Supervised Machine Learning Models Research Articles

Comparative Analysis of Machine Learning Algorithms for Used Car Price Prediction

After 2021, over 90 million passenger automobiles were produced, marking a significant increase in auto production. This growth has led to a flourishing used car market, which has become a highly lucrative sector. One of the most critical and fascinating areas of research within this market is automobile price prediction. Accurate price prediction models can greatly benefit buyers, sellers, and businesses in the used car industry. This paper presents a detailed comparative analysis of two supervised machine learning models: K-Nearest Neighbour and Support Vector Machine regression techniques, to predict used car prices. We utilized a comprehensive dataset of used cars sourced from the Kaggle website for training and testing our models. The K Nearest Neighbour algorithm is known for its simplicity and effectiveness in regression tasks. On the other hand, the Support Vector Machine regression technique uses a different approach, finding the optimal hyperplane that best fits the data. Both methods have their strengths and weaknesses, which we explored in this study. Our results indicated that both KNN and SVM models performed well in predicting used car prices, but with slight variations in accuracy. Consequently, the suggested models fit as the optimum models and have an accuracy of about 83 percent for KNN and 80 percent for SVM. The results indicate that the KNN model slightly outperforms the SVM model in predicting used car prices.

Multi-Objective Design of DNA-Stabilized Nanoclusters Using Variational Autoencoders With Automatic Feature Extraction.

DNA-stabilized silver nanoclusters (AgN-DNAs) have sequence-tuned compositions and fluorescence colors. High-throughput experiments together with supervised machine learning models have recently enabled design of DNA templates that select for AgN-DNA properties, including near-infrared (NIR) emission that holds promise for deep tissue bioimaging. However, these existing models do not enable simultaneous selection of multiple AgN-DNA properties, and require significant expert input for feature engineering and class definitions. This work presents a model for multiobjective, continuous-property design of AgN-DNAs with automatic feature extraction, based on variational autoencoders (VAEs). This model is generative, i.e., it learns both the forward mapping from DNA sequence to AgN-DNA properties and the inverse mapping from properties to sequence, and is trained on an experimental data set of DNA sequences paired with AgN-DNA fluorescence properties. Experimental testing shows that the model enables effective design of AgN-DNA emission, including bright NIR AgN-DNAs with 4-fold greater abundance compared to training data. In addition, Shapley analysis is employed to discern learned nucleobase patterns that correspond to fluorescence color and brightness. This generative model can be adapted for a range of biomolecular systems with sequence-dependent properties, enabling precise design of emerging biomolecular nanomaterials.

Better stopping through cross validation in an iterative ensemble smoother: A perspective from supervised machine learning

Iterative ensemble smoothers (IES) are among the popular reservoir data assimilation (RDA) algorithms for reservoir characterization. The actual deployment of an IES algorithm requires implementing certain stopping criteria, normally adopted for runtime control (e.g., by stopping the IES when it reaches the maximum number of iterations) and/or safeguarding the RDA performance (e.g., by preventing the simulated data from overfitting the actual observations). In practice, for various reasons, it is often challenging for existing stopping criteria to simultaneously achieve both purposes. One noticeable issue, as illustrated in this work, is that in many situations, the qualities of the estimated reservoir models may already start to deteriorate before a conventional stopping criterion activates to terminate the iteration process. Following this observation, one practically important question arises: Is it possible to further improve the efficacy of the IES algorithm by designing a different stopping criterion so that the IES can stop earlier, saving computational costs while achieving better RDA performance?As one of the rare attempts in the community, this work aims to investigate the use of a new IES stopping criterion that has the potential to provide an affirmative answer to the above question. In this regard, our main idea is based on the concept of cross validation (CV), routinely adopted in supervised machine learning (SML) problems for early stopping to prevent SML models from overfitting the training data. Despite the noticed similarities between RDA and SML problems, some fundamental differences exist, making it fail to work well if one directly extends a vanilla CV procedure from SML to RDA. To tackle this identified challenge, we design an efficient CV procedure tailored for RDA problems, and inspect the performance of an IES algorithm equipped with this CV procedure (IES-CV) in both synthetic and real field case studies. Our numerical investigation indicates that the IES-CV algorithm achieves promising RDA performance in all case studies, confirming the possibility that with the aid of a proper stopping criterion, an IES algorithm can terminate at an appropriate iteration step with near-optimal RDA performance. Beyond these numerical findings, it is also our hope that the current work may help improve the best practices of applying IES to RDA problems, taking advantage of the effective, CV-based stopping criterion.

