Supervised Machine Learning Methodology Research Articles

Forecasting Hydrogen Vehicle Refuelling for Sustainable Transportation: A Light Gradient-Boosting Machine Model

Forecasting Hydrogen Vehicle Refuelling for Sustainable Transportation: A Light Gradient-Boosting Machine Model

Comparing Clinician Estimates versus a Statistical Tool for Predicting Risk of Death within 45 Days of Admission for Cancer Patients.

While clinical practice guidelines recommend that oncologists discuss goals of care with patients who have advanced cancer, it is estimated that less than 20% of individuals admitted to the hospital with high-risk cancers have end-of-life discussions with their providers. While there has been interest in developing models for mortality prediction to trigger such discussions, few studies have compared how such models compare with clinical judgment to determine a patient's mortality risk. This study is a prospective analysis of 1,069 solid tumor medical oncology hospital admissions (n = 911 unique patients) from February 7 to June 7, 2022, at Memorial Sloan Kettering Cancer Center. Electronic surveys were sent to hospitalists, advanced practice providers, and medical oncologists the first afternoon following a hospital admission and they were asked to estimate the probability that the patient would die within 45 days. Provider estimates of mortality were compared with those from a predictive model developed using a supervised machine learning methodology, and incorporated routine laboratory, demographic, biometric, and admission data. Area under the receiver operating characteristic curve (AUC), calibration and decision curves were compared between clinician estimates and the model predictions. Within 45 days following hospital admission, 229 (25%) of 911 patients died. The model performed better than the clinician estimates (AUC 0.834 vs. 0.753, p < 0.0001). Integrating clinician predictions with the model's estimates further increased the AUC to 0.853 (p < 0.0001). Clinicians overestimated risk whereas the model was extremely well-calibrated. The model demonstrated net benefit over a wide range of threshold probabilities. The inpatient prognosis at admission model is a robust tool to assist clinical providers in evaluating mortality risk, and it has recently been implemented in the electronic medical record at our institution to improve end-of-life care planning for hospitalized cancer patients.

Predicting Prolonged Length of Hospital Stay and Identifying Risk Factors Following Total Ankle Arthroplasty: A Supervised Machine Learning Methodology

Ankle osteoarthritis (OA) is a debilitating condition that arises as a result of trauma or injury to the ankle and often progresses to chronic pain and loss of function that may require surgical intervention. Total ankle arthroplasty (TAA) has emerged as a means of operative treatment for end-stage ankle OA. Increased hospital length of stay (LOS) is a common adverse postoperative outcome that increases both the complications and cost of care associated with arthroplasty procedures. The purpose of this study was to employ four machine learning (ML) algorithms to predict LOS in patients undergoing TAA using the American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database. The ACS-NSQIP database was queried to identify adult patients undergoing elective TAA from 2008 to 2018. Four supervised ML classification algorithms were utilized and tasked with predicting increased hospital length of stay (LOS). Among these variables, female sex, ASA Class III, preoperative sodium, preoperative hematocrit, diabetes, preoperative creatinine, other arthritis, BMI, preoperative WBC, and Hispanic ethnicity carried the highest importance across predictions generated by 4 independent ML algorithms. Predictions generated by these algorithms were made with an average AUC of 0.7257, as well as an average accuracy of 73.98% and an average sensitivity and specificity of 48.47% and 79.38%, respectively. These findings may be useful for guiding decision-making within the perioperative period and may serve to identify patients at increased risk for a prolonged LOS.

Smart science: How artificial intelligence is revolutionizing pharmaceutical medicine

Abstract Artificial intelligence (AI) is a discipline within the field of computer science that encompasses the development and utilization of machines capable of emulating human behavior, particularly regarding the astute examination and interpretation of data. AI operates through the utilization of specialized algorithms, and it includes techniques such as deep (DL), and machine learning (ML), and natural language processing (NLP). As a result, AI has found its application in the study of pharmaceutical chemistry and healthcare. The AI models employed encompass a spectrum of methodologies, including unsupervised clustering techniques applied to drugs or patients to discern potential drug compounds or appropriate patient cohorts. Additionally, supervised ML methodologies are utilized to enhance the efficacy of therapeutic drug monitoring. Further, AI-aided prediction of the clinical outcomes of clinical trials can improve efficiency by prioritizing therapeutic intervention that are likely to succeed, hence benefiting the patient. AI may also help create personalized treatments by locating potential intervention targets and assessing their efficacy. Hence, this review provides insights into recent advances in the application of AI and different tools used in the field of pharmaceutical medicine.

