Tommi Suvitaival

Multivariate multi-way analysis of multi-source data

Motivation: Analysis of variance (ANOVA)-type methods are the default tool for the analysis of data with multiple covariates. These tools have been generalized to the multivariate analysis of high-throughput biological datasets, where the main challenge is the problem of small sample size and high dimensionality. However, the existing multi-way analysis methods are not designed for the currently increasingly important experiments where data is obtained from multiple sources. Common examples of such settings include integrated analysis of metabolic and gene expression profiles, or metabolic profiles from several tissues in our case, in a controlled multi-way experimental setup where disease status, medical treatment, gender and time-series are usual covariates. Results: We extend the applicability area of multivariate, multi-way ANOVA-type methods to multi-source cases by introducing a novel Bayesian model. The method is capable of finding covariate-related dependencies between the sources. It assumes the measurements consist of groups of similarly behaving variables, and estimates the multivariate covariate effects and their interaction effects for the discovered groups of variables. In particular, the method partitions the effects to those shared between the sources and to source-specific ones. The method is specifically designed for datasets with small sample sizes and high dimensionality. We apply the method to a lipidomics dataset from a lung cancer study with two-way experimental setup, where measurements from several tissues with mostly distinct lipids have been taken. The method is also directly applicable to gene expression and proteomics. Availability: An R-implementation is available at http://www.cis.hut.fi/projects/mi/software/multiWayCCA/ Contact: ilkka.huopaniemi@tkk.fi; samuel.kaski@tkk.fi

Two-way analysis of high-dimensional collinear data

We present a Bayesian model for two-way ANOVA-type analysis of high-dimensional, small sample-size datasets with highly correlated groups of variables. Modern cellular measurement methods are a main application area; typically the task is differential analysis between diseased and healthy samples, complicated by additional covariates requiring a multi-way analysis. The main complication is the combination of high dimensionality and low sample size, which renders classical multivariate techniques useless. We introduce a hierarchical model which does dimensionality reduction by assuming that the input variables come in similarly-behaving groups, and performs an ANOVA-type decomposition for the set of reduced-dimensional latent variables. We apply the methods to study lipidomic profiles of a recent large-cohort human diabetes study.

Lipidome as a predictive tool in progression to type 2 diabetes in Finnish men

BACKGROUND There is a need for early markers to track and predict the development of type 2 diabetes mellitus (T2DM) from the state of normal glucose tolerance through prediabetes. In this study we tested whether the plasma molecular lipidome has biomarker potential to predicting the onset of T2DM. METHODS We applied global lipidomic profiling on plasma samples from well-phenotyped men (107 cases, 216 controls) participating in the longitudinal METSIM study at baseline and at five-year follow-up. To validate the lipid markers, an additional study with a representative sample of adult male population (n=631) was also conducted. A total of 277 plasma lipids were analyzed using the lipidomics platform based on ultra-performance liquid chromatography coupled to time-of-flight mass spectrometry. Lipids with the highest predictive power for the development of T2DM were computationally selected, validated and compared to standard risk models without lipids. RESULTS A persistent lipid signature with higher levels of triacylglycerols and diacyl-phospholipids as well as lower levels of alkylacyl phosphatidylcholines was observed in progressors to T2DM. Lysophosphatidylcholine acyl C18:2 (LysoPC(18:2)), phosphatidylcholines PC(32:1), PC(34:2e) and PC(36:1), and triacylglycerol TG(17:1/18:1/18:2) were selected to the full model that included metabolic risk factors and FINDRISC variables. When further adjusting for BMI and age, these lipids had respective odds ratios of 0.32, 2.4, 0.50, 2.2 and 0.31 (all p<0.05) for progression to T2DM. The independently-validated predictive power improved in all pairwise comparisons between the lipid model and the respective standard risk model without the lipids (integrated discrimination improvement IDI>0; p<0.05). Notably, the lipid models remained predictive of the development of T2DM in the fasting plasma glucose-matched subset of the validation study. CONCLUSION This study indicates that a lipid signature characteristic of T2DM is present years before the diagnosis and improves prediction of progression to T2DM. Molecular lipid biomarkers were shown to have predictive power also in a high-risk group, where standard risk factors are not helpful at distinguishing progressors from non-progressors.