Cheminformatic Identification of Tyrosyl-DNA Phosphodiesterase 1 (Tdp1) Inhibitors: A Comparative Study of SMILES-Based Supervised Machine Learning Models.

Tyrosyl-DNA phosphodiesterase 1 (Tdp1) repairs damages in DNA induced by abortive topoisomerase 1 activity; however, maintenance of genetic integrity may sustain cellular division of neoplastic cells. It follows that Tdp1-targeting chemical inhibitors could synergize well with existing chemotherapy drugs to deny cancer growth; therefore, identification of Tdp1 inhibitors may advance precision medicine in oncology. Current computational research efforts focus primarily on molecular docking simulations, though datasets involving three-dimensional molecular structures are often hard to curate and computationally expensive to store and process. We propose the use of simplified molecular input line entry system (SMILES) chemical representations to train supervised machine learning (ML) models, aiming to predict potential Tdp1 inhibitors. An open-sourced consensus dataset containing the inhibitory activity of numerous chemicals against Tdp1 was obtained from Kaggle. Various ML algorithms were trained, ranging from simple algorithms to ensemble methods and deep neural networks. For algorithms requiring numerical data, SMILES were converted to chemical descriptors using RDKit, an open-sourced Python cheminformatics library. Out of 13 optimized ML models with rigorously tuned hyperparameters, the random forest model gave the best results, yielding a receiver operating characteristics-area under curve of 0.7421, testing accuracy of 0.6815, sensitivity of 0.6444, specificity of 0.7156, precision of 0.6753, and F1 score of 0.6595. Ensemble methods, especially the bootstrap aggregation mechanism adopted by random forest, outperformed other ML algorithms in classifying Tdp1 inhibitors from non-inhibitors using SMILES. The discovery of Tdp1 inhibitors could unlock more treatment regimens for cancer patients, allowing for therapies tailored to the patient's condition.

A Supervised Machine Learning Model for Dental Crowns Material Design for Manufacturing

A Supervised Machine Learning Model for Dental Crowns Material Design for Manufacturing

Baseline gut microbiota diversity and composition and albendazole efficacy in hookworm-infected individuals

Soil-transmitted helminth (STH) infections account for a significant global health burden, necessitating mass drug administration with benzimidazole-class anthelmintics, such as albendazole (ALB), for morbidity control. However, ALB efficacy shows substantial variability, presenting challenges for achieving consistent treatment outcomes. We have explored the potential impact of the baseline gut microbiota on ALB efficacy in hookworm-infected individuals through microbiota profiling and machine learning (ML) techniques. Our investigation included 89 stool samples collected from hookworm-infected individuals that were analyzed by microscopy and quantitative PCR (qPCR). Of these, 44 were negative by microscopy for STH infection using the Kato-Katz method and qPCR 21 days after treatment, which entails a cure rate of 49.4%. Microbiota characterization was based on amplicon sequencing of the V3–V4 16S ribosomal RNA gene region. Alpha and beta diversity analyses revealed no significant differences between participants who were cured and those who were not cured, suggesting that baseline microbiota diversity does not influence ALB treatment outcomes. Furthermore, differential abundance analysis at the phylum, family and genus levels yielded no statistically significant associations between bacterial communities and ALB efficacy. Utilizing supervised ML models failed to predict treatment response accurately. Our investigation did not provide conclusive insights into the relationship between gut microbiota and ALB efficacy. However, the results highlight the need for future research to incorporate longitudinal studies that monitor changes in the gut microbiota related to the infection and the cure with ALB, as well as functional metagenomics to better understand the interaction of the microbiome with the drug, and its role, if there is any, in modulating anthelmintic treatment outcomes in STH infections. Interdisciplinary approaches integrating microbiology, pharmacology, genetics and data science will be pivotal in advancing our understanding of STH infections and optimizing treatment strategies globally.Graphical