Improving infection prevention briefing through predictive predesign: a computational approach to architectural programming by evaluating socioecological risk factors

ABSTRACT This research aimed to use a computational approach to aid infection prevention strategies during the predesign phase of healthcare environment planning. The objective of this study was to decompose and analyze health system catchment area ‘outside design basis factors’ such as demographic, public health, and environmental characteristics that may contribute to infection causing pathogen spread in healthcare environments using state of Florida data as a context for analysis. Supervised Machine Learning methodologies were used to determine relationships between Socioecological System (SES) factors and observed incidence rates of specific Hospital Acquired Infections (HAI). Results of the analysis suggest that relevant regional population SES factors, when included as independent variables, can forecast the potential co-occurrence of specific types of HAI in regional health system service areas. Outcomes of the analysis suggest that demographic and regional characteristics are significantly related to the predictive incidence of Antimicrobial-resistant infections in healthcare settings in Florida Agency for Healthcare Administration (ACHA) regions. The outcomes of this research suggest the benefit of using SES computational modeling for decision-making in health system Architectural Programming and predesign. This process allows for manifesting meaningful patterns by applying quantitative methods to analyze patient population characteristics that may influence the prevalence of pathogen spread within the interior environments of healthcare inpatient settings. This information contributes to a deeper understanding of context-specific priorities for operationalizing human-centered Infection Prevention through Design (PtD) infrastructure and planning.

Extending regional habitat classification systems to ocean basin scale using predicted species distributions as proxies

The patchy nature and overall scarcity of available scientific data poses a challenge to holistic ecosystem-based management that considers the whole range of ecological, social, and economic aspects that affect ecosystem health and productivity in the deep sea. In particular, the evaluation of, for instance, the impact of human activities/climate change, the adequacy and representativity of MPA networks, and the valuation of ecosystem goods and services is hampered by the lack of detailed seafloor habitat maps and a univocal classification system. To maximize the use of current evidence-based management decision tools, this paper investigates the potential application of a supervised machine learning methodology to expand a well-established habitat classification system throughout an entire ocean basin. A multi-class Random Forest habitat classification model was built using the predicted distributions of 6 deep-sea fish and 6 cold-water corals as predictor variables (proxies). This model, found to correctly classify the area covered by an existing European seabed habitat classification system with ~90% accuracy, was used to provide a univocal deep-sea habitat classification for the North Atlantic. Until such time as global seabed mapping projects are complete, supervised machine learning approaches, as described here, can provide the full coverage classified maps and preliminary habitat inventories needed to underpin marine management decision making.

Exploring the role of ecology and social organisation in agropastoral societies: A Bayesian network approach

The present contribution focuses on investigating the interaction of people and environment in small-scale farming societies. Our study is centred on the particular way settlement location constraints economic strategy when technology is limited, and social division of work is not fully developed. Our intention is to investigate prehistoric socioeconomic organisation when farming began in the Old World along the Levant shores of Iberian Peninsula, the Neolithic phenomenon. We approach this subject extracting relevant information from a big set of ethnographic and ethnoarchaeological cases using Machine Learning methods. This paper explores the use of Bayesian networks as explanatory models of the independent variables–the environment- and dependent variables–social decisions-, and also as predictive models. The study highlights how subsistence strategies are modified by ecological and topographical variables of the settlement location and their relationship with social organisation. It also establishes the role of Bayesian networks as a suitable supervised Machine Learning methodology for investigating socio-ecological systems, introducing their use to build useful data-driven models to address relevant archaeological and anthropological questions.

Single nucleotide polymorphism genes and mitochondrial DNA haplogroups as biomarkers for early prediction of knee osteoarthritis structural progressors: use of supervised machine learning classifiers

