Prediction using step-wise L1, L2 regularization and feature selection for small data sets with large number of features

MicroRNA classification and discovery for major depressive disorder diagnosis: Towards a robust and interpretable machine learning approach

BACKGROUND

Major depressive disorder (MDD) is notably underdiagnosed and undertreated due to its complex nature and subjective diagnostic methods. Biomarker identification would help provide a clearer understanding of MDD aetiology. Although machine learning (ML) has been implemented in previous studies to study the alteration of microRNA (miRNA) levels in MDD cases, clinical translation has not been feasible due to the lack of interpretability (i.e. too many miRNAs for consideration) and stability.

METHODS

This study applied logistic regression (LR) model to the blood miRNA expression profile to differentiate patients with MDD (n = 60) from healthy controls (HCs, n = 60). Embedded (L1-regularised logistic regression) feature selector was utilised to extract clinically relevant miRNAs, and optimized for clinical application.

RESULTS

Patients with MDD could be differentiated from HCs with the area under the receiver operating characteristic curve (AUC) of 0.81 on testing data when all available miRNAs were considered (which served as a benchmark). Our LR model selected miRNAs up to 5 (known as LR-5 model) emerged as the best model because it achieved a moderate classification ability (AUC = 0.75), relatively high interpretability (feature number = 5) and stability (ϕ̂Z=0.55) compared to the benchmark. The top-ranking miRNAs identified by our model have demonstrated associations with MDD pathways involving cytokine signalling in the immune system, the reelin signalling pathway, programmed cell death and cellular responses to stress.

CONCLUSION

The LR-5 model, which is optimised based on ML design factors, may lead to a robust and clinically usable MDD diagnostic tool.

Machine Learning-Based Prediction of Cyclic Voltammetry Behavior of Substitution of Zinc and Cobalt in BiFeO3/Bi25FeO40 for Supercapacitor Applications

Artificial intelligence and machine learning have become indispensable tools across various disciplines in the present century. In that way, the role of artificial intelligence and machine learning in energy storage devices was investigated. As a preliminary study, the data derived from electrochemical studies were used for the prediction. The prediction of current from cyclic voltammetry (CV) studies was undertaken for bismuth ferrite (BFO), substitution of zinc in BFO (BFZO), and substitution of cobalt in BFO composite (BFCO). CV is a vital electrochemical technique used for studying the electrochemical behavior of any material. The electrochemical study provides insights into the energy storage behavior of the material through the specific capacitance. The machine learning models, such as Artificial Neural Network (ANN), Random Forest (RF), and XGBoost (XGB), are trained and implemented to predict current at different scan rates. These models are trained and validated using the data collected from a CHI 600E electrochemical workstation. Multiple trials of experiments were performed to build the most optimum model for the material. The predicted values provide promising results and align well with the experimental data. The XGBoost, ANN and RF models perform well for the CV data set with an average testing accuracy >97%. Also, a meta-model was created using stacking of the above three machine learning models which further improved the predictive performance, achieving a slightly higher average testing accuracy of over 97.73%. The outcomes from the models can promote the development of machine learning applications in the field of electrochemistry and provide insights into optimizing supercapacitor performance and design through data-driven approaches.

Co-registered optical coherence tomography and X-ray angiography for the prediction of fractional flow reserve

Cardiovascular disease (CVD) stands as the leading global cause of mortality, and coronary artery disease (CAD) has the highest prevalence, contributing to 42% of these fatalities. Recognizing the constraints inherent in the anatomical assessment of CAD, Fractional Flow Reserve (FFR) has emerged as a pivotal functional diagnostic metric. Herein, we assess the potential of employing an ensemble approach with deep neural networks (DNN) to predict invasively measured Fractional Flow Reserve (FFR) using raw anatomical data extracted from both optical coherence tomography (OCT) and X-ray coronary angiography (XA). In this study, we used a challenging dataset, with 46% of the lesions falling within the FFR range of 0.75 to 0.85. Despite this complexity, our model achieved an accuracy of 84.3%, demonstrating a sensitivity of 87.5% and a specificity of 81.4%. Our results demonstrate that incorporating both OCT and XA signals, co-registered, as inputs for the DNN model leads to an important increase in overall accuracy.