Drone delivery problem with multi-flight level: Machine learning based solution approach

This study provides a new perspective on the drone delivery problems (DDP) by conceptualizing the vertical space in multiple flight levels. The main advantage of drone delivery is efficiency in utilizing free three-dimension aerial space, enabling numerous travels at multiple flight levels. However, the operational efficiency tradeoff exists according to the flight level, particularly in metropolitan cities with countless skyscrapers. Operation on the upper level requires less detour on horizontal movement, but it needs more time on the vertical movement of drones to reach the upper level. This study introduces a novel DDP by dividing the vertical airspace into multiple flight levels, thereby providing an opportunity to increase overall delivery efficiency based on realistic constraints faced by cities. We formulate this problem into a mathematical model and suggest a new supervised machine learning approach called SPML (Sequential Prediction Machine Learning). The SPML has three phases. In the first phase, customers are sequenced by priority. The second phase uses a supervised machine learning model trained by the data collected from solving the mixed-integer linear programming (MILP) model to assign customers to the depot. The third phase is distributing jobs to drones by using dynamic programming.

A machine learning model based on CHAT-23 for early screening of autism in Chinese children.

Autism spectrum disorder (ASD) is a neurodevelopmental condition that significantly impacts the mental, emotional, and social development of children. Early screening for ASD typically involves the use of a series of questionnaires. With answers to these questionnaires, healthcare professionals can identify whether a child is at risk for developing ASD and refer them for further evaluation and diagnosis. CHAT-23 is an effective and widely used screening test in China for the early screening of ASD, which contains 23 different kinds of questions. We have collected clinical data from Wuxi, China. All the questions of CHAT-23 are regarded as different kinds of features for building machine learning models. We introduce machine learning methods into ASD screening, using the Max-Relevance and Min-Redundancy (mRMR) feature selection method to analyze the most important questions among all 23 from the collected CHAT-23 questionnaires. Seven mainstream supervised machine learning models were built and experiments were conducted. Among the seven supervised machine learning models evaluated, the best-performing model achieved a sensitivity of 0.909 and a specificity of 0.922 when the number of features was reduced to 9. This demonstrates the model's ability to accurately identify children for ASD with high precision, even with a more concise set of features. Our study focuses on the health of Chinese children, introducing machine learning methods to provide more accurate and effective early screening tests for autism. This approach not only enhances the early detection of ASD but also helps in refining the CHAT-23 questionnaire by identifying the most relevant questions for the diagnosis process.

Supervised machine learning for understanding and predicting the status of bistable eukaryotic plankton community in urbanized rivers

Understanding and predicting the ecological status of urbanized rivers is crucial for their restoration and management. However, the complex and nonlinear nature of ecological responses poses a challenge to the development of predictive models. Here, the study investigated and predicted the status of eukaryotic plankton communities in urbanized rivers by coupling environmental DNA metabarcoding, the alternative stable states theory, and supervised machine learning (SML) models. The results revealed two distinct states of eukaryotic plankton communities under similar environmental conditions: one state was characterized by the enrichment of a diverse phytoplankton population and the high relative abundance of protozoa, whereas the alternative state was characterized by abundant phytoplankton and fungi with an associated risk of algal blooms. Turbidity was identified as a key driver based on the SML model and Mantel test. Potential analysis demonstrated that the response pattern of eukaryotic plankton communities to turbidity was thresholds with hysteresis (Threshold1 = 17 NTU, Threshold2 = 24 NTU). A reduction in turbidity induced a regime shift in the eukaryotic plankton community toward an alternative state associated with a risk of algal blooms. In the prediction of ecological status, both SML models showed excellent performance (R2 > 0.80, RMSE < 0.1, Kappa > 0.70). Additionally, SHapley Additive exPlanations analysis identified turbidity, chlorophyll-a, chemical oxygen demand (COD), ammonia nitrogen and green algae's amplicon sequence variants as crucial features for prediction, with turbidity and COD showing a synergistic effect on ecological status. A framework was further proposed to enhance the understanding and prediction of ecological status in urbanized rivers. The obtained results of this study demonstrated the feasibility of using SML models to predict and explain the ecological status of urbanized rivers with alternative stable states. This provides valuable insights for the application of SML models in the restoration and management of urbanized rivers.

