Machine learning systems show significant promise for forecasting patient adverse events via risk scores. However, these risk scores implicitly encode assumptions about future interventions that the patient is likely to receive, based on the intervention policy present in the training data. Without this important context, predictions from such systems are less interpretable for clinicians. We propose a joint model of intervention policy and adverse event risk as a means to explicitly communicate the model's assumptions about future interventions. We develop such an intervention policy model on MIMIC-III, a real world de-identified ICU dataset, and discuss some use cases that highlight the utility of this approach. We show how combining typical risk scores, such as the likelihood of mortality, with future intervention probability scores leads to more interpretable clinical predictions.

Early prediction of patient outcomes is important for targeting preventive care. This protocol describes a practical workflow for developing deep-learning risk models that can predict various clinical and operational outcomes from structured electronic health record (EHR) data. The protocol comprises five main stages: formal problem definition, data pre-processing, architecture selection, calibration and uncertainty, and generalizability evaluation. We have applied the workflow to four endpoints (acute kidney injury, mortality, length of stay and 30-day hospital readmission). The workflow can enable continuous (e.g., triggered every 6 h) and static (e.g., triggered at 24 h after admission) predictions. We also provide an open-source codebase that illustrates some key principles in EHR modeling. This protocol can be used by interdisciplinary teams with programming and clinical expertise to build deep-learning prediction models with alternate data sources and prediction tasks. We present a practical workflow describing how to use deep learning to develop continuous-risk models that can predict various adverse outcomes from structured electronic health records.

Objective: Multitask learning (MTL) using electronic health records allows concurrent prediction of multiple endpoints. MTL has shown promise in improving model performance and training efficiency; however, it often suffers from negative transfer - impaired learning if tasks are not appropriately selected. We introduce a sequential subnetwork routing (SeqSNR) architecture that uses soft parameter sharing to find related tasks and encourage cross-learning between them. Materials and Methods: Using the MIMIC-III (Medical Information Mart for Intensive Care-III) dataset, we train deep neural network models to predict the onset of 6 endpoints including specific organ dysfunctions and general clinical outcomes: acute kidney injury, continuous renal replacement therapy, mechanical ventilation, vasoactive medications, mortality, and length of stay. We compare single-task (ST) models with naive multitask and SeqSNR in terms of discriminative performance and label efficiency. Results: SeqSNR showed a modest yet statistically significant performance boost across 4 of 6 tasks compared with ST and naive multitasking. When the size of the training dataset was reduced for a given task (label efficiency), SeqSNR outperformed ST for all cases showing an average area under the precision-recall curve boost of 2.1%, 2.9%, and 2.1% for tasks using 1%, 5%, and 10% of labels, respectively. Conclusions: The SeqSNR architecture shows superior label efficiency compared with ST and naive multitasking, suggesting utility in scenarios in which endpoint labels are difficult to ascertain.

Recurrent Neural Networks (RNNs) are often used for sequential modeling of adverse outcomes in electronic health records (EHRs) due to their ability to encode past clinical states. These deep, recurrent architectures have displayed increased performance compared to other modeling approaches in a number of tasks, fueling the interest in deploying deep models in clinical settings. One of the key elements in ensuring safe model deployment and building user trust is model explainability. Testing with Concept Activation Vectors (TCAV) has recently been introduced as a way of providing human-understandable explanations by comparing high-level concepts to the network's gradients. While the technique has shown promising results in real-world imaging applications, it has not been applied to structured temporal inputs. To enable an application of TCAV to sequential predictions in the EHR, we propose an extension of the method to time series data. We evaluate the proposed approach on an open EHR benchmark from the intensive care unit, as well as synthetic data where we are able to better isolate individual effects.

Molecular networking has become a key method to visualize and annotate the chemical space in non-targeted mass spectrometry data. We present feature-based molecular networking (FBMN) as an analysis method in the Global Natural Products Social Molecular Networking (GNPS) infrastructure that builds on chromatographic feature detection and alignment tools. FBMN enables quantitative analysis and resolution of isomers, including from ion mobility spectrometry.