BackgroundKnee osteoarthritis is the most prevalent chronic musculoskeletal debilitating disease. Current treatments are only symptomatic, and to improve this, we need a robust prediction model to stratify patients at an early stage according to the risk of joint structure disease progression. Some genetic factors, including single nucleotide polymorphism (SNP) genes and mitochondrial (mt)DNA haplogroups/clusters, have been linked to this disease. For the first time, we aim to determine, by using machine learning, whether some SNP genes and mtDNA haplogroups/clusters alone or combined could predict early knee osteoarthritis structural progressors.MethodsParticipants (901) were first classified for the probability of being structural progressors. Genotyping included SNP genes TP63, FTO, GNL3, DUS4L, GDF5, SUPT3H, MCF2L, and TGFA; mtDNA haplogroups H, J, T, Uk, and others; and clusters HV, TJ, KU, and C-others. They were considered for prediction with major risk factors of osteoarthritis, namely, age and body mass index (BMI). Seven supervised machine learning methodologies were evaluated. The support vector machine was used to generate gender-based models. The best input combination was assessed using sensitivity and synergy analyses. Validation was performed using tenfold cross-validation and an external cohort (TASOAC).ResultsFrom 277 models, two were defined. Both used age and BMI in addition for the first one of the SNP genes TP63, DUS4L, GDF5, and FTO with an accuracy of 85.0%; the second profits from the association of mtDNA haplogroups and SNP genes FTO and SUPT3H with 82.5% accuracy. The highest impact was associated with the haplogroup H, the presence of CT alleles for rs8044769 at FTO, and the absence of AA for rs10948172 at SUPT3H. Validation accuracy with the cross-validation (about 95%) and the external cohort (90.5%, 85.7%, respectively) was excellent for both models.ConclusionsThis study introduces a novel source of decision support in precision medicine in which, for the first time, two models were developed consisting of (i) age, BMI, TP63, DUS4L, GDF5, and FTO and (ii) the optimum one as it has one less variable: age, BMI, mtDNA haplogroup, FTO, and SUPT3H. Such a framework is translational and would benefit patients at risk of structural progressive knee osteoarthritis.

Machine Learning Methods for Hypercholesterolemia Long-Term Risk Prediction.

Cholesterol is a waxy substance found in blood lipids. Its role in the human body is helpful in the process of producing new cells as long as it is at a healthy level. When cholesterol exceeds the permissible limits, it works the opposite, causing serious heart health problems. When a person has high cholesterol (hypercholesterolemia), the blood vessels are blocked by fats, and thus, circulation through the arteries becomes difficult. The heart does not receive the oxygen it needs, and the risk of heart attack increases. Nowadays, machine learning (ML) has gained special interest from physicians, medical centers and healthcare providers due to its key capabilities in health-related issues, such as risk prediction, prognosis, treatment and management of various conditions. In this article, a supervised ML methodology is outlined whose main objective is to create risk prediction tools with high efficiency for hypercholesterolemia occurrence. Specifically, a data understanding analysis is conducted to explore the features association and importance to hypercholesterolemia. These factors are utilized to train and test several ML models to find the most efficient for our purpose. For the evaluation of the ML models, precision, recall, accuracy, F-measure, and AUC metrics have been taken into consideration. The derived results highlighted Soft Voting with Rotation and Random Forest trees as base models, which achieved better performance in comparison to the other models with an AUC of 94.5%, precision of 92%, recall of 91.8%, F-measure of 91.7% and an accuracy equal to 91.75%.

Genes and regulatory mechanisms associated with experimentally-induced bovine respiratory disease identified using supervised machine learning methodology

Bovine respiratory disease (BRD) is a multifactorial disease involving complex host immune interactions shaped by pathogenic agents and environmental factors. Advancements in RNA sequencing and associated analytical methods are improving our understanding of host response related to BRD pathophysiology. Supervised machine learning (ML) approaches present one such method for analyzing new and previously published transcriptome data to identify novel disease-associated genes and mechanisms. Our objective was to apply ML models to lung and immunological tissue datasets acquired from previous clinical BRD experiments to identify genes that classify disease with high accuracy. Raw mRNA sequencing reads from 151 bovine datasets (n = 123 BRD, n = 28 control) were downloaded from NCBI-GEO. Quality filtered reads were assembled in a HISAT2/Stringtie2 pipeline. Raw gene counts for ML analysis were normalized, transformed, and analyzed with MLSeq, utilizing six ML models. Cross-validation parameters (fivefold, repeated 10 times) were applied to 70% of the compiled datasets for ML model training and parameter tuning; optimized ML models were tested with the remaining 30%. Downstream analysis of significant genes identified by the top ML models, based on classification accuracy for each etiological association, was performed within WebGestalt and Reactome (FDR ≤ 0.05). Nearest shrunken centroid and Poisson linear discriminant analysis with power transformation models identified 154 and 195 significant genes for IBR and BRSV, respectively; from these genes, the two ML models discriminated IBR and BRSV with 100% accuracy compared to sham controls. Significant genes classified by the top ML models in IBR (154) and BRSV (195), but not BVDV (74), were related to type I interferon production and IL-8 secretion, specifically in lymphoid tissue and not homogenized lung tissue. Genes identified in Mannheimia haemolytica infections (97) were involved in activating classical and alternative pathways of complement. Novel findings, including expression of genes related to reduced mitochondrial oxygenation and ATP synthesis in consolidated lung tissue, were discovered. Genes identified in each analysis represent distinct genomic events relevant to understanding and predicting clinical BRD. Our analysis demonstrates the utility of ML with published datasets for discovering functional information to support the prediction and understanding of clinical BRD.

Massive young stellar objects in the Local Group irregular galaxy NGC 6822 identified using machine learning

ABSTRACT We present a supervised machine learning methodology to classify stellar populations in the Local Group dwarf-irregular galaxy NGC 6822. Near-IR colours (J − H, H − K, and J − K), K-band magnitudes and far-IR surface brightness (at 70 and 160 $\mu$m) measured from Spitzer and Herschel images are the features used to train a Probabilistic Random Forest (PRF) classifier. Point-sources are classified into eight target classes: young stellar objects (YSOs), oxygen- and carbon-rich asymptotic giant branch stars, red giant branch and red supergiant stars, active galactic nuclei, massive main-sequence stars, and Galactic foreground stars. The PRF identifies sources with an accuracy of ∼ 90 per cent across all target classes rising to ∼96 per cent for YSOs. We confirm the nature of 125 out of 277 literature YSO candidates with sufficient feature information, and identify 199 new YSOs and candidates. Whilst these are mostly located in known star-forming regions, we have also identified new star formation sites. These YSOs have mass estimates between ∼15 and 50 M⊙, representing the most massive YSO population in NGC 6822. Another 82 out of 277 literature candidates are definitively classified as non-YSOs by the PRF analysis. We characterize the star formation environment by comparing the spatial distribution of YSOs to those of gas and dust using archival images. We also explore the potential of using (unsupervised) t-distributed stochastic neighbour embedding maps for the identification of the same stellar population classified by the PRF.

Automatic detection of mud‐wall signatures in ground‐penetrating radar data

AbstractThe ground‐penetrating radar (GPR) method with the standard constant‐offset reflection mode allows to detect and map different types of archaeological structures, such as walls, foundations, floors and roads. The interpretation of the GPR data usually involves a detailed and time‐consuming analysis of large amounts of information, which entails nonnegligible chances of errors, especially under nonideal fieldwork conditions. The application of suitable automatic detection algorithms can be useful to more rapidly and successfully complete the interpretation task. In this work, we explore the use of supervised machine learning methodologies to automatically detect mud‐wall signatures in radargrams and to map the structures from these detections. Several algorithms, based on Viola–Jones cascade classifiers and the image feature descriptors Haar, histogram of oriented gradients and local binary patterns, were implemented. These algorithms were applied to datasets previously acquired in pre‐Inca and Inca‐Hispanic sites located in the Andean NW region of Argentina. The best algorithms provided very good detection rates for well‐preserved walls and acceptable rates for deteriorated walls, with a low number of spurious predictions. They also exhibited the ability to detect collapsed walls and fragments detached from them. These are remarkable results because mud walls are usually difficult to be detected by conventional analysis, owing to the complex and variable characteristics of their reflection patterns. The results of the automatic detection techniques were represented in plan views and three‐dimensional (3D) views that delineated in detail most of the structures of the sites. These algorithms are very fast, so applying them significantly reduces the interpretation times. In addition, once they have been trained using the patterns of one or several sites, they are directly applicable to other sites with similar characteristics.

Biventricular imaging markers to predict outcomes in non-compaction cardiomyopathy: a machine learning study.