Analysis of 27 supervised machine learning models for the co-gasification assessment of peanut shell and spent tea residue in an open-core downdraft gasifier

The producer gas (PG) obtained through thermochemical processing of the various renewable biomass types contributes to Sustainable Development Goal-7 (SDG-7). Hence, this study examined 27 supervised and multiple-input single-output ML models in terms of performance metrics and accuracy to predict CO, H2, CH4, and CO2 compositions, as well as PG's HHV in the co-gasification of peanut shell (PS) and spent tea residue (STR). Multiple trial experiments were tested with various equivalence ratios (ERs) and mixing ratios (MRs) on a 90 m3/h gas-yield open-core, down-draft gasifier with air as the gasification medium. About 18 models were chosen after evaluating their performance metrics in estimating relevant parameters. The accuracy of those models is assessed based on 12 sample runs with varying ER and MR. The findings revealed that 16 models from the linear, neural network, support vector machine, Gaussian process, tree, and ensemble categories were reliable ML models. The chosen 16 models' R2 values for CO, H2, CH4, CO2, and HHVPG are above 0.931, 0.937, 0.962, 0.806, and 0.971, respectively, with the exception of TRE. Additionally, the models demonstrated a prediction accuracy of >95 % for all of the investigated parameters in the co-gasification of PS and STR when compared to random experimental runs.

Long-read RNA sequencing reveals allele-specific N6-methyladenosine modifications.

Long-read sequencing technology enables highly accurate detection of allele-specific RNA expression, providing insights into the effects of genetic variation on splicing and RNA abundance. Furthermore, the ability to directly sequence RNA promises the detection of RNA modifications in tandem with ascertaining the allelic origin of each molecule. Here, we leverage these advantages to determine allele-biased patterns of N6-methyladenosine (m6A) modifications in native mRNA. We utilized human and mouse cells with known genetic variants to assign allelic origin of each mRNA molecule combined with a supervised machine learning model to detect read-level m6A modification ratios. Our analyses revealed the importance of sequences adjacent to the DRACH-motif in determining m6A deposition, in addition to allelic differences that directly alter the motif. Moreover, we discovered allele-specific m6A modification (ASM) events with no genetic variants in close proximity to the differentially modified nucleotide, demonstrating the unique advantage of using long reads and surpassing the capabilities of antibody-based short-read approaches. This technological advancement promises to advance our understanding of the role of genetics in determining mRNA modifications.

Prediction of medical admissions after psychiatric inpatient hospitalization in bipolar disorder: a retrospective cohort study.

Bipolar Disorder (BD) is a severe mental illness associated with high rates of general medical comorbidity, reduced life expectancy, and premature mortality. Although BD has been associated with high medical hospitalization, the factors that contribute to this risk remain largely unexplored. We used baseline medical and psychiatric records to develop a supervised machine learning model to predict general medical admissions after discharge from psychiatric hospitalization. In this retrospective three-year cohort study of 71 patients diagnosed with BD (mean age=52.19 years, females=56.33%), lasso regression models combining medical and psychiatric records, as well as those using them separately, were fitted and their predictive power was estimated using a leave-one-out cross-validation procedure. The proportion of medical admissions in patients with BD was higher compared with age- and sex-matched hospitalizations in the same region (25.4% vs. 8.48%). The lasso model fairly accurately predicted the outcome (area under the curve [AUC]=69.5%, 95%C.I.=55-84.1; sensitivity=61.1%, specificity=75.5%, balanced accuracy=68.3%). Notably, pre-existing cardiovascular, neurological, or osteomuscular diseases collectively accounted for more than 90% of the influence on the model. The accuracy of the model based on medical records was slightly inferior (AUC=68.7%, 95%C.I. = 54.6-82.9), while that of the model based on psychiatric records only was below chance (AUC=61.8%, 95%C.I.=46.2-77.4). Our findings support the need to monitor medical comorbidities during clinical decision-making to tailor and implement effective preventive measures in people with BD. Further research with larger sample sizes and prospective cohorts is warranted to replicate these findings and validate the predictive model.

