Dataset Information

Correcting mistakes in predicting distributions.

ABSTRACT: Motivation:Many applications monitor predictions of a whole range of features for biological datasets, e.g. the fraction of secreted human proteins in the human proteome. Results and error estimates are typically derived from publications. Results:Here, we present a simple, alternative approximation that uses performance estimates of methods to error-correct the predicted distributions. This approximation uses the confusion matrix (TP true positives, TN true negatives, FP false positives and FN false negatives) describing the performance of the prediction tool for correction. As proof-of-principle, the correction was applied to a two-class (membrane/not) and to a seven-class (localization) prediction. Availability and implementation:Datasets and a simple JavaScript tool available freely for all users at http://www.rostlab.org/services/distributions. Supplementary information:Supplementary data are available at Bioinformatics online.

SUBMITTER: Marot-Lassauzaie V

PROVIDER: S-EPMC6157078 | biostudies-literature | 2018 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Correcting mistakes in predicting distributions.

Marot-Lassauzaie Valérie V Bernhofer Michael M Rost Burkhard B

Bioinformatics (Oxford, England) 20181001 19

<h4>Motivation</h4>Many applications monitor predictions of a whole range of features for biological datasets, e.g. the fraction of secreted human proteins in the human proteome. Results and error estimates are typically derived from publications.<h4>Results</h4>Here, we present a simple, alternative approximation that uses performance estimates of methods to error-correct the predicted distributions. This approximation uses the confusion matrix (TP true positives, TN true negatives, FP false po ...[more]

PMID: 29762646

Similar Datasets

Project description:A species sensitivity distribution (SSD) is a cumulative distribution function of toxicity endpoints for a receptor group. A key assumption when deriving an SSD is that the toxicity data points are independent and identically distributed (iid). This assumption is tenuous, however, because closely related species are more likely to have similar sensitivities than are distantly related species. When the response of 1 species can be partially predicted by the response of another species, there is a dependency or autocorrelation in the data set. To date, phylogenetic relationships and the resulting dependencies in input data sets have been ignored in deriving SSDs. In this paper, we explore the importance of the phylogenetic signal in deriving SSDs using a case studies approach. The case studies involved toxicity data sets for aquatic autotrophs exposed to atrazine and aquatic and avian species exposed to chlorpyrifos. Full and partial data sets were included to explore the influences of differing phylogenetic signal strength and sample size. The phylogenetic signal was significant for some toxicity data sets (i.e., most chlorpyrifos data sets) but not for others (i.e., the atrazine data sets, the chlorpyrifos data sets for all insects, crustaceans, and birds). When a significant phylogenetic signal did occur, effective sample size was reduced. The reduction was large when the signal was strong. In spite of the reduced effective sample sizes, significant phylogenetic signals had little impact on fitted SSDs, even in the tails (e.g., hazardous concentration for 5th percentile species [HC5]). The lack of a phylogenetic signal impact occurred even when we artificially reduced original sample size and increased strength of the phylogenetic signal. We conclude that it is good statistical practice to account for the phylogenetic signal when deriving SSDs because most toxicity data sets do not meet the independence assumption. That said, SSDs and HC5s are robust to deviations from the independence assumption. Integr Environ Assess Manag 2019;00:1-13. © 2019 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals, Inc. on behalf of Society of Environmental Toxicology & Chemistry (SETAC).

Project description:Global coupled chemistry-climate models underestimate carbon monoxide (CO) in the Northern Hemisphere, exhibiting a pervasive negative bias against measurements peaking in late winter and early spring. While this bias has been commonly attributed to underestimation of direct anthropogenic and biomass burning emissions, chemical production and loss via OH reaction from emissions of anthropogenic and biogenic volatile organic compounds (VOCs) play an important role. Here we investigate the reasons for this underestimation using aircraft measurements taken in May and June 2016 from the Korea-United States Air Quality (KORUS-AQ) experiment in South Korea and the Air Chemistry Research in Asia (ARIAs) in the North China Plain (NCP). For reference, multispectral CO retrievals (V8J) from the Measurements of Pollution in the Troposphere (MOPITT) are jointly assimilated with meteorological observations using an ensemble adjustment Kalman filter (EAKF) within the global Community Atmosphere Model with Chemistry (CAM-Chem) and the Data Assimilation Research Testbed (DART). With regard to KORUS-AQ data, CO is underestimated by 42% in the control run and by 12% with the MOPITT assimilation run. The inversion suggests an underestimation of anthropogenic CO sources in many regions, by up to 80% for northern China, with large increments over the Liaoning Province and the North China Plain (NCP). Yet, an often-overlooked aspect of these inversions is that correcting the underestimation in anthropogenic CO emissions also improves the comparison with observational O3 datasets and observationally constrained box model simulations of OH and HO2. Running a CAM-Chem simulation with the updated emissions of anthropogenic CO reduces the bias by 29% for CO, 18% for ozone, 11% for HO2, and 27% for OH. Longer-lived anthropogenic VOCs whose model errors are correlated with CO are also improved, while short-lived VOCs, including formaldehyde, are difficult to constrain solely by assimilating satellite retrievals of CO. During an anticyclonic episode, better simulation of O3, with an average underestimation of 5.5 ppbv, and a reduction in the bias of surface formaldehyde and oxygenated VOCs can be achieved by separately increasing by a factor of 2 the modeled biogenic emissions for the plant functional types found in Korea. Results also suggest that controlling VOC and CO emissions, in addition to widespread NO x controls, can improve ozone pollution over East Asia.

Dataset Information

Correcting mistakes in predicting distributions.

Publications

Correcting mistakes in predicting distributions.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure