ABSTRACT
BACKGROUND: Proteomic stable isotope probing (SIP) is used in microbial ecology to trace a non-radioactive isotope from a labeled substrate into de novo synthesized proteins in specific populations that are actively assimilating and metabolizing the substrate in a complex microbial community. The Sipros algorithm is used in proteomic SIP to identify variably labeled proteins and quantify their isotopic enrichment levels (atom%) by performing enrichment-resolved database searching. RESULTS: In this study, Sipros was upgraded to improve the labeled protein identification, isotopic enrichment quantification, and database searching speed. The new Sipros 4 was compared with the existing Sipros 3, Calisp, and MetaProSIP in terms of the number of identifications and the accuracy and precision of atom% quantification on both the peptide and protein levels using standard E. coli cultures with 1.07 atom%, 2 atom%, 5 atom%, 25 atom%, 50 atom%, and 99 atom% 13C enrichment. Sipros 4 outperformed Calisp and MetaProSIP across all samples, especially in samples with ≥ 5 atom% 13C labeling. The computational speed on Sipros 4 was > 20 times higher than Sipros 3 and was on par with the overall speed of Calisp- and MetaProSIP-based pipelines. Sipros 4 also demonstrated higher sensitivity for the detection of labeled proteins in two 13C-SIP experiments on a real-world soil community. The labeled proteins were used to trace 13C from 13C-methanol and 13C-labeled plant exudates to the consuming soil microorganisms and their newly synthesized proteins. CONCLUSION: Overall, Sipros 4 improved the quality of the proteomic SIP results and reduced the computational cost of SIP database searching, which will make proteomic SIP more useful and accessible to the border community. Video Abstract.
Subject(s)
Algorithms , Isotope Labeling , Proteomics , Proteomics/methods , Escherichia coli/metabolism , Carbon Isotopes/metabolism , Tandem Mass Spectrometry/methods , ProteomeABSTRACT
Percutaneous renal biopsy is commonly used for kidney cancer diagnosis. However, the biopsy procedure remains challenging in sampling accuracy. Here we introduce a forward-viewing optical coherence tomography probe for differentiating tumor and normal tissues, aiming at precise biopsy guidance. Totally, ten human kidney samples, nine of which had malignant renal carcinoma and one had benign oncocytoma, were used for system evaluation. Based on their distinct imaging features, carcinoma could be efficiently distinguished from normal renal tissues. Additionally, oncocytoma could be differentiated from carcinoma. We developed convolutional neural networks for tissue recognition. Compared to the conventional attenuation coefficient method, convolutional neural network models provided more accurate carcinoma predictions. These models reached a tissue recognition accuracy of 99.1% on a hold-out set of four kidney samples. Furthermore, they could efficiently distinguish oncocytoma from carcinoma. In conclusion, our convolutional neural network-aided endoscopic imaging platform could enhance carcinoma diagnosis during percutaneous renal biopsy procedures.
ABSTRACT
Climate changes significantly impact greenhouse gas emissions from wetland soil. Specifically, wetland soil may be exposed to oxygen (O2) during droughts, or to sulfate (SO42-) as a result of sea level rise. How these stressors - separately and together - impact microbial food webs driving carbon cycling in the wetlands is still not understood. To investigate this, we integrated geochemical analysis, proteogenomics, and stoichiometric modeling to characterize the impact of elevated SO42- and O2 levels on microbial methane (CH4) and carbon dioxide (CO2) emissions. The results uncovered the adaptive responses of this community to changes in SO42- and O2 availability and identified altered microbial guilds and metabolic processes driving CH4 and CO2 emissions. Elevated SO42- reduced CH4 emissions, with hydrogenotrophic methanogenesis more suppressed than acetoclastic. Elevated O2 shifted the greenhouse gas emissions from CH4 to CO2. The metabolic effects of combined SO42- and O2 exposures on CH4 and CO2 emissions were similar to those of O2 exposure alone. The reduction in CH4 emission by increased SO42- and O2 was much greater than the concomitant increase in CO2 emission. Thus, greater SO42- and O2 exposure in wetlands is expected to reduce the aggregate warming effect of CH4 and CO2. Metaproteomics and stoichiometric modeling revealed a unique subnetwork involving carbon metabolism that converts lactate and SO42- to produce acetate, H2S, and CO2 when SO42- is elevated under oxic conditions. This study provides greater quantitative resolution of key metabolic processes necessary for the prediction of CH4 and CO2 emissions from wetlands under future climate scenarios.