Applications of peptide-based nanomaterials in targeting cancer therapy

To meet the demand for precision medicine, researchers are committed to developing novel strategies to reduce systemic toxicity and side effects in cancer treatment. Targeting peptides are widely applied due to their affinity and specificity, and their ability to be high-throughput screened, chemically synthesized and modified. More importantly, peptides can form ordered self-assembled structures through non-covalent supramolecular interactions, which can form nanostructures with different morphologies and functions, playing crucial roles in targeted diagnosis and treatment. Among them, in targeted immunotherapy, utilizing targeting peptides to block the binding between immune checkpoints and ligands, thereby activating the immune system to eliminate cancer cells, is an advanced therapeutic strategy. In this mini-review, we summarize the screening, self-assembly, and biomedical applications of targeting peptide-based nanomaterials. Furthermore, this mini-review summarizes the potential and optimization strategies of targeting peptides.

NTpred: a robust and precise machine learning framework for in silico identification of Tyrosine nitration sites in protein sequences

Post-translational modifications (PTMs) either enhance a protein's activity in various sub-cellular processes, or degrade their activity which leads toward failure of intracellular processes. Tyrosine nitration (NT) modification degrades protein's activity that initiates and propagates various diseases including neurodegenerative, cardiovascular, autoimmune diseases and carcinogenesis. Identification of NT modification supports development of novel therapies and drug discoveries for associated diseases. Identification of NT modification in biochemical labs is expensive, time consuming and error-prone. To supplement this process, several computational approaches have been proposed. However these approaches fail to precisely identify NT modification, due to the extraction of irrelevant, redundant and less discriminative features from protein sequences. This paper presents the NTpred framework that is competent in extracting comprehensive features from raw protein sequences using four different sequence encoders. To reap the benefits of different encoders, it generates four additional feature spaces by fusing different combinations of individual encodings. Furthermore, it eradicates irrelevant and redundant features from eight different feature spaces through a Recursive Feature Elimination process. Selected features of four individual encodings and four feature fusion vectors are used to train eight different Gradient Boosted Tree classifiers. The probability scores from the trained classifiers are utilized to generate a new probabilistic feature space, which is used to train a Logistic Regression classifier. On the BD1 benchmark dataset, the proposed framework outperforms the existing best-performing predictor in 5-fold cross validation and independent test evaluation with combined improvement of 13.7% in MCC and 20.1% in AUC. Similarly, on the BD2 benchmark dataset, the proposed framework outperforms the existing best-performing predictor with combined improvement of 5.3% in MCC and 1.0% in AUC. NTpred is publicly available for further experimentation and predictive use at: https://sds_genetic_analysis.opendfki.de/PredNTS/.

Advancing algorithmic drug product development: Recommendations for machine learning approaches in drug formulation

Artificial intelligence is a rapidly expanding area of research, with the disruptive potential to transform traditional approaches in the pharmaceutical industry, from drug discovery and development to clinical practice. Machine learning, a subfield of artificial intelligence, has fundamentally transformed in silico modelling and has the capacity to streamline clinical translation. This paper reviews data-driven modelling methodologies with a focus on drug formulation development. Despite recent advances, there is limited modelling guidance specific to drug product development and a trend towards suboptimal modelling practices, resulting in models that may not give reliable predictions in practice. There is an overwhelming focus on benchtop experimental outcomes obtained for a specific modelling aim, leaving the capabilities of data scraping or the use of combined modelling approaches yet to be fully explored. Moreover, the preference for high accuracy can lead to a reliance on black box methods over interpretable models. This further limits the widespread adoption of machine learning as black boxes yield models that cannot be easily understood for the purposes of enhancing product performance. In this review, recommendations for conducting machine learning research for drug product development to ensure trustworthiness, transparency, and reliability of the models produced are presented. Finally, possible future directions on how research in this area might develop are discussed to aim for models that provide useful and robust guidance to formulators.

Development and validation of risk prediction nomograms for acute respiratory failure in elderly patients with hip fracture

BACKGROUND

Hip fractures in the elderly often lead to acute respiratory failure, but there is currently no tool to assess the prognosis of such patients. This study aims to develop a risk prediction model for respiratory failure in these patients.

METHODS

A retrospective cross-sectional study was conducted using the Medical Information Mart for Intensive Care (MIMIC)-IV database, incorporating data from 3,266 patients with hip fractures aged over 55 years from 2008 to 2019. Data included demographic information, laboratory indicators, comorbidities, and treatment methods. Patients were divided into a training group (70%) and a validation group (30%). Least Absolute Shrinkage and Selection Operator (LASSO) regression was applied to select prognostic predictors, and a visualized nomogram model was constructed using multivariate logistic regression analysis. Model performance and clinical applicability were assessed. Statistical analyses were done using R4.2.2, with P < 0.05 deemed significant.

