A cautionary tale about properly vetting datasets used in supervised learning predicting metabolic pathway involvement.

The mapping of metabolite-specific data to pathways within cellular metabolism is a major data analysis step needed for biochemical interpretation. A variety of machine learning approaches, particularly deep learning approaches, have been used to predict these metabolite-to-pathway mappings, utilizing a training dataset of known metabolite-to-pathway mappings. A few such training datasets have been derived from the Kyoto Encyclopedia of Genes and Genomes (KEGG). However, several prior published machine learning approaches utilized an erroneous KEGG-derived training dataset that used SMILES molecular representations strings (KEGG-SMILES dataset) and contained a sizable proportion (~26%) duplicate entries. The presence of so many duplicates taint the training and testing sets generated from k-fold cross-validation of the KEGG-SMILES dataset. Therefore, the k-fold cross-validation performance of the resulting machine learning models was grossly inflated by the erroneous presence of these duplicate entries. Here we describe and evaluate the KEGG-SMILES dataset so that others may avoid using it. We also identify the prior publications that utilized this erroneous KEGG-SMILES dataset so their machine learning results can be properly and critically evaluated. In addition, we demonstrate the reduction of model k-fold cross-validation (CV) performance after de-duplicating the KEGG-SMILES dataset. This is a cautionary tale about properly vetting prior published benchmark datasets before using them in machine learning approaches. We hope others will avoid similar mistakes.

InChI isotopologue and isotopomer specifications.

This work presents a proposed extension to the International Union of Pure and Applied Chemistry (IUPAC) International Chemical Identifier (InChI) standard that allows the representation of isotopically-resolved chemical entities at varying levels of ambiguity in isotope location. This extension includes an improved interpretation of the current isotopic layer within the InChI standard and a new isotopologue layer specification for representing chemical intensities with ambiguous isotope localization. Both improvements support the unique isotopically-resolved chemical identification of features detected and measured in analytical instrumentation, specifically nuclear magnetic resonance and mass spectrometry. SCIENTIFIC CONTRIBUTION: This new extension to the InChI standard would enable improved annotation of analytical datasets characterizing chemical entities, supporting the FAIR (Findable, Accessible, Interoperable, and Reusable) guiding principles of data stewardship for chemical datasets, ultimately promoting Open Science in chemistry.

Predicting the Pathway Involvement of Metabolites Based on Combined Metabolite and Pathway Features.

A major limitation of most metabolomics datasets is the sparsity of pathway annotations for detected metabolites. It is common for less than half of the identified metabolites in these datasets to have a known metabolic pathway involvement. Trying to address this limitation, machine learning models have been developed to predict the association of a metabolite with a "pathway category", as defined by a metabolic knowledge base like KEGG. Past models were implemented as a single binary classifier specific to a single pathway category, requiring a set of binary classifiers for generating the predictions for multiple pathway categories. This past approach multiplied the computational resources necessary for training while diluting the positive entries in the gold standard datasets needed for training. To address these limitations, we propose a generalization of the metabolic pathway prediction problem using a single binary classifier that accepts the features both representing a metabolite and representing a pathway category and then predicts whether the given metabolite is involved in the corresponding pathway category. We demonstrate that this metabolite-pathway features pair approach not only outperforms the combined performance of training separate binary classifiers but demonstrates an order of magnitude improvement in robustness: a Matthews correlation coefficient of 0.784 ± 0.013 versus 0.768 ± 0.154.

Predicting The Pathway Involvement Of Metabolites Based on Combined Metabolite and Pathway Features.