AimsLeft ventricular non‐compaction cardiomyopathy (LVNC) is a genetic heart disease, with heart failure, arrhythmias, and embolic events as main clinical manifestations. The goal of this study was to analyse a large set of echocardiographic (echo) and cardiac magnetic resonance imaging (CMRI) parameters using machine learning (ML) techniques to find imaging predictors of clinical outcomes in a long‐term follow‐up of LVNC patients.Methods and resultsPatients with echo and/or CMRI criteria of LVNC, followed from January 2011 to December 2017 in the heart failure section of a tertiary referral cardiologic hospital, were enrolled in a retrospective study. Two‐dimensional colour Doppler echocardiography and subsequent CMRI were carried out. Twenty‐four hour Holter monitoring was also performed in all patients. Death, cardiac transplantation, heart failure hospitalization, aborted sudden cardiac death, complex ventricular arrhythmias (sustained and non‐sustained ventricular tachycardia), and embolisms (i.e. stroke, pulmonary thromboembolism and/or peripheral arterial embolism) were registered and were referred to as major adverse cardiovascular events (MACEs) in this study. Recruited for the study were 108 LVNC patients, aged 38.3 ± 15.5 years, 48.1% men, diagnosed by echo and CMRI criteria. They were followed for 5.8 ± 3.9 years, and MACEs were registered. CMRI and echo parameters were analysed via a supervised ML methodology. Forty‐seven (43.5%) patients had at least one MACE. The best performance of imaging variables was achieved by combining four parameters: left ventricular (LV) ejection fraction (by CMRI), right ventricular (RV) end‐systolic volume (by CMRI), RV systolic dysfunction (by echo), and RV lower diameter (by CMRI) with accuracy, sensitivity, and specificity rates of 75.5%, 77%, 75%, respectively.ConclusionsOur findings show the importance of biventricular assessment to detect the severity of this cardiomyopathy and to plan for early clinical intervention. In addition, this study shows that even patients with normal LV function and negative late gadolinium enhancement had MACE. ML is a promising tool for analysing a large set of parameters to stratify and predict prognosis in LVNC patients.

Identifying disease genes using machine learning and gene functional similarities, assessed through Gene Ontology.

Identifying disease genes from a vast amount of genetic data is one of the most challenging tasks in the post-genomic era. Also, complex diseases present highly heterogeneous genotype, which difficult biological marker identification. Machine learning methods are widely used to identify these markers, but their performance is highly dependent upon the size and quality of available data. In this study, we demonstrated that machine learning classifiers trained on gene functional similarities, using Gene Ontology (GO), can improve the identification of genes involved in complex diseases. For this purpose, we developed a supervised machine learning methodology to predict complex disease genes. The proposed pipeline was assessed using Autism Spectrum Disorder (ASD) candidate genes. A quantitative measure of gene functional similarities was obtained by employing different semantic similarity measures. To infer the hidden functional similarities between ASD genes, various types of machine learning classifiers were built on quantitative semantic similarity matrices of ASD and non-ASD genes. The classifiers trained and tested on ASD and non-ASD gene functional similarities outperformed previously reported ASD classifiers. For example, a Random Forest (RF) classifier achieved an AUC of 0. 80 for predicting new ASD genes, which was higher than the reported classifier (0.73). Additionally, this classifier was able to predict 73 novel ASD candidate genes that were enriched for core ASD phenotypes, such as autism and obsessive-compulsive behavior. In addition, predicted genes were also enriched for ASD co-occurring conditions, including Attention Deficit Hyperactivity Disorder (ADHD). We also developed a KNIME workflow with the proposed methodology which allows users to configure and execute it without requiring machine learning and programming skills. Machine learning is an effective and reliable technique to decipher ASD mechanism by identifying novel disease genes, but this study further demonstrated that their performance can be improved by incorporating a quantitative measure of gene functional similarities. Source code and the workflow of the proposed methodology are available at https://github.com/Muh-Asif/ASD-genes-prediction.

Supervised Machine Learning Approach Discovers Protective Sequence for Avoiding Sexual Victimization in Criminal Suit Documents

Previous research has studied effective self-protective behaviors, such as a victim’s physical resistance leading to the avoidance of sexual victimization. However, there are few studies on effective self-protective behavioral sequences, such as an offender’s physical violence followed by the victim’s physical resistance. Our study aims to clarify these sequences through a supervised machine learning approach. The samples consisted of 88 official documents on sexual assaults involving women, committed by male offenders incarcerated in a Japanese local prison. These crimes were classified as completed or attempted cases based on judges’ evaluations. All phrases in each crime description were also partitioned and coded according to the Japanese Penal Code. The support vector machine identified the most likely sequences of behaviors to predict completed and attempted cases. Approximately 90% of cases were correctly predicted through the identification of behavior sequences. The sequence involving an offender’s violence followed by the victim’s physical resistance predicted attempted sexual assault. However, the sequence involving a victim’s general resistance followed by the offender’s violence predicted completed sexual assault. Victims’ and offender’s behaviors need to be interpreted from behavioral sequence perspectives rather than a single action perspective. The supervised machine learning methodologies may extract self-protective behavioral sequences in documents more effectively than other methodologies. The self-protective sequence is a fundamental part of resistance during sexual assault. Training focused on protective sequence contributes to the improvement of resistance training and rape avoidance rates.