Subject(s)
Carbon Dioxide , Methane , Oxygen , Proteomics , Sulfates , Wetlands , Sulfates/metabolism , Oxygen/metabolism , Proteomics/methods , Methane/metabolism , Carbon Dioxide/metabolism , Soil Microbiology , Microbiota , Bacteria/metabolism , Bacteria/classification , Bacteria/genetics , Climate ChangeABSTRACT
Direct RNA nanopore sequencing reveals changes in gene expression, polyadenylation, splicing, m6A methylation, and pseudouridylation in response to influenza virus exposure in primary human bronchial epithelial cells. This study focuses on the epitranscriptomic profile of genes in the host immune response. In addition to polyadenylated noncoding RNA, we purified and sequenced nonpolyadenylated noncoding RNA and observed changes in expression, N6-methyl-adenosine (m6A), and pseudouridylation (Ψ) in these novel RNA. Two recently discovered lincRNA with roles in immune response, Chaserr and LEADR , became highly methylated in response to influenza exposure. Several H/ACA type snoRNAs that guide pseudouridylation are decreased in expression in response to influenza, and there is a corresponding decrease in the pseudouridylation of two novel lncRNA. Thus, novel epitranscriptomic changes revealed by direct RNA sequencing with nanopore technology provides unique insights into the host epitranscriptomic changes in epithelial gene networks that respond to influenza virus infection.
ABSTRACT
Creating intervention messages for smoking cessation is a labor-intensive process. Advances in Large Language Models (LLMs) offer a promising alternative for automated message generation. Two critical questions remain: 1) How to optimize LLMs to mimic human expert writing, and 2) Do LLM-generated messages meet clinical standards? We systematically examined the message generation and evaluation processes through three studies investigating prompt engineering (Study 1), decoding optimization (Study 2), and expert review (Study 3). We employed computational linguistic analysis in LLM assessment and established a comprehensive evaluation framework, incorporating automated metrics, linguistic attributes, and expert evaluations. Certified tobacco treatment specialists assessed the quality, accuracy, credibility, and persuasiveness of LLM-generated messages, using expert-written messages as the benchmark. Results indicate that larger LLMs, including ChatGPT, OPT-13B, and OPT-30B, can effectively emulate expert writing to generate well-written, accurate, and persuasive messages, thereby demonstrating the capability of LLMs in augmenting clinical practices of smoking cessation interventions.
ABSTRACT
Summary: Shotgun proteomics is widely used in many system biology studies to determine the global protein expression profiles of tissues, cultures, and microbiomes. Many non-distributed computer algorithms have been developed for users to process proteomics data on their local computers. However, the amount of data acquired in a typical proteomics study has grown rapidly in recent years, owing to the increasing throughput of mass spectrometry and the expanding scale of study designs. This presents a big data challenge for researchers to process proteomics data in a timely manner. To overcome this challenge, we developed a cloud-based parallel computing application to offer end-to-end proteomics data analysis software as a service (SaaS). A web interface was provided to users to upload mass spectrometry-based proteomics data, configure parameters, submit jobs, and monitor job status. The data processing was distributed across multiple nodes in a supercomputer to achieve scalability for large datasets. Our study demonstrated SaaS for proteomics as a viable solution for the community to scale up the data processing using cloud computing. Availability and implementation: This application is available online at https://sipros.oscer.ou.edu/ or https://sipros.unt.edu for free use. The source code is available at https://github.com/Biocomputing-Research-Group/CloudProteoAnalyzer under the GPL version 3.0 license.