A major limitation of most metabolomics datasets is the sparsity of pathway annotations of detected metabolites. It is common for less than half of identified metabolites in these datasets to have known metabolic pathway involvement. Trying to address this limitation, machine learning models have been developed to predict the association of a metabolite with a "pathway category", as defined by one of the metabolic knowledgebases like the Kyoto Encyclopedia of Gene and Genomes. Most of these models are implemented as a single binary classifier specific to a single pathway category, requiring a set of binary classifiers for generating predictions for multiple pathway categories. This single binary classifier per pathway category approach both multiplies the computational resources necessary for training while diluting the positive entries in gold standard datasets needed for training. To address the limitations of training separate classifiers, we propose a generalization of the metabolic pathway prediction problem using a single binary classifier that accepts both features representing a metabolite and features representing a generic pathway category and then predicts whether the given metabolite is involved in the corresponding pathway category. We demonstrate that this metabolite-pathway features-pair approach is not only competitive with the combined performance of training separate binary classifiers, but it outperforms the previous benchmark models.

Neurotensin accelerates atherosclerosis and increases circulating levels of short-chain and saturated triglycerides.

BACKGROUND AND AIMS: Obesity and type 2 diabetes are significant risk factors for atherosclerotic cardiovascular disease (CVD) worldwide, but the underlying pathophysiological links are poorly understood. Neurotensin (NT), a 13-amino-acid hormone peptide, facilitates intestinal fat absorption and contributes to obesity in mice fed a high-fat diet. Elevated levels of pro-NT (a stable NT precursor produced in equimolar amounts relative to NT) are associated with obesity, type 2 diabetes, and CVD in humans. Whether NT is a causative factor in CVD is unknown. METHODS: Nt+/+ and Nt-/- mice were either injected with adeno-associated virus encoding PCSK9 mutants or crossed with Ldlr-/- mice and fed a Western diet. Atherosclerotic plaques were analyzed by en face analysis, Oil Red O and CD68 staining. In humans, we evaluated the association between baseline pro-NT and growth of carotid bulb thickness after 16.4 years. Lipidomic profiles were analyzed. RESULTS: Atherosclerotic plaque formation is attenuated in Nt-deficient mice through mechanisms that are independent of reductions in circulating cholesterol and triglycerides but associated with remodeling of the plasma triglyceride pool. An increasing plasma concentration of pro-NT predicts atherosclerotic events in coronary and cerebral arteries independent of all major traditional risk factors, indicating a strong link between NT and atherosclerosis. This plasma lipid profile analysis confirms the association of pro-NT with remodeling of the plasma triglyceride pool in atherosclerotic events. CONCLUSIONS: Our findings are the first to directly link NT to increased atherosclerosis and indicate the potential role for NT in preventive and therapeutic strategies for CVD.

md_harmonize: A Python Package for Atom-Level Harmonization of Public Metabolic Databases.

A major challenge to integrating public metabolic resources is the use of different nomenclatures by individual databases. This paper presents md_harmonize, an open-source Python package for harmonizing compounds and metabolic reactions across various metabolic databases. The md_harmonize package utilizes a neighborhood-specific graph coloring method for generating a unique identifier for each compound via atom identifiers based on a compound's chemical structure. The resulting harmonized compounds and reactions can be used for various downstream analyses, including the construction of atom-resolved metabolic networks and models for metabolic flux analysis. Parts of the md_harmonize package have been optimized using a variety of computational techniques to allow certain NP-complete problems handled by the software to be tractable for these specific use-cases. The software is available on GitHub and through the Python Package Index, with end-user documentation hosted on GitHub Pages.

Benchmark Dataset for Training Machine Learning Models to Predict the Pathway Involvement of Metabolites.

Metabolic pathways are a human-defined grouping of life sustaining biochemical reactions, metabolites being both the reactants and products of these reactions. But many public datasets include identified metabolites whose pathway involvement is unknown, hindering metabolic interpretation. To address these shortcomings, various machine learning models, including those trained on data from the Kyoto Encyclopedia of Genes and Genomes (KEGG), have been developed to predict the pathway involvement of metabolites based on their chemical descriptions; however, these prior models are based on old metabolite KEGG-based datasets, including one benchmark dataset that is invalid due to the presence of over 1500 duplicate entries. Therefore, we have developed a new benchmark dataset derived from the KEGG following optimal standards of scientific computational reproducibility and including all source code needed to update the benchmark dataset as KEGG changes. We have used this new benchmark dataset with our atom coloring methodology to develop and compare the performance of Random Forest, XGBoost, and multilayer perceptron with autoencoder models generated from our new benchmark dataset. Best overall weighted average performance across 1000 unique folds was an F1 score of 0.8180 and a Matthews correlation coefficient of 0.7933, which was provided by XGBoost binary classification models for 11 KEGG-defined pathway categories.