Thermostability of Proteins Revisited Through Machine Learning Methodologies: From Nucleotide Sequence to Structure

Background: Predicting thermostability of a protein, given its sequence or structure is a challenging job. Predicting which mutations can render mesophilic proteins thermostable is further challenging. A guided approach for the same is still elusive. Prediction can be done on the three hierarchies of protein organization: nucleotide sequence of the gene, primary amino acid sequences and the three dimensional structure of proteins. However it is still unclear that which level leads to a better predictor of protein thermostability and what combination of parameters in each level is responsible for thermostability of proteins. Methods: Present paper addresses this question by testing numerous unsupervised and supervised machine learning methodologies with multitude of prediction parameters to discriminate thermostable and mesostable proteins and to elucidate parameters in each level that lead to protein thermostability. The present method used 33 features of nucleotide sequences, 19 amino acid composition features and 17 structural features. Numerous machine learning approaches: 11 weighting algorithms, 5 unsupervised, 3 supervised learners and 2 lazy modeling algorithms with different parameters were tested and applied on the three datasets. Results: Results showed that amino acid datasets outperforms the nucleotide and structural datasets in generating the best prediction model for protein thermostability. LibSVM with polynomial kernel when applied on amino acid dataset with 10 fold cross validation results in the best prediction model for classifying thermostable proteins with 90.97% accuracy of prediction. Results also show that CG codons are not deterministic of thermostability. Decrease in Gln content, increase in hydrophobic residues and increase in main chain to main chain hydrogen bonds followed by gamma turns and aromatic aromatic interactions are responsible for protein thermostability. Conclusion: Present study shows that studying different datasets can bring about the importance of features in each hierarchy of protein organization which is responsible for protein thermostability. The results support the long drawn conclusion that protein amino acid composition is highly correlated to protein thermostability. Thus further insight into all these features can lead to the development of a better predictor and classifier of thermostable proteins. The results support the long drawn conclusion that protein amino acid composition is correlated to protein thermostability. The study also shows that testing of multitude of criteria is important before reaching at a definitive conclusion. Keywords: Thermostability, codon, amino acid, tertiary structure, predictor, machine learning.

Comparison of expectation-maximization clustering and logistic regression on categorization of planets with known properties

Analysis of the exoplanet data is the top priority of astrophysicists today. With the increasing incoming information there is a need for an efficient and reliable algorithm. The data is taken from exoplanet data explorer which was cross checked and filtered with NASA’s known categorization. These were then sorted into 5 categories: Dwarfs, Terrestrial, Icy, Jovian and Giant planets. This paper compares expectation-maximization clustering algorithm as an unsupervised and logistic regression as a supervised machine learning methodologies. Comparatively, logistic regression outperformed EM, indicating it cannot be used to sort through the incoming data. Further analysis is necessary.

Computations of diffuse fraction of global irradiance: Part 2 – Neural Networks

Solar energy is the feedstock for various applications of renewable energy systems, thus, the necessity of calculating and using global tilted irradiance is acknowledged for the computations of the performance and monitoring of Photovoltaic (PV) Parks and other solar energy applications. Thus, the aim of our research is to develop a model for the correlation of diffuse fraction (kd) and the clearness index (kt), that can then be used for the evaluation of the diffuse irradiance given the global irradiance.In a companion paper, existing simple empirical models were reviewed and compared based on 10years of data from Cyprus and then, analytical approaches for the computation of diffuse fraction were employed, where solar altitude was introduced as an additional parameter in the calculations.In the present paper, the same dataset was used, and three additional parameters were introduced to the calculations: global irradiance on the horizontal plane, extraterrestrial irradiance on the horizontal plane and the time of the day. These parameters were chosen due to the strong dependence of the diffuse fraction/clearness index correlation to the season/day of the year and time of day. Due to the non-linear influence of these parameters to the kt–kd correlations and the additional interaction between them, the employment of analytical methods is not applicable. Thus, non-parametric regression analysis was adopted, using supervised machine learning methodologies such as Artificial Neural Networks, which are able to learn the key information patterns from multivariate input.Comparing the non-parametric regression to the analytical models developed in the companion paper, it is shown herewith that the accuracy of the models was slightly improved. The statistical indicators MBE, RMSE and R2 of the best fit model were −4.69%, 21.54% and 0.90 respectively.

