeling can provide powerful frameworks for understanding disease etiology. Here, one such computational framework based on redescription mining in both its incarnations, static and dynamic, is discussed.

The static framework provides bioinformatic tools applicable to multifaceted datasets, containing genetic, transcriptomic, proteomic, and clinical data for diseased patients and normal subjects.

The dynamic redescription framework provides systems biology tools to model complex sets of regulatory, metabolic and signaling pathways in the initiation and progression of a disease.

As an example, the case of chronic fatigue syndrome (CFS) is considered, which has so far remained intractable and unpredictable in its etiology and nosology. The redescription mining approaches can be applied to the Centers for Disease Control and Prevention's Wichita (KS, USA) dataset, integrating transcriptomic, epidemiological and clinical data, and can also be used to study how pathways in the hypothalamic-pituitary-adrenal axis affect CFS patients.

Chronic fatigue syndrome (CFS) is a complex, multisymptom illness of unknown etiology. A variety of operational case definitions based on symptom report have been developed that share some common clinical features. Patients often come to clinical presentation after months or, more typically, years of symptomatic distress. Comorbid presentation with psychiatric illnesses has been noted.

Due to these fundamental issues, the impact of patient selection and the specification of the methods of outcome assessment loom large in therapeutic studies of CFS. While a substantial body of research has focused on increasing our understanding of the basic pathobiology of CFS, there have been comparatively fewer studies that have addressed the problems of patient characterization and outcome assessment.

The role of clinical methodology in the study of the therapeutics of CFS is not trivial, and may confound our understanding of pragmatic recommendations for treatment.