Benchmark dataset for training machine learning models to predict the pathway involvement of metabolites.

Metabolic pathways are a human-defined grouping of life sustaining biochemical reactions, metabolites being both the reactants and products of these reactions. But many public datasets include identified metabolites whose pathway involvement is unknown, hindering metabolic interpretation. To address these shortcomings, various machine learning models, including those trained on data from the Kyoto Encyclopedia of Genes and Genomes (KEGG), have been developed to predict the pathway involvement of metabolites based on their chemical descriptions; however, these prior models are based on old metabolite KEGG-based datasets, including one benchmark dataset that is invalid due to the presence of over 1500 duplicate entries. Therefore, we have developed a new benchmark dataset derived from the KEGG following optimal standards of scientific computational reproducibility and including all source code needed to update the benchmark dataset as KEGG changes. We have used this new benchmark dataset with our atom coloring methodology to develop and compare the performance of Random Forest, XGBoost, and multilayer perceptron with autoencoder models generated from our new benchmark dataset. Best overall weighted average performance across 1000 unique folds was an F1-score of 0.8180 and Matthews correlation coefficient of 0.7933, which was provided by XGBoost binary classification models for 11 KEGG-defined pathway categories.

MESSES: Software for Transforming Messy Research Datasets into Clean Submissions to Metabolomics Workbench for Public Sharing.

In recent years, the FAIR guiding principles and the broader concept of open science has grown in importance in academic research, especially as funding entities have aggressively promoted public sharing of research products. Key to public research sharing is deposition of datasets into online data repositories, but it can be a chore to transform messy unstructured data into the forms required by these repositories. To help generate Metabolomics Workbench depositions, we have developed the MESSES (Metadata from Experimental SpreadSheets Extraction System) software package, implemented in the Python 3 programming language and supported on Linux, Windows, and Mac operating systems. MESSES helps transform tabular data from multiple sources into a Metabolomics Workbench specific deposition format. The package provides three commands, extract, validate, and convert, that implement a natural data transformation workflow. Moreover, MESSES facilitates richer metadata capture than is typically attempted by manual efforts. The source code and extensive documentation is hosted on GitHub and is also available on the Python Package Index for easy installation.

The metabolomics workbench file status website: a metadata repository promoting FAIR principles of metabolomics data.

BACKGROUND: An updated version of the mwtab Python package for programmatic access to the Metabolomics Workbench (MetabolomicsWB) data repository was released at the beginning of 2021. Along with updating the package to match the changes to MetabolomicsWB's 'mwTab' file format specification and enhancing the package's functionality, the included validation facilities were used to detect and catalog file inconsistencies and errors across all publicly available datasets in MetabolomicsWB. RESULTS: The MetabolomicsWB File Status website was developed to provide continuous validation of MetabolomicsWB data files and a useful interface to all found inconsistencies and errors. This list of detectable issues/errors include format parsing errors, format compliance issues, access problems via MetabolomicsWB's REST interface, and other small inconsistencies that can hinder reusability. The website uses the mwtab Python package to pull down and validate each available analysis file and then generates an html report. The website is updated on a weekly basis. Moreover, the Python website design utilizes GitHub and GitHub.io, providing an easy to replicate template for implementing other metadata, virtual, and meta- repositories. CONCLUSIONS: The MetabolomicsWB File Status website provides a metadata repository of validation metadata to promote the FAIR use of existing metabolomics datasets from the MetabolomicsWB data repository.