The rise and fall of supervised machine learning techniques

Machine learning is of immense importance in bioinformatics and biomedical science more generally (Larranaga et al., 2006; Tarca et al., 2007). In particular, supervised machine learning has been used to great effect in numerous bioinformatics prediction methods. Through many years of editing and reviewing manuscripts, we noticed that some supervised machine learning techniques seem to be gaining in popularity while others seemed, at least to our eyes, to be looking ‘unfashionable’. We were motivated to create a league table of machine learning techniques to learn what is hot and what is not in the machine learning field. In this editorial, we only include those that we considered major league and leave analysis of the minor league methods as an exercise for the interested reader. To create our league table, we created a list of supervised machine learning techniques commonly used in bioinformatics and their common synonyms, plural forms and abbreviations. We then searched this list against the PubMed titles and abstracts to identify the number of papers published per year for each machine learning technique. To match as many papers as possible, searches were case insensitive and allowed for variation in hyphenation. To our surprise, the artificial neural network (ANN) is not only the dominant league leader in 2011 but has been in this position since at least the 1970s (see Fig. 1). However, in recent years the usage of support vector machines (SVMs) grew tremendously, and we predict that SVMs will challenge ANNs for the dominant position in the coming decade. Since 2007 the number of publications using ANNs has decreased by 21%, which we hypothesize may be directly attributed to researchers increasingly using SVMs in place of ANNs. SVMs caught up with and overtook Markov models in 2004 to gain second spot in our machine learning league. Fig. 1. The growth of supervised machine learning methods in PubMed. As for the question of ‘what is hot?’, one can see that Random forests are a rapidly growing method with not a single mention of them before 2003 and now a total of 407 papers published to date. We were hoping to find techniques that were not so hot and perhaps going out of fashion. The results show that none of the major league methods has gone out of fashion, but we do see moderate decreases in the use of both ANNs and Markov models in the literature. We were also curious to find out if certain machine learning techniques were used in combination with each other. To investigate this, we looked at what machine learning methods are co-mentioned in articles (See Fig. 2). For all pairs of methods from the Supervised Machine Learning Top-5, we counted the number of abstracts that mention both methods and normalized the counts with the number of co-occurrences that would be expected by chance (based on the frequencies with which the methods are mentioned over the years). The strongest correlation (185 times higher than random expectation) is seen between decision trees and random forests, which is to be expected as random forests are ensembles of decision trees. Apart from this, the next strongest correlation (88 times higher than random expectation) is found between the two newest methods on the list, namely SVMs and random forests. We hypothesize that this is due to many researchers using these algorithms through machine learning frameworks such as Weka (Frank et al., 2004), which allows many different algorithms to easily be applied to the same dataset. Fig. 2. Heatmap showing the co-occurrence of machine learning techniques within articles. Applications of supervised machine learning methodology continue to grow in the biomedical literature. Despite new methods growing in usage, for example support vector machines and random forests, we see little evidence that any widely adopted methods are falling out of use. Conflict of Interest: none declared.

Atypical cytostatic mechanism of N-1-sulfonylcytosine derivatives determined by in vitro screening and computational analysis

We have previously shown that N-1-sulfonylpyrimidine derivatives have strong antiproliferative activity on human tumor cell lines, whereby 1-(p-toluenesulfonyl)cytosine showed good selectivity with regard to normal cells and was easily synthesized on a large scale. In the present work we have used an interdisciplinary approach to elucidate the compounds' mechanistic class. An augmented number of cell lines (11) has allowed a computational search for compounds with similar activity profiles and/or mechanistic class by integrating our data with the comprehensive DTP-NCI database. We applied supervised machine learning methodology (Random Forest classifier), which offers information complementary to unsupervised algorithms commonly used for analysis of cytostatic activity profiles, such as self-organizing maps. The computational results taken together with cell cycle perturbation and apoptosis analysis of the cell lines point to an unusual mechanism of cytostatic action, possibly a combination of nucleic acid antimetabolite activity and a novel molecular mechanism.