Machine Learning–Based Prediction of Chronic Shunt-Dependent Hydrocephalus After Spontaneous Subarachnoid Hemorrhage

BackgroundChronic posthemorrhagic hydrocephalus often arises following spontaneous subarachnoid hemorrhage (SAH). Timely identification of patients predisposed to develop chronic shunt-dependent hydrocephalus may significantly enhance clinical outcomes. MethodsWe performed an analysis of 510 SAH-patients treated at our institution between 2013 and 2018. Clinical and radiological variables, including age, sex, Hunt & Hess grade, Fisher-Score, external ventricular drainage placement, central nervous system infection, aneurysm characteristics, and treatment modalities, were evaluated. Supervised machine learning models, trained and compared using Python and scikit-learn, were employed to predict chronic shunt-dependent hydrocephalus. Model performance was rigorously assessed through repeated cross-validation. To facilitate transparency and collaboration, we publicly released the dataset and code on GitHub (https://github.com/RISCSoftware/shuntclf) and developed an interactive web application (https://huggingface.co/spaces/risc42/shuntclf). ResultsAmong the evaluated machine learning models, logistic regression exhibited superior performance, with an AUC-ROC of 0.819 and an AUC-PR of 0.482, along with the highest F1 score of 0.473. Although the balanced accuracy scores of the models were generally proximate, ranging from 0.735 to 0.764, logistic regression consistently outperform others in key metrics such as AUC-ROC and AUC-PR. Conversely, female gender and absence of aneurysm within the anterior communicating artery were associated with reduced shunt requirement likelihood. ConclusionMachine learning models, including logistic regression, demonstrate strong predictive capability for early chronic shunt-dependent hydrocephalus following spontaneous SAH, which may potentially contribute to more timely shunt placement interventions. This predictive capability is supported by our web interface, which simplifies the application of these models, aiding clinicians in efficiently determining the need for shunt placement.

Optimizing Lung Cancer Diagnosis with Machine Learning and Feature Selection Methods

Lung cancer is a prevalent disease, with nearly 238,000 new cases diagnosed in 2023. This study utilizes clinical predictors from a Kaggle dataset containing 309 observations across 15 variables to aid in lung cancer diagnosis. The variables include swallowing difficulty, peer pressure, gender, allergy, yellow fingers, anxiety, wheezing, alcohol consumption, chronic disease, chest pain, coughing, fatigue, smoking, age, and shortness of breath. The research aims to develop and compare various supervised machine learning models for classifying and predicting lung cancer, while also identifying key clinical tests and parameters using unsupervised statistical models. The dataset was divided into training and test sets, balanced, and preprocessed for unbiased training. Feature selection and machine learning models were applied to identify crucial predictors. The study explored tree models, logistic regression, Naïve Bayes, support vector machine (SVM), ensemble, neural network, and kernel models. Among these, the linear SVM achieved the highest accuracy of 93.75% with 5-fold cross-validation. However, it showed overfitting, with a lower test accuracy of 82.55%. The Gaussian Naïve Bayes model emerged as the optimal choice, providing consistent performance between validation and test cases. It achieved the highest cross-validation classification accuracy of 82.81% using only 9 variables: swallowing difficulty, peer pressure, gender, allergy, yellow fingers, anxiety, wheezing, alcohol consumption, and chronic disease. This model allows for effective training with fewer predictors without compromising classification

Forecasting infectious diseases in Brazilian cities: Integrating socio-economic and geographic data from related cities through a machine learning approach

Supervised machine learning models and public surveillance data have been employed for infectious disease forecasting in many settings. These models leverage various data sources capturing drivers of disease spread, such as climate conditions or human behavior. However, few models have incorporated the organizational structure of different geographic locations for forecasting. Traveling waves of seasonal outbreaks have been reported for dengue, influenza, and other infectious diseases, and many of the drivers of infectious disease dynamics may be shared across different cities, either due to their geographic or socioeconomic proximity. In this study, we developed a machine learning model to predict case counts of four infectious diseases across Brazilian cities one week ahead by incorporating information from related cities. We compared selecting related cities using both geographic distance and GDP per capita. Incorporating information from geographically proximate cities improved predictive performance for two of the four diseases, specifically COVID-19 and Zika. We also discuss the impact on forecasts in the presence of anomalous contagion patterns and the limitations of the proposed methodology.