Serum metabolite profile associates with the development of metabolic co-morbidities in first-episode psychosis

Psychotic patients are at high risk for developing obesity, metabolic syndrome and type 2 diabetes. These metabolic co-morbidities are hypothesized to be related to both treatment side effects as well as to metabolic changes occurring during the psychosis. Earlier metabolomics studies have shown that blood metabolite levels are predictive of insulin resistance and type 2 diabetes in the general population as well as sensitive to the effects of antipsychotics. In this study, we aimed to identify the metabolite profiles predicting future weight gain and other metabolic abnormalities in psychotic patients. We applied comprehensive metabolomics to investigate serum metabolite profiles in a prospective study setting in 36 first-episode psychosis patients during the first year of the antipsychotic treatment and 19 controls. While corroborating several earlier findings when comparing cases and controls and the effects of the antipsychotic medication, we also found that prospective weight gain in psychotic patients was associated with increased levels of triacylglycerols with low carbon number and double-bond count at baseline, that is, lipids known to be associated with increased liver fat. Our study suggests that metabolite profiles may be used to identify the psychotic patients most vulnerable to develop metabolic co-morbidities, and may point to a pharmacological approach to counteract the antipsychotic-induced weight gain.

Stronger findings from mass spectral data through multi-peak modeling

BackgroundMass spectrometry-based metabolomic analysis depends upon the identification of spectral peaks by their mass and retention time. Statistical analysis that follows the identification currently relies on one main peak of each compound. However, a compound present in the sample typically produces several spectral peaks due to its isotopic properties and the ionization process of the mass spectrometer device. In this work, we investigate the extent to which these additional peaks can be used to increase the statistical strength of differential analysis.ResultsWe present a Bayesian approach for integrating data of multiple detected peaks that come from one compound. We demonstrate the approach through a simulated experiment and validate it on ultra performance liquid chromatography-mass spectrometry (UPLC-MS) experiments for metabolomics and lipidomics. Peaks that are likely to be associated with one compound can be clustered by the similarity of their chromatographic shape. Changes of concentration between sample groups can be inferred more accurately when multiple peaks are available.ConclusionsWhen the sample-size is limited, the proposed multi-peak approach improves the accuracy at inferring covariate effects. An R implementation and data are available at http://research.ics.aalto.fi/mi/software/peakANOVA/.

Identification of a plasma signature of psychotic disorder in children and adolescents from the Avon Longitudinal Study of Parents and Children (ALSPAC) cohort

The identification of an early biomarker of psychotic disorder is important as early treatment is associated with improved patient outcome. Metabolomic and lipidomic approaches in combination with multivariate statistical analysis were applied to identify plasma alterations in children (age 11) (38 cases vs 67 controls) and adolescents (age 18) (36 cases vs 117 controls) preceeding or coincident with the development of psychotic disorder (PD) at age 18 in the Avon Longitudinal Study of Parents and Children (ALSPAC). Overall, 179 lipids were identified at age 11, with 32 found to be significantly altered between the control and PD groups. Following correction for multiple comparisons, 8 of these lipids remained significant (lysophosphatidlycholines (LPCs) LPC(18:1), LPC(18:2), LPC(20:3); phosphatidlycholines (PCs) PC(32:2; PC(34:2), PC(36:4), PC(0-34-3) and sphingomyelin (SM) SM(d18:1/24:0)), all of which were elevated in the PD group. At age 18, 23 lipids were significantly different between the control and PD groups, although none remained significant following correction for multiple comparisons. In conclusion, the findings indicate that the lipidome is altered in the blood during childhood, long before the development of psychotic disorder. LPCs in particular are elevated in those who develop PD, indicating inflammatory abnormalities and altered phospholipid metabolism. These findings were not found at age 18, suggesting there may be ongoing alterations in the pathophysiological processes from prodrome to onset of PD.