ABSTRACT
Ammonia-oxidizing microorganisms (AOM) contribute to one of the largest nitrogen fluxes in the global nitrogen budget. Four distinct lineages of AOM: ammonia-oxidizing archaea (AOA), beta- and gamma-proteobacterial ammonia-oxidizing bacteria (ß-AOB and γ-AOB) and complete ammonia oxidizers (comammox), are thought to compete for ammonia as their primary nitrogen substrate. In addition, many AOM species can utilize urea as an alternative energy and nitrogen source through hydrolysis to ammonia. How the coordination of ammonia and urea metabolism in AOM influences their ecology remains poorly understood. Here we use stable isotope tracing, kinetics and transcriptomics experiments to show that representatives of the AOM lineages employ distinct regulatory strategies for ammonia or urea utilization, thereby minimizing direct substrate competition. The tested AOA and comammox species preferentially used ammonia over urea, while ß-AOB favoured urea utilization, repressed ammonia transport in the presence of urea and showed higher affinity for urea than for ammonia. Characterized γ-AOB co-utilized both substrates. These results reveal contrasting niche adaptation and coexistence patterns among the major AOM lineages.
Subject(s)
Archaea , Bacteria , Archaea/metabolism , Bacteria/metabolism , Ammonia/metabolism , Nitrogen/metabolism , Oxidation-Reduction , Nitrification , Phylogeny , Soil Microbiology , Urea/metabolismABSTRACT
Epidural anesthesia helps manage pain during different surgeries. Nonetheless, the precise placement of the epidural needle remains a challenge. In this study, we developed a probe based on polarization-sensitive optical coherence tomography (PS-OCT) to enhance the epidural anesthesia needle placement. The probe was tested on six porcine spinal samples. The multimodal imaging guidance used the OCT intensity mode and three distinct PS-OCT modes: (1) phase retardation, (2) optic axis, and (3) degree of polarization uniformity (DOPU). Each mode enabled the classification of different epidural tissues through distinct imaging characteristics. To further streamline the tissue recognition procedure, convolutional neural network (CNN) were used to autonomously identify the tissue types within the probe's field of view. ResNet50 models were developed for all four imaging modes. DOPU imaging was found to provide the highest cross-testing accuracy of 91.53%. These results showed the improved precision by PS-OCT in guiding epidural anesthesia needle placement.
Subject(s)
Anesthesia, Epidural , Tomography, Optical Coherence , Animals , Swine , Tomography, Optical Coherence/methods , Multimodal Imaging , Neural Networks, ComputerABSTRACT
Percutaneous renal biopsy (PRB) is commonly used for kidney cancer diagnosis. However, current PRB remains challenging in sampling accuracy. This study introduces a forward-viewing optical coherence tomography (OCT) probe for differentiating tumor and normal tissues, aiming at precise PRB guidance. Five human kidneys and renal carcinoma samples were used to evaluate the performance of our probe. Based on their distinct OCT imaging features, tumor and normal renal tissues can be accurately distinguished. We examined the attenuation coefficient for tissue classification and achieved 98.19% tumor recognition accuracy, but underperformed for distinguishing normal tissues. We further developed convolutional neural networks (CNN) and evaluated two CNN architectures: ResNet50 and InceptionV3, yielding 99.51% and 99.48% accuracies for tumor recognition, and over 98.90% for normal tissues recognition. In conclusion, combining OCT and CNN significantly enhanced the PRB guidance, offering a promising guidance technology for improved kidney cancer diagnosis.
ABSTRACT
Chagas disease (CD), caused by Trypanosoma cruzi (T. cruzi) protozoa, is a complicated parasitic illness with inadequate medical measures for diagnosing infection and monitoring treatment success. To address this gap, we analyzed changes in the metabolome of T. cruzi-infected mice via liquid chromatography tandem mass spectrometry of clinically accessible biofluids: saliva, urine, and plasma. Urine was the most indicative of infection status across mouse and parasite genotypes. Metabolites perturbed by infection in urine include kynurenate, acylcarnitines, and threonylcarbamoyladenosine. Based on these results, we sought to implement urine as a tool for the assessment of CD treatment success. Strikingly, it was found that mice with parasite clearance following benznidazole antiparasitic treatment had an overall urine metabolome comparable to that of mice that failed to clear parasites. These results provide a complementary hypothesis to explain clinical trial data in which benznidazole treatment did not improve patient outcomes in late-stage disease, even in patients with successful parasite clearance. Overall, this study provides insights into new small-molecule-based CD diagnostic methods and a new approach to assess functional responses to treatment.
Subject(s)
Chagas Disease , Parasites , Trypanocidal Agents , Trypanosoma cruzi , Humans , Mice , Animals , Trypanocidal Agents/pharmacology , Trypanocidal Agents/therapeutic use , Chagas Disease/parasitologyABSTRACT
Polyethylene (PE) is the most widely produced synthetic polymer and the most abundant plastic waste worldwide due to its recalcitrance to biodegradation and low recycle rate. Microbial degradation of PE has been reported, but the underlying mechanisms are poorly understood. Here, we isolated a Rhodococcus strain A34 from 609 day enriched cultures derived from naturally weathered plastic waste and identified the potential key PE degradation enzymes. After 30 days incubation with A34, 1% weight loss was achieved. Decreased PE molecular weight, appearance of C-O and CâO on PE, palmitic acid in the culture supernatant, and pits on the PE surface were observed. Proteomics analysis identified multiple key PE oxidation and depolymerization enzymes including one multicopper oxidase, one lipase, six esterase, and a few lipid transporters. Network analysis of proteomics data demonstrated the close relationships between PE degradation and metabolisms of phenylacetate, amino acids, secondary metabolites, and tricarboxylic acid cycles. The metabolic roadmap generated here provides critical insights for optimization of plastic degradation condition and assembly of artificial microbial communities for efficient plastic degradation.
Subject(s)
Microbiota , Polyethylene , Biodegradation, Environmental , Membrane Transport Proteins , Molecular WeightABSTRACT
Chagas Disease (CD), caused by Trypanosoma cruzi (T. cruzi) protozoa, is a complicated parasitic illness with inadequate medical measures for diagnosing infection and monitoring treatment success. To address this gap, we analyzed changes in the metabolome of T. cruzi-infected mice via liquid chromatography tandem mass spectrometry analysis of clinically-accessible biofluids: saliva, urine, and plasma. Urine was the most indicative of infection status, across mouse and parasite genotypes. Metabolites perturbed by infection in the urine include kynurenate, acylcarnitines, and threonylcarbamoyladenosine. Based on these results, we sought to implement urine as a tool for assessment of CD treatment success. Strikingly, it was found that mice with parasite clearance following benznidazole antiparasitic treatment had comparable overall urine metabolome to mice that failed to clear parasites. These results match with clinical trial data in which benznidazole treatment did not improve patient outcomes in late-stage disease. Overall, this study provides insights into new small molecule-based CD diagnostic methods and a new approach to assess functional treatment response.
ABSTRACT
Many complex diseases share common genetic determinants and are comorbid in a population. We hypothesized that the co-occurrences of diseases and their overlapping genetic etiology can be exploited to simultaneously improve multiple diseases' polygenic risk scores (PRS). This hypothesis was tested using a multi-task learning (MTL) approach based on an explainable neural network architecture. We found that parallel estimations of the PRS for 17 prevalent cancers in a pan-cancer MTL model were generally more accurate than independent estimations for individual cancers in comparable single-task learning (STL) models. Such performance improvement conferred by positive transfer learning was also observed consistently for 60 prevalent non-cancer diseases in a pan-disease MTL model. Interpretation of the MTL models revealed significant genetic correlations between the important sets of single nucleotide polymorphisms used by the neural network for PRS estimation. This suggested a well-connected network of diseases with shared genetic basis.