RESULTS

Seven key factors, including age, height, albumin, chloride, pneumonia, acute kidney injury (AKI), and heparin use, were associated with respiratory failure risk. The model demonstrated good performance with area under the curve (AUC) values of 0.77 and 0.73 in the training and validation sets, respectively. The calibration curve showed good agreement, and decision curve analysis (DCA) indicated the model's clinical benefit.

CONCLUSIONS

This risk prediction model can effectively predict respiratory failure in hip fracture patients, assisting clinicians in identifying high-risk individuals and providing evidence-based references for treatment strategies.

Fly ash degree of reaction in hypersaline NaCl and CaCl2 brines: Effects of calcium-based additives

The pozzolanic reaction of fly ashes with calcium-based additives can be effectively used to solidify and chemically stabilize (S&S process) highly concentrated brines inside a cementitious matrix. However, complex interactions between the fly ash, the additive, and the brine typically affect the phases formed at equilibrium, and the resulting solid capacity to successfully encapsulate the brine and its contaminants. Here, the performances of two types of fly ash (a Class C and Class F fly ash) are assessed when combined with different additives (two types of cement, or lime with and without NaAlO2), and two types of brine (NaCl or CaCl2) over a range of concentrations (0 ≤ [Cl-] ≤ 2 M). The best performing matrices - i.e., the matrices with the highest Cl-containing phases content - were identified using XRD and TGA. The experimental results were then combined with thermodynamic modeling to dissociate the contribution of the fly ash from that of the additives. All results were implemented in a machine learning model that showed good accuracy at predicting the fly ash degree of reaction, allowing for the robust prediction of extended systems performance when combined with thermodynamic modeling.

Empirical Study of Overfitting in Deep Learning for Predicting Breast Cancer Metastasis

Overfitting may affect the accuracy of predicting future data because of weakened generalization. In this research, we used an electronic health records (EHR) dataset concerning breast cancer metastasis to study the overfitting of deep feedforward neural networks (FNNs) prediction models. We studied how each hyperparameter and some of the interesting pairs of hyperparameters were interacting to influence the model performance and overfitting. The 11 hyperparameters we studied were activate function, weight initializer, number of hidden layers, learning rate, momentum, decay, dropout rate, batch size, epochs, L1, and L2. Our results show that most of the single hyperparameters are either negatively or positively corrected with model prediction performance and overfitting. In particular, we found that overfitting overall tends to negatively correlate with learning rate, decay, batch size, and L2, but tends to positively correlate with momentum, epochs, and L1. According to our results, learning rate, decay, and batch size may have a more significant impact on both overfitting and prediction performance than most of the other hyperparameters, including L1, L2, and dropout rate, which were designed for minimizing overfitting. We also find some interesting interacting pairs of hyperparameters such as learning rate and momentum, learning rate and decay, and batch size and epochs.

Variation in the infant health effects of the women, infants, and children program by predicted risk using novel machine learning methods

The Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) has an extensive literature documenting positive effects on infant health outcomes, specifically preterm birth, low birthweight, small size for gestational age, and infant mortality. However, existing studies focus on average effects for these relatively infrequent outcomes, thus providing no evidence for how WIC affects those at greatest risk of negative infant health outcomes. Our study focuses on documenting how WIC's infant health effects vary by level of risk. In doing so, we leverage a uniquely rich database describing maternal and infant outcomes and risk factors. Additionally, we use high dimensional data to generate predictions of risk and combine these predictions with the novel double machine learning method to stratify the effects of WIC by predicted risk. Our estimates of WIC's average treatment effects align with those in the existing literature. More importantly, we document significant variation in the effects of WIC on infant health by predicted risk level. Our results show that WIC is most beneficial among those at greatest risk of poor outcomes.

A pairwise strategy for imputing predictive features when combining multiple datasets

MOTIVATION

In the training of predictive models using high-dimensional genomic data, multiple studies' worth of data are often combined to increase sample size and improve generalizability. A drawback of this approach is that there may be different sets of features measured in each study due to variations in expression measurement platform or technology. It is often common practice to work only with the intersection of features measured in common across all studies, which results in the blind discarding of potentially useful feature information that is measured in individual or subsets of studies.

