Detection of medications associated with Alzheimer's disease using ensemble methods and cooperative game theory.

OBJECTIVE: To study the feasibility of evaluating feature importance with Shapley Values and ensemble methods in the context of pharmacoepidemiology and medication safety. METHODS: We detected medications associated with Alzheimer's disease (AD) by examining the additive feature attribution with combined approach of Gradient Boosting and Shapley Values in the Medication use and Alzheimer's disease (MEDALZ) study, a nested case-control study of 70,719 verified AD cases in Finland. Our methodological approach is to do binary classification using Gradient boosting (an ensemble of weak classifiers) in a supervised learning manner. Then we apply Shapley Values (from cooperative game theory) to analyze how feature combinations affect the classification result. Medication use with a five to one year time-window before AD diagnosis was ascertained from Prescription register. RESULTS: Antipsychotics with low or medium dose, antidepressants with medium to high dose, and cardiovascular medications with medium to high dose were identified as the contributing features for separating cases with AD from controls. Medium to high amount of irregularity in the purchase pattern were an indicating feature for separating AD cases from controls. The similarity of medication purchases between AD cases and controls made the feature evaluation challenging. CONCLUSIONS: The combined approach of Gradient Boosting and feature evaluation with Shapley Values identified features that were consistent with findings from previous hypothesis-driven studies. Additionally, the results from the additive feature attribution identified new candidates for future studies on AD risk factors. Our approach also shows promise for studies based on observational studies, where feature identification and interactions in populations are of interest; and the applicability of using Shapley Values for evaluating feature relevance in pattern recognition tasks.

Genetic background of extreme violent behavior.

In developed countries, the majority of all violent crime is committed by a small group of antisocial recidivistic offenders, but no genes have been shown to contribute to recidivistic violent offending or severe violent behavior, such as homicide. Our results, from two independent cohorts of Finnish prisoners, revealed that a monoamine oxidase A (MAOA) low-activity genotype (contributing to low dopamine turnover rate) as well as the CDH13 gene (coding for neuronal membrane adhesion protein) are associated with extremely violent behavior (at least 10 committed homicides, attempted homicides or batteries). No substantial signal was observed for either MAOA or CDH13 among non-violent offenders, indicating that findings were specific for violent offending, and not largely attributable to substance abuse or antisocial personality disorder. These results indicate both low monoamine metabolism and neuronal membrane dysfunction as plausible factors in the etiology of extreme criminal violent behavior, and imply that at least about 5-10% of all severe violent crime in Finland is attributable to the aforementioned MAOA and CDH13 genotypes.

Effects of an isocaloric healthy Nordic diet on insulin sensitivity, lipid profile and inflammation markers in metabolic syndrome -- a randomized study (SYSDIET).

BACKGROUND: Different healthy food patterns may modify cardiometabolic risk. We investigated the effects of an isocaloric healthy Nordic diet on insulin sensitivity, lipid profile, blood pressure and inflammatory markers in people with metabolic syndrome. METHODS: We conducted a randomized dietary study lasting for 18-24 weeks in individuals with features of metabolic syndrome (mean age 55 years, BMI 31.6 kg m(-2) , 67% women). Altogether 309 individuals were screened, 200 started the intervention after 4-week run-in period, and 96 (proportion of dropouts 7.9%) and 70 individuals (dropouts 27%) completed the study, in the Healthy diet and Control diet groups, respectively. Healthy diet included whole-grain products, berries, fruits and vegetables, rapeseed oil, three fish meals per week and low-fat dairy products. An average Nordic diet served as a Control diet. Compliance was monitored by repeated 4-day food diaries and fatty acid composition of serum phospholipids. RESULTS: Body weight remained stable, and no significant changes were observed in insulin sensitivity or blood pressure. Significant changes between the groups were found in non-HDL cholesterol (-0.18, mmol L(-1) 95% CI -0.35; -0.01, P = 0.04), LDL to HDL cholesterol (-0.15, -0.28; -0.00, P = 0.046) and apolipoprotein B to apolipoprotein A1 ratios (-0.04, -0.07; -0.00, P = 0.025) favouring the Healthy diet. IL-1 Ra increased during the Control diet (difference -84, -133; -37 ng L(-1) , P = 0.00053). Intakes of saturated fats (E%, beta estimate 4.28, 0.02; 8.53, P = 0.049) and magnesium (mg, -0.23, -0.41; -0.05, P = 0.012) were associated with IL-1 Ra. CONCLUSIONS: Healthy Nordic diet improved lipid profile and had a beneficial effect on low-grade inflammation.

Lipoprotein subclass profiles in individuals with varying degrees of glucose tolerance: a population-based study of 9399 Finnish men.