Identifying and sharing per-and polyfluoroalkyl substances hot-spot areas and exposures in drinking water.

Exposure to per- and polyfluoroalkyl substances (PFAS) in drinking water is widely recognized as a public health concern. Decision-makers who are responsible for managing PFAS drinking water risks lack the tools to acquire the information they need. In response to this need, we provide a detailed description of a Kentucky dataset that allows decision-makers to visualize potential hot-spot areas and evaluate drinking water systems that may be susceptible to PFAS contamination. The dataset includes information extracted from publicly available sources to create five different maps in ArcGIS Online and highlights potential sources of PFAS contamination in the environment in relation to drinking water systems. As datasets of PFAS drinking water sampling continue to grow as part of evolving regulatory requirements, we used this Kentucky dataset as an example to promote the reuse of this dataset and others like it. We incorporated the FAIR (Findable, Accessible, Interoperable, and Reusable) principles by creating a Figshare item that includes all data and associated metadata with these five ArcGIS maps.

A proposed FAIR approach for disseminating geospatial information system maps.

We present a draft Minimum Information About Geospatial Information System (MIAGIS) standard for facilitating public deposition of geospatial information system (GIS) datasets that follows the FAIR (Findable, Accessible, Interoperable and Reusable) principles. The draft MIAGIS standard includes a deposition directory structure and a minimum javascript object notation (JSON) metadata formatted file that is designed to capture critical metadata describing GIS layers and maps as well as their sources of data and methods of generation. The associated miagis Python package facilitates the creation of this MIAGIS metadata file and directly supports metadata extraction from both Esri JSON and GEOJSON GIS data formats plus options for extraction from user-specified JSON formats. We also demonstrate their use in crafting two example depositions of ArcGIS generated maps. We hope this draft MIAGIS standard along with the supporting miagis Python package will assist in establishing a GIS standards group that will develop the draft into a full standard for the wider GIS community as well as a future public repository for GIS datasets.

A Case for Accelerating Standards to Achieve the FAIR Principles of Environmental Health Research Experimental Data.

BACKGROUND: Funding agencies, publishers, and other stakeholders are pushing environmental health science investigators to improve data sharing; to promote the findable, accessible, interoperable, and reusable (FAIR) principles; and to increase the rigor and reproducibility of the data collected. Accomplishing these goals will require significant cultural shifts surrounding data management and strategies to develop robust and reliable resources that bridge the technical challenges and gaps in expertise. OBJECTIVE: In this commentary, we examine the current state of managing data and metadata-referred to collectively as (meta)data-in the experimental environmental health sciences. We introduce new tools and resources based on in vivo experiments to serve as examples for the broader field. METHODS: We discuss previous and ongoing efforts to improve (meta)data collection and curation. These include global efforts by the Functional Genomics Data Society to develop metadata collection tools such as the Investigation, Study, Assay (ISA) framework, and the Center for Expanded Data Annotation and Retrieval. We also conduct a case study of in vivo data deposited in the Gene Expression Omnibus that demonstrates the current state of in vivo environmental health data and highlights the value of using the tools we propose to support data deposition. DISCUSSION: The environmental health science community has played a key role in efforts to achieve the goals of the FAIR guiding principles and is well positioned to advance them further. We present a proposed framework to further promote these objectives and minimize the obstacles between data producers and data scientists to maximize the return on research investments. https://doi.org/10.1289/EHP11484.

kegg_pull: a software package for the RESTful access and pulling from the Kyoto Encyclopedia of Gene and Genomes.