RESULTS

We characterize the loss in predictive performance incurred by using only the intersection of feature information available across all studies when training predictors using gene expression data from microarray and sequencing datasets. We study the properties of linear and polynomial regression for imputing discarded features and demonstrate improvements in the external performance of prediction functions through simulation and in gene expression data collected on breast cancer patients. To improve this process, we propose a pairwise strategy that applies any imputation algorithm to two studies at a time and averages imputed features across pairs. We demonstrate that the pairwise strategy is preferable to first merging all datasets together and imputing any resulting missing features. Finally, we provide insights on which subsets of intersected and study-specific features should be used so that missing-feature imputation best promotes cross-study replicability.

AVAILABILITY AND IMPLEMENTATION

The code is available at https://github.com/YujieWuu/Pairwise_imputation.

SUPPLEMENTARY INFORMATION

Supplementary information is available at Bioinformatics online.

Analysis of Parkinson's Disease Using an Imbalanced-Speech Dataset by Employing Decision Tree Ensemble Methods

Parkinson's disease (PD) currently affects approximately 10 million people worldwide. The detection of PD positive subjects is vital in terms of disease prognostics, diagnostics, management and treatment. Different types of early symptoms, such as speech impairment and changes in writing, are associated with Parkinson disease. To classify potential patients of PD, many researchers used machine learning algorithms in various datasets related to this disease. In our research, we study the dataset of the PD vocal impairment feature, which is an imbalanced dataset. We propose comparative performance evaluation using various decision tree ensemble methods, with or without oversampling techniques. In addition, we compare the performance of classifiers with different sizes of ensembles and various ratios of the minority class and the majority class with oversampling and undersampling. Finally, we combine feature selection with best-performing ensemble classifiers. The result shows that AdaBoost, random forest, and decision tree developed for the RUSBoost imbalanced dataset perform well in performance metrics such as precision, recall, F1-score, area under the receiver operating characteristic curve (AUROC) and the geometric mean. Further, feature selection methods, namely lasso and information gain, were used to screen the 10 best features using the best ensemble classifiers. AdaBoost with information gain feature selection method is the best performing ensemble method with an F1-score of 0.903.

Validating administrative data to identify complex surgical site infections following cardiac implantable electronic device implantation: a comparison of traditional methods and machine learning

Background

Cardiac implantable electronic device (CIED) surgical site infections (SSIs) have been outpacing the increases in implantation of these devices. While traditional surveillance of these SSIs by infection prevention and control would likely be the most accurate, this is not practical in many centers where resources are constrained. Therefore, we explored the validity of administrative data at identifying these SSIs.

Methods

We used a cohort of all patients with CIED implantation in Calgary, Alberta where traditional surveillance was done for infections from Jan 1, 2013 to December 31, 2019. We used this infection subgroup as our “gold standard” and then utilized various combinations of administrative data to determine which best optimized the sensitivity and specificity at identifying infection. We evaluated six approaches to identifying CIED infection using administrative data, which included four algorithms using International Classification of Diseases codes and/or Canadian Classification of Health Intervention codes, and two machine learning models. A secondary objective of our study was to assess if machine learning techniques with training of logistic regression models would outperform our pre-selected codes.

Results

We determined that all of the pre-selected algorithms performed well at identifying CIED infections but the machine learning model was able to produce the optimal method of identification with an area under the receiver operating characteristic curve (AUC) of 96.8%. The best performing pre-selected algorithm yielded an AUC of 94.6%.

Conclusions

Our findings suggest that administrative data can be used to effectively identify CIED infections. While machine learning performed the most optimally, in centers with limited analytic capabilities a simpler algorithm of pre-selected codes also has excellent yield. This can be valuable for centers without traditional surveillance to follow trends in SSIs over time and identify when rates of infection are increasing. This can lead to enhanced interventions for prevention of SSIs.

Towards a Deep-Learning Approach for Prediction of Fractional Flow Reserve from Optical Coherence Tomography

Cardiovascular disease (CVD) is the number one cause of death worldwide, and coronary artery disease (CAD) is the most prevalent CVD, accounting for 42% of these deaths. In view of the limitations of the anatomical evaluation of CAD, Fractional Flow Reserve (FFR) has been introduced as a functional diagnostic index. Herein, we evaluate the feasibility of using deep neural networks (DNN) in an ensemble approach to predict the invasively measured FFR from raw anatomical information that is extracted from optical coherence tomography (OCT). We evaluate the performance of various DNN architectures under different formulations: regression, classification—standard, and few-shot learning (FSL) on a dataset containing 102 intermediate lesions from 80 patients. The FSL approach that is based on a convolutional neural network leads to slightly better results compared to the standard classification: the per-lesion accuracy, sensitivity, and specificity were 77.5%, 72.9%, and 81.5%, respectively. However, since the 95% confidence intervals overlap, the differences are statistically not significant. The main findings of this study can be summarized as follows: (1) Deep-learning (DL)-based FFR prediction from reduced-order raw anatomical data is feasible in intermediate coronary artery lesions; (2) DL-based FFR prediction provides superior diagnostic performance compared to baseline approaches that are based on minimal lumen diameter and percentage diameter stenosis; and (3) the FFR prediction performance increases quasi-linearly with the dataset size, indicating that a larger train dataset will likely lead to superior diagnostic performance.