OBJECTIVES: We investigated serum concentrations of lipoprotein subclass particles and their lipid components determined by proton nuclear magnetic resonance spectroscopy in a population-based study. DESIGN AND METHODS: A total of 9399 Finnish men were included in the study: 3034 men with normal fasting glucose and normal glucose tolerance; 4345 with isolated impaired fasting glucose (IFG); 312 with isolated impaired glucose tolerance (IGT); 1058 with both IFG and IGT; and 650 with newly diagnosed type 2 diabetes (New DM). Lipoprotein subclasses included chylomicrons (CM) and largest VLDL particles, other VLDL particles (five subclasses), intermediate-density lipoprotein (IDL), LDL (three subclasses) and HDL (four subclasses). The phospholipid, triglyceride (TG), cholesterol, free cholesterol and cholesterol ester levels of the lipoprotein particles were measured. RESULTS: Abnormal glucose tolerance (especially IGT and New DM) was significantly associated with increased concentrations of VLDL subclass particles and their components (with the exception of very small VLDL particles). After further adjustment for total TGs and HDL cholesterol, increased lipid concentrations in the CM/largest VLDL particles and in most of the other VLDL particles remained significant in individuals with isolated IGT, IFG+IGT and New DM. There was a consistent trend towards a decrease in large and an increase in small HDL particle concentrations in individuals with hyperglycaemia even after adjustment for serum total TGs and HDL cholesterol. CONCLUSIONS: Abnormal glucose tolerance modifies the concentrations of lipoprotein subclass particles and their lipid components in the circulation and is also related to compositional changes in these particles.

Association of indices of liver and adipocyte insulin resistance with 19 confirmed susceptibility loci for type 2 diabetes in 6,733 non-diabetic Finnish men.

AIMS/HYPOTHESIS: Of the confirmed type 2 diabetes susceptibility loci only a few are known to affect insulin sensitivity. We examined the association of indices of hepatic and adipocyte insulin resistance (IR) with 19 confirmed type 2 diabetes risk loci in a large population-based study. METHODS: Non-diabetic participants (n = 8,460, age 57.3 ± 7.0 years, BMI 26.8 ± 3.8 kg/m(2); mean ± SD) from a population-based cohort underwent an OGTT. Of them, 6,733 non-diabetic men were genotyped for single nucleotide polymorphisms (SNPs) in or near PPARG2 (also known as PPARG), KCNJ11, TCF7L2, SLC30A8, HHEX, CDKN2B, IGF2BP2, CDKAL1, HNF1B, WFS1, JAZF1, CDC123, TSPAN8, THADA, ADAMTS9, NOTCH2, KCNQ1, MTNR1B and SNP rs7480010. We investigated hepatic IR with a new index of liver IR. The adipocyte IR index was defined as a product of fasting NEFA and plasma insulin levels. RESULTS: Type 2 diabetes risk SNPs in or near KCNJ11 and HHEX were significantly (p < 0.0013), and those in or near CDKN2B, NOTCH2 and MTNR1B were nominally (p < 0.05), associated with decreased liver IR index. The Pro12 allele of PPARG2 was significantly associated with a high adipocyte IR index and nominally associated with high liver IR. CONCLUSIONS/INTERPRETATION: The Pro12 allele of PPARG2 seems to impair insulin's antilipolytic effect, leading to high NEFA release in the fasting state and IR. In addition, the type 2 diabetes risk alleles of KCNJ11 and HHEX, which are known to impair insulin secretion, were associated with increased hepatic insulin sensitivity.

A novel surrogate index for hepatic insulin resistance.

AIMS/HYPOTHESIS: In epidemiological and genetic studies surrogate indices are needed to investigate insulin resistance in different insulin-sensitive tissues. Our objective was to develop a surrogate index for hepatic insulin resistance. METHODS: A sample of 368 non-diabetic participants (age 43.0 ± 8.2 years, BMI 26.0 ± 4.0 kg/m(2); mean ± SD) whose endogenous glucose production (EGP) was measured with [6-6(2)H(2)]glucose in the fasting state and during the euglycaemic-hyperinsulinaemic clamp were included in the study. EGP multiplied by fasting plasma insulin (FPI) concentration was the reference measurement for liver insulin resistance (liver IR). Liver IR index was calculated with linear regression analysis including age, obesity indices, lipids, lipoproteins and several variables regulating glucose metabolism. RESULTS: The following variables were significantly associated with liver IR in multiple forward stepwise regression analysis: insulin AUC in an OGTT, fat mass, HDL-cholesterol and BMI. Liver IR index correlated significantly with EGP×FPI (r = 0.65, p < 0.001). In participants with abnormal glucose tolerance, the correlation of liver IR with EGP×FPI was slightly stronger (r = 0.69, p < 0.001) than in those with normal glucose tolerance (r = 0.62, p < 0.001). CONCLUSIONS/INTERPRETATION: We generated a novel surrogate index for liver insulin resistance correlating strongly with EGP × FPI.

Integration of genomic data for pharmacology and toxicology using Internet resources.

Genome based technologies such as sequencing and gene expression profiling using microarrays are creating massive amounts of data. Results from these studies have provided unique insights into targets, biochemical pathways, and biological systems affected by drug or xenobiotic chemical treatments. Moreover, these genomic technologies offer the potential to identify biomarkers for pharmacological development or toxicological prediction. Nonetheless, microarray studies involving a single compound produce useful although limited data. To gain further power from these individual studies, the ability to combine datasets through integration schemes has become imperative. In the current study, we describe and analyze currently available Internet resources designed to address this problem. Many functionalities, such as ability to cross reference orthologous genes across species or to combine same technology platform data, are present in these resources. Nonetheless, these resources are limited in the number of technology platforms they can support. While the ability to integrate all currently existing gene expression datasets remains enigmatic, the current tools provide a partial solution that may still yield unique insights into the affects of exogenous molecules at the level of gene expression.