Machine learning approaches to detect hepatocyte chromatin alterations from iron oxide nanoparticle exposure

This study focuses on developing machine learning models to detect subtle alterations in hepatocyte chromatin organization due to Iron (II, III) oxide nanoparticle exposure, hypothesizing that exposure will significantly alter chromatin texture. A total of 2000 hepatocyte nuclear regions of interest (ROIs) from mouse liver tissue were analyzed, and for each ROI, 5 different parameters were calculated: Long Run Emphasis, Short Run Emphasis, Run Length Nonuniformity, and 2 wavelet coefficient energies obtained after the discrete wavelet transform. These parameters served as input for supervised machine learning models, specifically random forest and gradient boosting classifiers. The models demonstrated relatively robust performance in distinguishing hepatocyte chromatin structures belonging to the group exposed to IONPs from the controls. The study's findings suggest that iron oxide nanoparticles induce substantial changes in hepatocyte chromatin distribution and underscore the potential of AI techniques in advancing hepatocyte evaluation in physiological and pathological conditions.

A Machine Learning-Based Prediction of Malaria Occurrence in Kenya

For many years’ malaria has been a health public concern in Kenya as well as many parts of Africa and other parts of the world. The purpose of this study is to develop and evaluate a supervised machine learning model to predict malaria occurrence (final malaria test results) in Kenya. The study investigated twelve predictor variables on the outcome variable (malaria test results), where five machine learning models namely; k-nearest neighbors, support vector machines, random forest, tree bagging, and boosting, were estimated. During the model evaluation, random forest emerged as the best overall model in the classification and prediction of final malaria test results. The model attained a higher classification accuracy of 97.33%, sensitivity of 71.1%, specificity of 98.4%, balanced accuracy of 84.7% and an area under the curve of 98.3%. From the final model, the presence of plasmodium falciparum emerged most important feature, followed by region, endemic zone and anemic level. The feature with the least importance in predicting final malaria test results was having mosquito nets. In conclusion, employing Machine learning algorithms enhances early detection, optimizing resource allocation for interventions, and ultimately reducing the incidence and impact of malaria in the Kenya. The study recommends allocation of resources and funds to areas with the presence of plasmodium falciparum, region susceptible to malaria, endemic zones and anemic prone areas.

Middle ear-acquired cholesteatoma diagnosis based on CT scan image mining using supervised machine learning models

BackgroundDistinguishing between middle ear cholesteatoma and chronic suppurative otitis media (CSOM) is an ongoing challenge. While temporal bone computed tomography (CT) scan is highly accurate for diagnosing middle ear conditions, its specificity in discerning between cholesteatoma and CSOM is only moderate. To address this issue, we utilized trained machine learning models to enhance the specificity of temporal bone CT scan in diagnosing middle ear cholesteatoma. Our database consisted of temporal bone CT scan native images from 122 patients diagnosed with middle ear cholesteatoma and a control group of 115 patients diagnosed with CSOM, with both groups labeled based on surgical findings. We preprocessed the native images to isolate the region of interest and then utilized the Inception V3 convolutional neural network for image embedding into data vectors. Classification was performed using machine learning models including support vector machine (SVM), k-nearest neighbors (k-NN), random forest, and neural network. Statistical metrics employed to interpret the results included classification accuracy, precision, recall, F1 score, confusion matrix, area under the receiver operating characteristic curve (AUC), and FreeViz diagram.ResultsOur training dataset comprised 5390 images, and the testing dataset included 125 different images. The neural network, k-NN, and SVM models demonstrated significantly higher relevance in terms of classification accuracy, precision, and recall compared to the random forest model. For instance, the F1 scores were 0.974, 0.987, and 0.897, respectively, for the former three models, in contrast to 0.661 for the random forest model.ConclusionThe performance metrics of the presented trained machine learning models hold promising prospects as potentially clinically useful aids.