BACKGROUND: The Kyoto Encyclopedia of Genes and Genomes (KEGG) provides organized genomic, biomolecular, and metabolic information and knowledge that is reasonably current and highly useful for a wide range of analyses and modeling. KEGG follows the principles of data stewardship to be findable, accessible, interoperable, and reusable (FAIR) by providing RESTful access to their database entries via their web-accessible KEGG API. However, the overall FAIRness of KEGG is often limited by the library and software package support available in a given programming language. While R library support for KEGG is fairly strong, Python library support has been lacking. Moreover, there is no software that provides extensive command line level support for KEGG access and utilization. RESULTS: We present kegg_pull, a package implemented in the Python programming language that provides better KEGG access and utilization functionality than previous libraries and software packages. Not only does kegg_pull include an application programming interface (API) for Python programming, it also provides a command line interface (CLI) that enables utilization of KEGG for a wide range of shell scripting and data analysis pipeline use-cases. As kegg_pull's name implies, both the API and CLI provide versatile options for pulling (downloading and saving) an arbitrary (user defined) number of database entries from the KEGG API. Moreover, this functionality is implemented to efficiently utilize multiple central processing unit cores as demonstrated in several performance tests. Many options are provided to optimize fault-tolerant performance across a single or multiple processes, with recommendations provided based on extensive testing and practical network considerations. CONCLUSIONS: The new kegg_pull package enables new flexible KEGG retrieval use cases not available in previous software packages. The most notable new feature that kegg_pull provides is its ability to robustly pull an arbitrary number of KEGG entries with a single API method or CLI command, including pulling an entire KEGG database. We provide recommendations to users for the most effective use of kegg_pull according to their network and computational circumstances.

Bilirubin Nanoparticle Treatment in Obese Mice Inhibits Hepatic Ceramide Production and Remodels Liver Fat Content.

Studies have indicated that increasing plasma bilirubin levels might be useful for preventing and treating hepatic lipid accumulation that occurs with metabolic diseases such as obesity and diabetes. We have previously demonstrated that mice with hyperbilirubinemia had significantly less lipid accumulation in a diet-induced non-alcoholic fatty liver disease (NAFLD) model. However, bilirubin's effects on individual lipid species are currently unknown. Therefore, we used liquid chromatography-mass spectroscopy (LC-MS) to determine the hepatic lipid composition of obese mice with NAFLD treated with bilirubin nanoparticles or vehicle control. We placed the mice on a high-fat diet (HFD) for 24 weeks and then treated them with bilirubin nanoparticles or vehicle control for 4 weeks while maintaining the HFD. Bilirubin nanoparticles suppressed hepatic fat content overall. After analyzing the lipidomics data, we determined that bilirubin inhibited the accumulation of ceramides in the liver. The bilirubin nanoparticles significantly lowered the hepatic expression of two essential enzymes that regulate ceramide production, Sgpl1 and Degs1. Our results demonstrate that the bilirubin nanoparticles improve hepatic fat content by reducing ceramide production, remodeling the liver fat content, and improving overall metabolic health.

Polycomb deficiency drives a FOXP2-high aggressive state targetable by epigenetic inhibitors.

Inhibitors of the Polycomb Repressive Complex 2 (PRC2) histone methyltransferase EZH2 are approved for certain cancers, but realizing their wider utility relies upon understanding PRC2 biology in each cancer system. Using a genetic model to delete Ezh2 in KRAS-driven lung adenocarcinomas, we observed that Ezh2 haplo-insufficient tumors were less lethal and lower grade than Ezh2 fully-insufficient tumors, which were poorly differentiated and metastatic. Using three-dimensional cultures and in vivo experiments, we determined that EZH2-deficient tumors were vulnerable to H3K27 demethylase or BET inhibitors. PRC2 loss/inhibition led to de-repression of FOXP2, a transcription factor that promotes migration and stemness, and FOXP2 could be suppressed by BET inhibition. Poorly differentiated human lung cancers were enriched for an H3K27me3-low state, representing a subtype that may benefit from BET inhibition as a single therapy or combined with additional EZH2 inhibition. These data highlight diverse roles of PRC2 in KRAS-driven lung adenocarcinomas, and demonstrate the utility of three-dimensional cultures for exploring epigenetic drug sensitivities for cancer.

Academic Tracker: Software for tracking and reporting publications associated with authors and grants.