Subject(s)
Learning , Neural Networks, Computer , Humans , Risk Factors , Multifactorial Inheritance/genetics , Polymorphism, Single Nucleotide/genetics , Genetic Predisposition to Disease/geneticsABSTRACT
Carbohydrate utilization is critical to microbial survival. The phosphotransferase system (PTS) is a well-documented microbial system with a prominent role in carbohydrate metabolism, which can transport carbohydrates through forming a phosphorylation cascade and regulate metabolism by protein phosphorylation or interactions in model strains. However, those PTS-mediated regulated mechanisms have been underexplored in non-model prokaryotes. Here, we performed massive genome mining for PTS components in nearly 15,000 prokaryotic genomes from 4,293 species and revealed a high prevalence of incomplete PTSs in prokaryotes with no association to microbial phylogeny. Among these incomplete PTS carriers, a group of lignocellulose degrading clostridia was identified to have lost PTS sugar transporters and carry a substitution of the conserved histidine residue in the core PTS component, HPr (histidine-phosphorylatable phosphocarrier). Ruminiclostridium cellulolyticum was then selected as a representative to interrogate the function of incomplete PTS components in carbohydrate metabolism. Inactivation of the HPr homolog reduced rather than increased carbohydrate utilization as previously indicated. In addition to regulating distinct transcriptional profiles, PTS associated CcpA (Catabolite Control Protein A) homologs diverged from previously described CcpA with varied metabolic relevance and distinct DNA binding motifs. Furthermore, the DNA binding of CcpA homologs is independent of HPr homolog, which is determined by structural changes at the interface of CcpA homologs, rather than in HPr homolog. These data concordantly support functional and structural diversification of PTS components in metabolic regulation and bring novel understanding of regulatory mechanisms of incomplete PTSs in cellulose-degrading clostridia.
Subject(s)
Bacterial Proteins , Phosphoenolpyruvate Sugar Phosphotransferase System , Bacterial Proteins/genetics , Bacterial Proteins/metabolism , Cellulose , Histidine , Phosphoenolpyruvate Sugar Phosphotransferase System/genetics , Phosphoenolpyruvate Sugar Phosphotransferase System/chemistry , Phosphoenolpyruvate Sugar Phosphotransferase System/metabolism , Phosphotransferases/genetics , Carbohydrates , Firmicutes/genetics , DNAABSTRACT
BACKGROUND: Childhood Leukemia is the most common type of cancer among children. Nearly 39% of cancer-induced childhood deaths are attributable to Leukemia. Nevertheless, early intervention has long been underdeveloped. Moreover, there are still a group of children succumbing to their cancer due to the cancer care resource disparity. Therefore, it calls for an accurate predictive approach to improve childhood Leukemia survival and mitigate these disparities. Existing survival predictions rely on a single best model, which fails to consider model uncertainties in predictions. Prediction from a single model is brittle, with model uncertainty neglected, and inaccurate prediction could lead to serious ethical and economic consequences. METHODS: To address these challenges, we develop a Bayesian survival model to predict patient-specific survivals by taking model uncertainty into account. Specifically, we first develop a survival model predict time-varying survival probabilities. Second, we place different prior distributions over various model parameters and estimate their posterior distribution with full Bayesian inference. Third, we predict the patient-specific survival probabilities changing with respect to time by considering model uncertainty induced by posterior distribution. RESULTS: Concordance index of the proposed model is 0.93. Moreover, the standardized survival probability of the censored group is higher than that of the deceased group. CONCLUSIONS: Experimental results indicate that the proposed model is robust and accurate in predicting patient-specific survivals. It can also help clinicians track the contribution of multiple clinical attributes, thereby enabling well-informed intervention and timely medical care for childhood Leukemia.