Immune-based Machine learning Prediction of Diagnosis and Illness State in Schizophrenia and Bipolar Disorder

BackgroundSchizophrenia and bipolar disorder frequently face significant delay in diagnosis, leading to being missed or misdiagnosed in early stages. Both disorders have also been associated with trait and state immune abnormalities. Recent machine learning-based studies have shown encouraging results using diagnostic biomarkers in predictive models, but few have focused on immune-based markers. Our main objective was to develop supervised machine learning models to predict diagnosis and illness state in schizophrenia and bipolar disorder using only a panel of peripheral kynurenine metabolites and cytokines. MethodsThe cross-sectional I-GIVE cohort included hospitalized acute bipolar patients (n = 205), stable bipolar outpatients (n = 116), hospitalized acute schizophrenia patients (n = 111), stable schizophrenia outpatients (n = 75) and healthy controls (n = 185). Serum kynurenine metabolites, namely tryptophan (TRP), kynurenine (KYN), kynurenic acid (KA), quinaldic acid (QUINA), xanthurenic acid (XA), quinolinic acid (QUINO) and picolinic acid (PICO) were quantified using liquid chromatography-tandem mass spectrometry (LC-MS/MS), while V-plex Human Cytokine Assays were used to measure cytokines (interleukin-6 (IL-6), IL-8, IL-17, IL-12/IL23-P40, tumor necrosis factor-alpha (TNF-ɑ), interferon-gamma (IFN-γ)). Supervised machine learning models were performed using JMP Pro 17.0.0. We compared a primary analysis using nested cross-validation to a split set as sensitivity analysis. Post-hoc, we re-ran the models using only the significant features to obtain the key markers. ResultsThe models yielded a good Area Under the Curve (AUC) (0.804, Positive Prediction Value (PPV) = 86.95; Negative Prediction Value (NPV) = 54.61) for distinguishing all patients from controls. This implies that a positive test is highly accurate in identifying the patients, but a negative test is inconclusive. Both schizophrenia patients and bipolar patients could each be separated from controls with a good accuracy (SCZ AUC 0.824; BD AUC 0.802). Overall, increased levels of IL-6, TNF-ɑ and PICO and decreased levels of IFN-γ and QUINO were predictive for an individual being classified as a patient. Classification of acute versus stable patients reached a fair AUC of 0.713. The differentiation between schizophrenia and bipolar disorder yielded a poor AUC of 0.627. ConclusionsThis study highlights the potential of using immune-based measures to build predictive classification models in schizophrenia and bipolar disorder, with IL-6, TNF-ɑ, IFN-γ, QUINO and PICO as key candidates. While machine learning models successfully distinguished schizophrenia and bipolar disorder from controls, the challenges in differentiating schizophrenic from bipolar patients likely reflect shared immunological pathways by the both disorders and confounding by a larger state-specific effect. Larger multi-centric studies and multi-domain models are needed to enhance reliability and translation into clinic.

Synthetic lidar point cloud generation using deep generative models for improved driving scene object recognition

The imbalanced distribution of different object categories poses a challenge for training accurate object recognition models in driving scenes. Supervised machine learning models trained on imbalanced data are biased and easily overfit the majority classes, such as vehicles and pedestrians, which appear more frequently in driving scenes. We propose a novel data augmentation approach for object recognition in lidar point cloud of driving scenes, which leverages probabilistic generative models to produce synthetic point clouds for the minority classes and complement the original imbalanced dataset. We evaluate five generative models based on different statistical principles, including Gaussian mixture model, variational autoencoder, generative adversarial network, adversarial autoencoder and the diffusion model. Experiments with a real-world autonomous driving dataset show that the synthetic point clouds generated for the minority classes by the Latent Generative Adversarial Network result in significant improvement of object recognition performance for both minority and majority classes. The codes are available at https://github.com/AAAALEX-XIANG/Synthetic-Lidar-Generation.