CTT: CNN Meets Transformer for Tracking

Siamese networks are one of the most popular directions in the visual object tracking based on deep learning. In Siamese networks, the feature pyramid network (FPN) and the cross-correlation complete feature fusion and the matching of features extracted from the template and search branch, respectively. However, object tracking should focus on the global and contextual dependencies. Hence, we introduce a delicate residual transformer structure which contains a self-attention mechanism called encoder-decoder into our tracker as the part of neck. Under the encoder-decoder structure, the encoder promotes the interaction between the low-level features extracted from the target and search branch by the CNN to obtain global attention information, while the decoder replaces cross-correlation to send global attention information into the head module. We add a spatial and channel attention component in the target branch, which can further improve the accuracy and robustness of our proposed model for a low price. Finally, we detailly evaluate our tracker CTT on GOT-10k, VOT2019, OTB-100, LaSOT, NfS, UAV123 and TrackingNet benchmarks, and our proposed method obtains competitive results with the state-of-the-art algorithms.

Utility of Cognitive Neural Features for Predicting Mental Health Behaviors

Cognitive dysfunction underlies common mental health behavioral symptoms including depression, anxiety, inattention, and hyperactivity. In this study of 97 healthy adults, we aimed to classify healthy vs. mild-to-moderate self-reported symptoms of each disorder using cognitive neural markers measured with an electroencephalography (EEG). We analyzed source-reconstructed EEG data for event-related spectral perturbations in the theta, alpha, and beta frequency bands in five tasks, a selective attention and response inhibition task, a visuospatial working memory task, a Flanker interference processing task, and an emotion interference task. From the cortical source activation features, we derived augmented features involving co-activations between any two sources. Logistic regression on the augmented feature set, but not the original feature set, predicted the presence of psychiatric symptoms, particularly for anxiety and inattention with >80% sensitivity and specificity. We also computed current flow closeness and betweenness centralities to identify the “hub” source signal predictors. We found that the Flanker interference processing task was the most useful for assessing the connectivity hubs in general, followed by the inhibitory control go-nogo paradigm. Overall, these interpretable machine learning analyses suggest that EEG biomarkers collected on a rapid suite of cognitive assessments may have utility in classifying diverse self-reported mental health symptoms.

Evolutionary Sparse Learning for Phylogenomics

We introduce a supervised machine learning approach with sparsity constraints for phylogenomics, referred to as evolutionary sparse learning (ESL). ESL builds models with genomic loci—such as genes, proteins, genomic segments, and positions—as parameters. Using the Least Absolute Shrinkage and Selection Operator, ESL selects only the most important genomic loci to explain a given phylogenetic hypothesis or presence/absence of a trait. ESL models do not directly involve conventional parameters such as rates of substitutions between nucleotides, rate variation among positions, and phylogeny branch lengths. Instead, ESL directly employs the concordance of variation across sequences in an alignment with the evolutionary hypothesis of interest. ESL provides a natural way to combine different molecular and nonmolecular data types and incorporate biological and functional annotations of genomic loci in model building. We propose positional, gene, function, and hypothesis sparsity scores, illustrate their use through an example, and suggest several applications of ESL. The ESL framework has the potential to drive the development of a new class of computational methods that will complement traditional approaches in evolutionary genomics, particularly for identifying influential loci and sequences given a phylogeny and building models to test hypotheses. ESL’s fast computational times and small memory footprint will also help democratize big data analytics and improve scientific rigor in phylogenomics.