Subject(s)
Leukemia , Child , Humans , Bayes Theorem , Probability , UncertaintyABSTRACT
Complex interactions exist among microorganisms in a community to carry out ecological processes and adapt to changing environments. Here, we constructed a quad-culture consisting of a cellulolytic bacterium (Ruminiclostridium cellulolyticum), a hydrogenotrophic methanogen (Methanospirillum hungatei), an acetoclastic methanogen (Methanosaeta concilii), and a sulfate-reducing bacterium (Desulfovibrio vulgaris). The four microorganisms in the quad-culture cooperated via cross-feeding to produce methane using cellulose as the only carbon source and electron donor. The community metabolism of the quad-culture was compared with those of the R. cellulolyticum-containing tri-cultures, bi-cultures, and mono-culture. Methane production was higher in the quad-culture than the sum of the increases in the tri-cultures, which was attributed to a positive synergy of four species. In contrast, cellulose degradation by the quad-culture was lower than the additive effects of the tri-cultures which represented a negative synergy. The community metabolism of the quad-culture was compared between a control condition and a treatment condition with sulfate addition using metaproteomics and metabolic profiling. Sulfate addition enhanced sulfate reduction and decreased methane and CO2 productions. The cross-feeding fluxes in the quad-culture in the two conditions were modeled using a community stoichiometric model. Sulfate addition strengthened metabolic handoffs from R. cellulolyticum to M. concilii and D. vulgaris and intensified substrate competition between M. hungatei and D. vulgaris. Overall, this study uncovered emergent properties of higher-order microbial interactions using a four-species synthetic community. IMPORTANCE A synthetic community was designed using four microbial species that together performed distinct key metabolic processes in the anaerobic degradation of cellulose to methane and CO2. The microorganisms exhibited expected interactions, such as cross-feeding of acetate from a cellulolytic bacterium to an acetoclastic methanogen and competition of H2 between a sulfate reducing bacterium and a hydrogenotrophic methanogen. This validated our rational design of the interactions between microorganisms based on their metabolic roles. More interestingly, we also found positive and negative synergies as emergent properties of high-order microbial interactions among three or more microorganisms in cocultures. These microbial interactions can be quantitatively measured by adding and removing specific members. A community stoichiometric model was constructed to represent the fluxes in the community metabolic network. This study paved the way toward a more predictive understanding of the impact of environmental perturbations on microbial interactions sustaining geochemically significant processes in natural systems.
Subject(s)
Euryarchaeota , Methane , Methane/metabolism , Cellulose/metabolism , Anaerobiosis , Carbon Dioxide/metabolism , Bacteria/metabolism , Euryarchaeota/metabolism , Sulfates/metabolismABSTRACT
Metaproteomics based on high-throughput tandem mass spectrometry (MS/MS) plays a crucial role in characterizing microbiome functions. The acquired MS/MS data is searched against a protein sequence database to identify peptides, which are then used to infer a list of proteins present in a metaproteome sample. While the problem of protein inference has been well-studied for proteomics of single organisms, it remains a major challenge for metaproteomics of complex microbial communities because of the large number of degenerate peptides shared among homologous proteins in different organisms. This challenge calls for improved discrimination of true protein identifications from false protein identifications given a set of unique and degenerate peptides identified in metaproteomics. MetaLP was developed here for protein inference in metaproteomics using an integrative linear programming method. Taxonomic abundance information extracted from metagenomics shotgun sequencing or 16s rRNA gene amplicon sequencing, was incorporated as prior information in MetaLP. Benchmarking with mock, human gut, soil, and marine microbial communities demonstrated significantly higher numbers of protein identifications by MetaLP than ProteinLP, PeptideProphet, DeepPep, PIPQ, and Sipros Ensemble. In conclusion, MetaLP could substantially improve protein inference for complex metaproteomes by incorporating taxonomic abundance information in a linear programming model.