Stronger findings for metabolomics through Bayesian modeling of multiple peaks and compound correlations

Motivation: Data analysis for metabolomics suffers from uncertainty because of the noisy measurement technology and the small sample size of experiments. Noise and the small sample size lead to a high probability of false findings. Further, individual compounds have natural variation between samples, which in many cases renders them unreliable as biomarkers. However, the levels of similar compounds are typically highly correlated, which is a phenomenon that we model in this work. Results: We propose a hierarchical Bayesian model for inferring differences between groups of samples more accurately in metabolomic studies, where the observed compounds are collinear. We discover that the method decreases the error of weak and non-existent covariate effects, and thereby reduces false-positive findings. To achieve this, the method makes use of the mass spectral peak data by clustering similar peaks into latent compounds, and by further clustering latent compounds into groups that respond in a coherent way to the experimental covariates. We demonstrate the method with three simulated studies and validate it with a metabolomic benchmark dataset. Availability and implementation: An implementation in R is available at http://research.ics.aalto.fi/mi/software/peakANOVA/. Contact: samuel.kaski@aalto.fi.

Cross-organism toxicogenomics with group factor analysis

We investigate the problem of detecting toxicogenomic associations that generalize across organisms, that is, statistical dependencies between transcriptional responses of multiple organisms and toxicological outcomes. We apply an interpretable probabilistic model to detect cross-organism toxicogenomic associations and propose an approach for drug toxicity analysis based on the interactive retrieval of drugs with similar toxicogenomic properties. We show that our approach can give relevant information about the properties of a drug even when direct prediction of toxicity is not feasible. Moreover, we show that a search from a cross-organism database can improve accuracy in the analysis.

Cross-species translation of multi-way biomarkers

We present a Bayesian translational model for matching patterns in data sets which have neither co-occurring samples nor variables, but only a similar experiment design dividing the samples into two or more categories. The model estimates covariate effects related to this design and separates the factors that are shared across the data sets from those specific to one data set. The model is designed to find similarities in medical studies, where there is great need for methods for linking laboratory experiments with model organisms to studies of human diseases and new treatments.

Platform for systems medicine research and diagnostic applications in psychotic disorders—The METSY project

Psychotic disorders are associated with metabolic abnormalities including alterations in glucose and lipid metabolism. A major challenge in the treatment of psychosis is to identify patients with vulnerable metabolic profiles who may be at risk of developing cardiometabolic co-morbidities. It is established that both central and peripheral metabolic organs use lipids to control energy balance and regulate peripheral insulin sensitivity. The endocannabinoid system, implicated in the regulation of glucose and lipid metabolism, has been shown to be dysregulated in psychosis. It is currently unclear how these endocannabinoid abnormalities relate to metabolic changes in psychosis. Here we review recent research in the field of metabolic co-morbidities in psychotic disorders as well as the methods to study them and potential links to the endocannabinoid system. We also describe the bioinformatics platforms developed in the EU project METSY for the investigations of the biological etiology in patients at risk of psychosis and in first episode psychosis patients. The METSY project was established with the aim to identify and evaluate multi-modal peripheral and neuroimaging markers that may be able to predict the onset and prognosis of psychiatric and metabolic symptoms in patients at risk of developing psychosis and first episode psychosis patients. Given the intrinsic complexity and widespread role of lipid metabolism, a systems biology approach which combines molecular, structural and functional neuroimaging methods with detailed metabolic characterisation and multi-variate network analysis is essential in order to identify how lipid dysregulation may contribute to psychotic disorders. A decision support system, integrating clinical, neuropsychological and neuroimaging data, was also developed in order to aid clinical decision making in psychosis. Knowledge of common and specific mechanisms may aid the etiopathogenic understanding of psychotic and metabolic disorders, facilitate early disease detection, aid treatment selection and elucidate new targets for pharmacological treatments.