Application of the Machine Learning LightGBM Model to the Prediction of the Water Levels of the Lower Columbia River

Due to the strong nonlinear interaction with river discharge, tides in estuaries are characterised as nonstationary and their mechanisms are yet to be fully understood. It remains highly challenging to accurately predict estuarine water levels. Machine learning methods, which offer a unique ability to simulate the unknown relationships between variables, have been increasingly used in a large number of research areas. This study applies the LightGBM model to predicting the water levels along the lower reach of the Columbia River. The model inputs consist of the discharges from two upstream rivers (Columbia and Willamette Rivers) and the tide characteristics, including the tide range at the estuary mouth (Astoria) and tide constituents. The model is optimized with the selected parameters. The results show that the LightGBM model can achieve high prediction accuracy, with the root-mean-square-error values of water level being reduced to 0.14 m and the correlation coefficient and skill score being in the ranges of 0.975–0.987 and 0.941–0.972, respectively, which are statistically better than those obtained from physics-based models such as the nonstationary tidal harmonic analysis model (NS_TIDE). The importance of subtide constituents in interacting with the river discharge in the estuary is clearly revealed from the model results.

Tracking and Monitoring Mood Stability of Patients With Major Depressive Disorder by Machine Learning Models Using Passive Digital Data: Prospective Naturalistic Multicenter Study

BACKGROUND

Major depressive disorder (MDD) is a common mental illness characterized by persistent sadness and a loss of interest in activities. Using smartphones and wearable devices to monitor the mental condition of patients with MDD has been examined in several studies. However, few studies have used passively collected data to monitor mood changes over time.

OBJECTIVE

The aim of this study is to examine the feasibility of monitoring mood status and stability of patients with MDD using machine learning models trained by passively collected data, including phone use data, sleep data, and step count data.

METHODS

We constructed 950 data samples representing time spans during three consecutive Patient Health Questionnaire-9 assessments. Each data sample was labeled as Steady or Mood Swing, with subgroups Steady-remission, Steady-depressed, Mood Swing-drastic, and Mood Swing-moderate based on patients' Patient Health Questionnaire-9 scores from three visits. A total of 252 features were extracted, and 4 feature selection models were applied; 6 different combinations of types of data were experimented with using 6 different machine learning models.

RESULTS

A total of 334 participants with MDD were enrolled in this study. The highest average accuracy of classification between Steady and Mood Swing was 76.67% (SD 8.47%) and that of recall was 90.44% (SD 6.93%), with features from all types of data being used. Among the 6 combinations of types of data we experimented with, the overall best combination was using call logs, sleep data, step count data, and heart rate data. The accuracies of predicting between Steady-remission and Mood Swing-drastic, Steady-remission and Mood Swing-moderate, and Steady-depressed and Mood Swing-drastic were over 80%, and the accuracy of predicting between Steady-depressed and Mood Swing-moderate and the overall Steady to Mood Swing classification accuracy were over 75%. Comparing all 6 aforementioned combinations, we found that the overall prediction accuracies between Steady-remission and Mood Swing (drastic and moderate) are better than those between Steady-depressed and Mood Swing (drastic and moderate).

CONCLUSIONS

Our proposed method could be used to monitor mood changes in patients with MDD with promising accuracy by using passively collected data, which can be used as a reference by doctors for adjusting treatment plans or for warning patients and their guardians of a relapse.

TRIAL REGISTRATION

Chinese Clinical Trial Registry ChiCTR1900021461; http://www.chictr.org.cn/showprojen.aspx?proj=36173.

Reaching the End-Game for GWAS: Machine Learning Approaches for the Prioritization of Complex Disease Loci

Genome-wide association studies (GWAS) have revealed thousands of genetic loci that underpin the complex biology of many human traits. However, the strength of GWAS - the ability to detect genetic association by linkage disequilibrium (LD) - is also its limitation. Whilst the ever-increasing study size and improved design have augmented the power of GWAS to detect effects, differentiation of causal variants or genes from other highly correlated genes associated by LD remains the real challenge. This has severely hindered the biological insights and clinical translation of GWAS findings. Although thousands of disease susceptibility loci have been reported, causal genes at these loci remain elusive. Machine learning (ML) techniques offer an opportunity to dissect the heterogeneity of variant and gene signals in the post-GWAS analysis phase. ML models for GWAS prioritization vary greatly in their complexity, ranging from relatively simple logistic regression approaches to more complex ensemble models such as random forests and gradient boosting, as well as deep learning models, i.e., neural networks. Paired with functional validation, these methods show important promise for clinical translation, providing a strong evidence-based approach to direct post-GWAS research. However, as ML approaches continue to evolve to meet the challenge of causal gene identification, a critical assessment of the underlying methodologies and their applicability to the GWAS prioritization problem is needed. This review investigates the landscape of ML applications in three parts: selected models, input features, and output model performance, with a focus on prioritizations of complex disease associated loci. Overall, we explore the contributions ML has made towards reaching the GWAS end-game with consequent wide-ranging translational impact.