Subject(s)
Programming, Linear , Tandem Mass Spectrometry , Humans , RNA, Ribosomal, 16S/genetics , Proteins/chemistry , Peptides/chemistryABSTRACT
The immune system of some genetically susceptible children can be triggered by certain environmental factors to produce islet autoantibodies (IA) against pancreatic ß cells, which greatly increases their risk for Type-1 diabetes. An environmental factor under active investigation is the gut microbiome due to its important role in immune system education. Here, we study gut metagenomes that are de-novo-assembled in 887 at-risk children in the Environmental Determinants of Diabetes in the Young (TEDDY) project. Our results reveal a small set of core protein families, present in >50% of the subjects, which account for 64% of the sequencing reads. Time-series binning generates 21,536 high-quality metagenome-assembled genomes (MAGs) from 883 species, including 176 species that hitherto have no MAG representation in previous comprehensive human microbiome surveys. IA seroconversion is positively associated with 2373 MAGs and negatively with 1549 MAGs. Comparative genomics analysis identifies lipopolysaccharides biosynthesis in Bacteroides MAGs and sulfate reduction in Anaerostipes MAGs as functional signatures of MAGs with positive IA-association. The functional signatures in the MAGs with negative IA-association include carbohydrate degradation in lactic acid bacteria MAGs and nitrate reduction in Escherichia MAGs. Overall, our results show a distinct set of gut microorganisms associated with IA seroconversion and uncovered the functional genomics signatures of these IA-associated microorganisms.
Subject(s)
Diabetes Mellitus, Type 1 , Gastrointestinal Microbiome , Microbiota , Autoantibodies , Child , Diabetes Mellitus, Type 1/genetics , Gastrointestinal Microbiome/genetics , Humans , Infant , Metagenome/genetics , Metagenomics/methods , SeroconversionABSTRACT
Epidural anesthesia requires injection of anesthetic into the epidural space in the spine. Accurate placement of the epidural needle is a major challenge. To address this, we developed a forward-view endoscopic optical coherence tomography (OCT) system for real-time imaging of the tissue in front of the needle tip during the puncture. We tested this OCT system in porcine backbones and developed a set of deep learning models to automatically process the imaging data for needle localization. A series of binary classification models were developed to recognize the five layers of the backbone, including fat, interspinous ligament, ligamentum flavum, epidural space, and spinal cord. The classification models provided an average classification accuracy of 96.65%. During puncture, it is important to maintain a safe distance between the needle tip and the dura mater. Regression models were developed to estimate that distance based on the OCT imaging data. Based on the Inception architecture, our models achieved a mean absolute percentage error of 3.05% ± 0.55%. Overall, our results validated the technical feasibility of using this novel imaging strategy to automatically recognize different tissue structures and measure the distances ahead of the needle tip during the epidural needle placement.
Subject(s)
Anesthesia, Epidural , Deep Learning , Anesthesia, Epidural/methods , Animals , Epidural Space/diagnostic imaging , Needles , Swine , Tomography, Optical Coherence/methodsABSTRACT
While neural networks can provide high predictive performance, it was a challenge to identify the salient features and important feature interactions used for their predictions. This represented a key hurdle for deploying neural networks in many biomedical applications that require interpretability, including predictive genomics. In this paper, linearizing neural network architecture (LINA) was developed here to provide both the first-order and the second-order interpretations on both the instance-wise and the model-wise levels. LINA combines the representational capacity of a deep inner attention neural network with a linearized intermediate representation for model interpretation. In comparison with DeepLIFT, LIME, Grad*Input and L2X, the first-order interpretation of LINA had better Spearman correlation with the ground-truth importance rankings of features in synthetic datasets. In comparison with NID and GEH, the second-order interpretation results from LINA achieved better precision for identification of the ground-truth feature interactions in synthetic datasets. These algorithms were further benchmarked using predictive genomics as a real-world application. LINA identified larger numbers of important single nucleotide polymorphisms (SNPs) and salient SNP interactions than the other algorithms at given false discovery rates. The results showed accurate and versatile model interpretation using LINA.