The early prediction of deterioration could have an important role in supporting healthcare professionals, as an estimated 11% of deaths in hospital follow a failure to promptly recognize and treat deteriorating patients1. To achieve this goal requires predictions of patient risk that are continuously updated and accurate, and delivered at an individual level with sufficient context and enough time to act. Here we develop a deep learning approach for the continuous risk prediction of future deterioration in patients, building on recent work that models adverse events from electronic health records2-17 and using acute kidney injury-a common and potentially life-threatening condition18-as an exemplar. Our model was developed on a large, longitudinal dataset of electronic health records that cover diverse clinical environments, comprising 703,782 adult patients across 172 inpatient and 1,062 outpatient sites. Our model predicts 55.8% of all inpatient episodes of acute kidney injury, and 90.2% of all acute kidney injuries that required subsequent administration of dialysis, with a lead time of up to 48 h and a ratio of 2 false alerts for every true alert. In addition to predicting future acute kidney injury, our model provides confidence assessments and a list of the clinical features that are most salient to each prediction, alongside predicted future trajectories for clinically relevant blood tests9. Although the recognition and prompt treatment of acute kidney injury is known to be challenging, our approach may offer opportunities for identifying patients at risk within a time window that enables early treatment.

To visualize the personalized distributions of pathogens and chemical environments, including microbial metabolites, pharmaceuticals, and their metabolic products, within and between human lungs afflicted with cystic fibrosis (CF), we generated three-dimensional (3D) microbiome and metabolome maps of six explanted lungs from three cystic fibrosis patients. These 3D spatial maps revealed that the chemical environments differ between patients and within the lungs of each patient. Although the microbial ecosystems of the patients were defined by the dominant pathogen, their chemical diversity was not. Additionally, the chemical diversity between locales in the lungs of the same individual sometimes exceeded interindividual variation. Thus, the chemistry and microbiome of the explanted lungs appear to be not only personalized but also regiospecific. Previously undescribed analogs of microbial quinolones and antibiotic metabolites were also detected. Furthermore, mapping the chemical and microbial distributions allowed visualization of microbial community interactions, such as increased production of quorum sensing quinolones in locations where Pseudomonas was in contact with Staphylococcus and Granulicatella, consistent with in vitro observations of bacteria isolated from these patients. Visualization of microbe-metabolite associations within a host organ in early-stage CF disease in animal models will help elucidate the complex interplay between the presence of a given microbial structure, antibiotics, metabolism of antibiotics, microbial virulence factors, and host responses.IMPORTANCE Microbial infections are now recognized to be polymicrobial and personalized in nature. Comprehensive analysis and understanding of the factors underlying the polymicrobial and personalized nature of infections remain limited, especially in the context of the host. By visualizing microbiomes and metabolomes of diseased human lungs, we reveal how different the chemical environments are between hosts that are dominated by the same pathogen and how community interactions shape the chemical environment or vice versa. We highlight that three-dimensional organ mapping methods represent hypothesis-building tools that allow us to design mechanistic studies aimed at addressing microbial responses to other microbes, the host, and pharmaceutical drugs.

To visualize the personalized distributions of pathogens, chemical environments including microbial metabolites, pharmaceuticals, and their metabolic products within and between human lungs afflicted with cystic fibrosis, we generated 3D microbiome and metabolome maps of six explanted lungs from three cystic fibrosis patients. These 3D spatial maps revealed that the chemical environments are variable between patients and within the lungs of each patient. Although the patients’ microbial ecosystems were defined by the dominant pathogen, their chemical diversity was not. Additionally, the chemical diversity between locales in lungs of the same individual sometimes exceeded inter-individual variation. Thus, the chemistry and microbiome of the explanted lungs appear to be not only personalized but also regiospecific. Previously undescribed analogs of microbial quinolones and antibiotic metabolites were also detected. Furthermore, mapping the chemical and microbial distributions allowed visualization of microbial community interactions, such as increased production of quorum sensing quinolones in locations where was in contact with and , consistent with observations of bacteria isolated from these patients. Visualization of microbe-metabolite associations within a host organ in early-stage CF disease in animal models will help elucidate a complex interplay between the presence of a given microbial structure, antibiotics, metabolism of antibiotics, microbial virulence factors, and host responses.