In recent years, United States federal funding agencies, including the National Institutes of Health (NIH) and the National Science Foundation (NSF), have implemented public access policies to make research supported by funding from these federal agencies freely available to the public. Enforcement is primarily through annual and final reports submitted to these funding agencies, where all peer-reviewed publications must be registered through the appropriate mechanism as required by the specific federal funding agency. Unreported and/or incorrectly reported papers can result in delayed acceptance of annual and final reports and even funding delays for current and new research grants. So, it's important to make sure every peer-reviewed publication is reported properly and in a timely manner. For large collaborative research efforts, the tracking and proper registration of peer-reviewed publications along with generation of accurate annual and final reports can create a large administrative burden. With large collaborative teams, it is easy for these administrative tasks to be overlooked, forgotten, or lost in the shuffle. In order to help with this reporting burden, we have developed the Academic Tracker software package, implemented in the Python 3 programming language and supporting Linux, Windows, and Mac operating systems. Academic Tracker helps with publication tracking and reporting by comprehensively searching major peer-reviewed publication tracking web portals, including PubMed, Crossref, ORCID, and Google Scholar, given a list of authors. Academic Tracker provides highly customizable reporting templates so information about the resulting publications is easily transformed into appropriate formats for tracking and reporting purposes. The source code and extensive documentation is hosted on GitHub (https://moseleybioinformaticslab.github.io/academic_tracker/) and is also available on the Python Package Index (https://pypi.org/project/academic_tracker) for easy installation.

Scan-Centric, Frequency-Based Method for Characterizing Peaks from Direct Injection Fourier Transform Mass Spectrometry Experiments.

We present a novel, scan-centric method for characterizing peaks from direct injection multi-scan Fourier transform mass spectra of complex samples that utilizes frequency values derived directly from the spacing of raw m/z points in spectral scans. Our peak characterization method utilizes intensity-independent noise removal and normalization of scan-level data to provide a much better fit of relative intensity to natural abundance probabilities for low abundance isotopologues that are not present in all of the acquired scans. Moreover, our method calculates both peak- and scan-specific statistics incorporated within a series of quality control steps that are designed to robustly derive peak centers, intensities, and intensity ratios with their scan-level variances. These cross-scan characterized peaks are suitable for use in our previously published peak assignment methodology, Small Molecule Isotope Resolved Formula Enumeration (SMIRFE).

Hepatic kinome atlas: An in-depth identification of kinase pathways in liver fibrosis of humans and rodents.

BACKGROUND AND AIMS: Resolution of pathways that converge to induce deleterious effects in hepatic diseases, such as in the later stages, have potential antifibrotic effects that may improve outcomes. We aimed to explore whether humans and rodents display similar fibrotic signaling networks. APPROACH AND RESULTS: We assiduously mapped kinase pathways using 340 substrate targets, upstream bioinformatic analysis of kinase pathways, and over 2000 random sampling iterations using the PamGene PamStation kinome microarray chip technology. Using this technology, we characterized a large number of kinases with altered activity in liver fibrosis of both species. Gene expression and immunostaining analyses validated many of these kinases as bona fide signaling events. Surprisingly, the insulin receptor emerged as a considerable protein tyrosine kinase that is hyperactive in fibrotic liver disease in humans and rodents. Discoidin domain receptor tyrosine kinase, activated by collagen that increases during fibrosis, was another hyperactive protein tyrosine kinase in humans and rodents with fibrosis. The serine/threonine kinases found to be the most active in fibrosis were dystrophy type 1 protein kinase and members of the protein kinase family of kinases. We compared the fibrotic events over four models: humans with cirrhosis and three murine models with differing levels of fibrosis, including two models of fatty liver disease with emerging fibrosis. The data demonstrate a high concordance between human and rodent hepatic kinome signaling that focalizes, as shown by our network analysis of detrimental pathways. CONCLUSIONS: Our findings establish a comprehensive kinase atlas for liver fibrosis, which identifies analogous signaling events conserved among humans and rodents.