Methods development for "Omics" data

National Institute Of Environmental Health Sciences

Investigators

Linked publications, trials & patents

Abstract

In continued work on dose response modeling, in collaboration with Dr. Mathew Wheeler we have developed a suite of tools for dose response modeling. The need to analyze the complex relationships observed in high-throughput toxicogenomic and other omic platforms has resulted in an explosion of methodological advances in computational toxicology. However, advancements in the literature often outpace the development of software researchers can implement in their pipelines, and existing software is frequently based on pre-specified workflows built from well-vetted assumptions that may not be optimal for novel research questions. Accordingly, there is a need for a stable platform and open-source codebase attached to a programming language that allows users to program new algorithms. To fill this gap, the Biostatistics and Computational Biology Branch of the National Institute of Environmental Health Sciences, in cooperation with the National Toxicology Program (NTP) and US Environmental Protection Agency (EPA), developed ToxicR, an open-source R programming package. The ToxicR platform implements many of the standard analyses used by the NTP and EPA, including doseresponse analyses for continuous and dichotomous data that employ Bayesian, maximum likelihood, and model averaging methods, as well as many standard tests the NTP uses in rodent toxicology and carcinogenicity studies, such as the poly-K and Jonckheere trend tests. ToxicR is built on the same codebase as current versions of the EPAs Benchmark Dose software and NTPs BMDExpress software but has increased flexibility because it directly accesses this software. To demonstrate ToxicR, we developed a custom workflow to illustrate its capabilities for analyzing toxicogenomic data. The unique features of ToxicR will allow researchers in other fields to add modules, increasing its functionality in the future. Additionally, with Dr. Nat McNell, we have evaluated approaches for epidemiological weighting schemes in machine learning methods. Despite the prominent use of complex survey data and the growing popularity of machine learning methods in epidemiologic research, few machine learning software implementations offer options for handling complex samples. A major challenge impeding the broader incorporation of machine learning into epidemiologic research is incomplete guidance for analyzing complex survey data, including the importance of sampling weights for valid prediction in target populations. Using data from 15, 820 participants in the 19881994 National Health and Nutrition Examination Survey cohort, we determined whether ignoring weights in gradient boosting models of all-cause mortality affected prediction, as measured by the F1 score and corresponding 95% confidence intervals. In simulations, we additionally assessed the impact of sample size, weight variability, predictor strength, and model dimensionality. In the National Health and Nutrition Examination Survey data, unweighted model performance was inflated compared to the weighted model (F1 score 81.9% 95% confidence interval: 81.2%, 82.7% vs 77.4% 95% confidence interval: 76.1%, 78.6%). However, the error was mitigated if the F1 score was subsequently recalculated with observed outcomes from the weighted dataset (F1: 77.0%; 95% confidence interval: 75.7%, 78.4%). In simulations, this finding held in the largest sample size (N = 10,000) under all analytic conditions assessed. For sample sizes <5,000, sampling weights had little impact in simulations that more closely resembled a simple random sample (low weight variability) or in models with strong predictors, but findings were inconsistent under other analytic scenarios. Failing to account for sampling weights in gradient boosting models may limit generalizability for data from complex surveys, dependent on sample size and other analytic properties. In the absence of software for configuring weighted algorithms, post-hoc re-calculations of unweighted model performance using weighted observed outcomes may more accurately reflect model prediction in target populations than ignoring weights entirely. With Dr. David Reif's group, we developed the ToxPi*GIS Toolkit. It is a collection of methods for creating interactive feature layers that contain ToxPi profiles. It currently includes an ArcGIS Toolbox (ToxPiToolbox.tbx) for drawing location-specific ToxPi profiles in a single feature layer, a collection of modular Python scripts that create predesigned layer files containing ToxPi feature layers from the command line, and a collection of Python routines for useful data manipulation and preprocessing. We present workflows documenting ToxPi feature layer creation, sharing, and embedding for both novice and advanced users looking for additional customizability. Map visualizations created with the ToxPi*GIS Toolkit can be made freely available on public URLs, allowing users without ArcGIS Pro access or expertise to view and interact with them. Novice users with ArcGIS Pro access can create de novo custom maps, and advanced users can exploit additional customization options. The ArcGIS Toolbox provides a simple means for generating ToxPi feature layers. Ongoing projects building onto methods for detecting gene-environment interactions are currently ongoing, using variance QTLs to prioritize single nucleotide polymorphisms for detecting gene-gene interactions. Additionally, Dr. Ziyue Wang is working on developing new normalization approaches for microbiome data.