Our understanding of the spatial variation in the chemical and microbial makeup of an entire human organ remains limited, in part due to the size and heterogeneity of human organs and the complexity of the associated metabolome and microbiome. To address this challenge, we developed a workflow to enable the cartography of metabolomic and microbiome data onto a three-dimensional (3D) organ reconstruction built off radiological images. This enabled the direct visualization of the microbial and chemical makeup of a human lung from a cystic fibrosis patient. We detected host-derived molecules, microbial metabolites, medications, and region-specific metabolism of medications and placed it in the context of microbial distributions in the lung. Our tool further created browsable maps of a 3D microbiome/metabolome reconstruction map on a radiological image of a human lung and forms an interactive resource for the scientific community.

One of the goals of forensic science is to identify individuals and their lifestyle by analyzing the trace signatures left behind in built environments. Here, microbiome and metabolomic methods were used to see how its occupants used an office and to also gain insights into the lifestyle characteristics such as diet, medications, and personal care products of the occupants. 3D molecular cartography, a molecular visualization technology, was used in combination with mass spectrometry and microbial inventories to highlight human-environmental interactions. Molecular signatures were correlated with the individuals as well as their interactions with this indoor environment. There are person-specific chemical and microbial signatures associated with this environment that directly relate who had touched objects such as computers, computer mice, cell phones, desk phone, table or desks. By combining molecular and microbial investigation forensic strategies, this study offers novel insights to investigators who value the reconstructing of human lifestyle and characterization of human environmental interaction.

It is a common problem in natural product therapeutic lead discovery programs that despite good bioassay results in the initial extract, the active compound(s) may not be isolated during subsequent bioassay-guided purification. Herein, we present the concept of bioactive molecular networking to find candidate active molecules directly from fractionated bioactive extracts. By employing tandem mass spectrometry, it is possible to accelerate the dereplication of molecules using molecular networking prior to subsequent isolation of the compounds, and it is also possible to expose potentially bioactive molecules using bioactivity score prediction. Indeed, bioactivity score prediction can be calculated with the relative abundance of a molecule in fractions and the bioactivity level of each fraction. For that reason, we have developed a bioinformatic workflow able to map bioactivity score in molecular networks and applied it for discovery of antiviral compounds from a previously investigated extract of Euphorbia dendroides where the bioactive candidate molecules were not discovered following a classical bioassay-guided fractionation procedure. It can be expected that this approach will be implemented as a systematic strategy, not only in current and future bioactive lead discovery from natural extract collections but also for the reinvestigation of the untapped reservoir of bioactive analogues in previous bioassay-guided fractionation efforts.

Increasing appreciation of the gut microbiome's role in health motivates understanding the molecular composition of human feces. To analyze such complex samples, we developed a platform coupling targeted and untargeted metabolomics. The approach is facilitated through split flow from one UPLC, joint timing triggered by contact closure relays, and a script to retrieve the data. It is designed to detect specific metabolites of interest with high sensitivity, allows for correction of targeted information, enables better quantitation thus providing an advanced analytical tool for exploratory studies. Procrustes analysis revealed that untargeted approach provides a better correlation to microbiome data, associating specific metabolites with microbes that produce or process them. With the subset of over one hundred human fecal samples from the American Gut project, the implementation of the described coupled workflow revealed that targeted analysis using combination of single transition per compound with retention time misidentifies 30% of the targeted data and could lead to incorrect interpretations. At the same time, the targeted analysis extends detection limits and dynamic range, depending on the compounds, by orders of magnitude. A software application has been developed as a part of the workflow to allows for quantitative assessments based on calibration curves. Using this approach, we detect expected microbially modified molecules such as secondary bile acids and unexpected microbial molecules including Pseudomonas-associated quinolones and rhamnolipids in feces, setting the stage for metabolome-microbiome-wide association studies (MMWAS).