Data science and machine learning in anesthesiology

Machine learning (ML) is revolutionizing anesthesiology research. Unlike classical research methods that are largely inference-based, ML is geared more towards making accurate predictions. ML is a field of artificial intelligence concerned with developing algorithms and models to perform prediction tasks in the absence of explicit instructions. Most ML applications, despite being highly variable in the topics that they deal with, generally follow a common workflow. For classification tasks, a researcher typically tests various ML models and compares the predictive performance with the reference logistic regression model. The main advantage of ML lies in its ability to deal with many features with complex interactions and its specific focus on maximizing predictive performance. However, emphasis on data-driven prediction can sometimes neglect mechanistic understanding. This article mainly focuses on the application of supervised ML to electronic health record (EHR) data. The main limitation of EHR-based studies is in the difficulty of establishing causal relationships. However, the associated low cost and rich information content provide great potential to uncover hitherto unknown correlations. In this review, the basic concepts of ML are introduced along with important terms that any ML researcher should know. Practical tips regarding the choice of software and computing devices are also provided. Towards the end, several examples of successful ML applications in anesthesiology are discussed. The goal of this article is to provide a basic roadmap to novice ML researchers working in the field of anesthesiology.

Parallel Implementation on FPGA of Support Vector Machines Using Stochastic Gradient Descent

Sequential Minimal Optimization (SMO) is the traditional training algorithm for Support Vector Machines (SVMs). However, SMO does not scale well with the size of the training set. For that reason, Stochastic Gradient Descent (SGD) algorithms, which have better scalability, are a better option for massive data mining applications. Furthermore, even with the use of SGD, training times can become extremely large depending on the data set. For this reason, accelerators such as Field-programmable Gate Arrays (FPGAs) are used. This work describes an implementation in hardware, using FPGA, of a fully parallel SVM using Stochastic Gradient Descent. The proposed FPGA implementation of an SVM with SGD presents speedups of more than 10,000× relative to software implementations running on a quad-core processor and up to 319× compared to state-of-the-art FPGA implementations while requiring fewer hardware resources. The results show that the proposed architecture is a viable solution for highly demanding problems such as those present in big data analysis.

Predicting tachycardia as a surrogate for instability in the intensive care unit

Tachycardia is a strong though non-specific marker of cardiovascular stress that proceeds hemodynamic instability. We designed a predictive model of tachycardia using multi-granular intensive care unit (ICU) data by creating a risk score and dynamic trajectory. A subset of clinical and numerical signals were extracted from the Multiparameter Intelligent Monitoring in Intensive Care II database. A tachycardia episode was defined as heart rate ≥ 130/min lasting for ≥ 5 min, with ≥ 10% density. Regularized logistic regression (LR) and random forest (RF) classifiers were trained to create a risk score for upcoming tachycardia. Three different risk score models were compared for tachycardia and control (non-tachycardia) groups. Risk trajectory was generated from time windows moving away at 1 min increments from the tachycardia episode. Trajectories were computed over 3 hours leading up to the episode for three different models. From 2809 subjects, 787 tachycardia episodes and 707 control periods were identified. Patients with tachycardia had increased vasopressor support, longer ICU stay, and increased ICU mortality than controls. In model evaluation, RF was slightly superior to LR, which accuracy ranged from 0.847 to 0.782, with area under the curve from 0.921 to 0.842. Risk trajectory analysis showed average risks for tachycardia group evolved to 0.78 prior to the tachycardia episodes, while control group risks remained < 0.3. Among the three models, the internal control model demonstrated evolving trajectory approximately 75 min before tachycardia episode. Clinically relevant tachycardia episodes can be predicted from vital sign time series using machine learning algorithms.

Prediction of RNA- and DNA-Binding Proteins Using Various Machine Learning Classifiers

BACKGROUND

Nucleic acid-binding proteins play major roles in different biological processes, such as transcription, splicing and translation. Therefore, the nucleic acid-binding function prediction of proteins is a step toward full functional annotation of proteins. The aim of our research was the improvement of nucleic-acid binding function prediction.

METHODS

In the current study, nine machine-learning algorithms were used to predict RNA- and DNA-binding proteins and also to discriminate between RNA-binding proteins and DNA-binding proteins. The electrostatic features were utilized for prediction of each function in corresponding adapted protein datasets. The leave-one-out cross-validation process was used to measure the performance of employed classifiers.