Our skin, our belongings, the world surrounding us, and the environment we live in are covered with molecular traces. Detecting and characterizing these molecular traces is necessary to understand the environmental impact on human health and disease, and to decipher complex molecular interactions between humans and other species, particularly microbiota. We recently introduced 3D molecular cartography for mapping small organic molecules (including metabolites, lipids, and environmental molecules) found on various surfaces, including the human body. Here, we provide a protocol and open-source software for 3D molecular cartography. The protocol includes step-by-step procedures for sample collection and processing, liquid chromatography-mass spectrometry (LC-MS)-based metabolomics, quality control (QC), molecular identification using MS/MS, data processing, and visualization with 3D models of the sampled environment. The LC-MS method was optimized for a broad range of small organic molecules. We enable scientists to reproduce our previously obtained results, and illustrate the broad utility of our approach with molecular maps of a rosemary plant and an ATM keypad after a PIN code was entered. To promote reproducibility, we introduce cartographical snapshots: files that describe a particular map and visualization settings, and that can be shared and loaded to reproduce the visualization. The protocol enables molecular cartography to be performed in any mass spectrometry laboratory and, in principle, for any spatially mapped data. We anticipate applications, in particular, in medicine, ecology, agriculture, biotechnology, and forensics. The protocol takes 78 h for a molecular map of 100 spots, excluding the reagent setup.

Untargeted metabolomics of environmental samples routinely detects thousands of small molecules, the vast majority of which cannot be identified. Meta-mass shift chemical (MeMSChem) profiling was developed to identify mass differences between related molecules using molecular networks. This approach illuminates metabolome-wide relationships between molecules and the putative chemical groups that differentiate them (e.g., H2, CH2, COCH2). MeMSChem profiling was used to analyze a publicly available metabolomic dataset of coral, algal, and fungal mat holobionts (i.e., the host and its associated microbes and viruses) sampled from some of Earth's most remote and pristine coral reefs. Each type of holobiont had distinct mass shift profiles, even when the analysis was restricted to molecules found in all samples. This result suggests that holobionts modify the same molecules in different ways and offers insights into the generation of molecular diversity. Three genera of stony corals had distinct patterns of molecular relatedness despite their high degree of taxonomic relatedness. MeMSChem profiles also partially differentiated between individuals, suggesting that every coral reef holobiont is a potential source of novel chemical diversity.

The cars we drive, the homes we live in, the restaurants we visit, and the laboratories and offices we work in are all a part of the modern human habitat. Remarkably, little is known about the diversity of chemicals present in these environments and to what degree molecules from our bodies influence the built environment that surrounds us and vice versa. We therefore set out to visualize the chemical diversity of five built human habitats together with their occupants, to provide a snapshot of the various molecules to which humans are exposed on a daily basis. The molecular inventory was obtained through untargeted liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis of samples from each human habitat and from the people that occupy those habitats. Mapping MS-derived data onto 3D models of the environments showed that frequently touched surfaces, such as handles (e.g., door, bicycle), resemble the molecular fingerprint of the human skin more closely than other surfaces that are less frequently in direct contact with humans (e.g., wall, bicycle frame). Approximately 50% of the MS/MS spectra detected were shared between people and the environment. Personal care products, plasticizers, cleaning supplies, food, food additives, and even medications that were found to be a part of the human habitat. The annotations indicate that significant transfer of chemicals takes place between us and our built environment. The workflows applied here will lay the foundation for future studies of molecular distributions in medical, forensic, architectural, space exploration, and environmental applications.

We describe development and application of the new SNP-MED modular software system, designed to examine the influence of single-nucleotide polymorphisms (SNPs) on the function of genes associated with the risk of socially significant diseases. The SNP-MED system includes Genomics, Proteomics, and Gene Networks’ software components, and the Information Resource Database.

Unipro UGENE is an open-source bioinformatics toolkit that integrates popular tools along with original instruments for molecular biologists within a unified user interface. Nowadays, most bioinformatics desktop applications, including UGENE, make use of a local data model while processing different types of data. Such an approach causes an inconvenience for scientists working cooperatively and relying on the same data. This refers to the need of making multiple copies of certain files for every workplace and maintaining synchronization between them in case of modifications. Therefore, we focused on delivering a collaborative work into the UGENE user experience. Currently, several UGENE installations can be connected to a designated shared database and users can interact with it simultaneously. Such databases can be created by UGENE users and be used at their discretion. Objects of each data type, supported by UGENE such as sequences, annotations, multiple alignments, etc., can now be easily imported from or exported to a remote storage. One of the main advantages of this system, compared to existing ones, is the almost simultaneous access of client applications to shared data regardless of their volume. Moreover, the system is capable of storing millions of objects. The storage itself is a regular database server so even an inexpert user is able to deploy it. Thus, UGENE may provide access to shared data for users located, for example, in the same laboratory or institution. UGENE is available at: http://ugene.net/download.html.