RESULTS

Radial basis function classifier gave the best results in predicting RNA- and DNA-binding proteins in comparison with other classifiers applied. In discriminating between RNA- and DNA-binding proteins, multilayer perceptron classifier was the best one.

CONCLUSION

Our findings show that the prediction of nucleic acid-binding function based on these simple electrostatic features can be improved by applied classifiers. Moreover, a reasonable progress to distinguish between RNA- and DNA-binding proteins has been achieved.

Predicting Unplanned Transfers to the Intensive Care Unit: A Machine Learning Approach Leveraging Diverse Clinical Elements

Background

Early warning scores aid in the detection of pediatric clinical deteriorations but include limited data inputs, rarely include data trends over time, and have limited validation.

Objective

Machine learning methods that make use of large numbers of predictor variables are now commonplace. This work examines how different types of predictor variables derived from the electronic health record affect the performance of predicting unplanned transfers to the intensive care unit (ICU) at three large children’s hospitals.

Methods

We trained separate models with data from three different institutions from 2011 through 2013 and evaluated models with 2014 data. Cases consisted of patients who transferred from the floor to the ICU and met one or more of 5 different priori defined criteria for suspected unplanned transfers. Controls were patients who were never transferred to the ICU. Predictor variables for the models were derived from vitals, labs, acuity scores, and nursing assessments. Classification models consisted of L1 and L2 regularized logistic regression and neural network models. We evaluated model performance over prediction horizons ranging from 1 to 16 hours.

Results

Across the three institutions, the c-statistic values for our best models were 0.892 (95% CI 0.875-0.904), 0.902 (95% CI 0.880-0.923), and 0.899 (95% CI 0.879-0.919) for the task of identifying unplanned ICU transfer 6 hours before its occurrence and achieved 0.871 (95% CI 0.855-0.888), 0.872 (95% CI 0.850-0.895), and 0.850 (95% CI 0.825-0.875) for a prediction horizon of 16 hours. For our first model at 80% sensitivity, this resulted in a specificity of 80.5% (95% CI 77.4-83.7) and a positive predictive value of 5.2% (95% CI 4.5-6.2).

Conclusions

Feature-rich models with many predictor variables allow for patient deterioration to be predicted accurately, even up to 16 hours in advance.

Support vector-based Takagi-Sugeno fuzzy system for the prediction of binding affinity of peptides

High dimensional, complex and non-linear nature of the post-genome data often adversely affects the performance of predictive models. There are two methods that have been widely used to model such non-linear systems, namely Fuzzy System (FS) and Support Vector Machine (SVM). FS is good at modelling uncertainty and yielding a set of interpretable IF-THEN rules, but suffers from the curse of dimensionality whereas SVM is a method that has been shown to effectively deal with large number of dimensions leading to better generalization ability. In this paper, a hybrid system is therefore proposed to improve FS with the aid of SVM-based regression method and successfully applied to the prediction of binding affinity of peptides, which is regarded as one of the most complex modelling problems in the post-genome era due to the diversity of peptides discovered. The proposed hybrid method yields comparatively better results than what has been presented in the recently published papers, therefore can also be considered for other bioinformatics applications.

Predicting antitumor activity of peptides by consensus of regression models trained on a small data sample

Predicting antitumor activity of compounds using regression models trained on a small number of compounds with measured biological activity is an ill-posed inverse problem. Yet, it occurs very often within the academic community. To counteract, up to some extent, overfitting problems caused by a small training data, we propose to use consensus of six regression models for prediction of biological activity of virtual library of compounds. The QSAR descriptors of 22 compounds related to the opioid growth factor (OGF, Tyr-Gly-Gly-Phe-Met) with known antitumor activity were used to train regression models: the feed-forward artificial neural network, the k-nearest neighbor, sparseness constrained linear regression, the linear and nonlinear (with polynomial and Gaussian kernel) support vector machine. Regression models were applied on a virtual library of 429 compounds that resulted in six lists with candidate compounds ranked by predicted antitumor activity. The highly ranked candidate compounds were synthesized, characterized and tested for an antiproliferative activity. Some of prepared peptides showed more pronounced activity compared with the native OGF; however, they were less active than highly ranked compounds selected previously by the radial basis function support vector machine (RBF SVM) regression model. The ill-posedness of the related inverse problem causes unstable behavior of trained regression models on test data. These results point to high complexity of prediction based on the regression models trained on a small data sample.