Average ROC AUC Research Articles

Predictive analytical model for ectopic pregnancy diagnosis: Statistics vs. machine learning.

ObjectiveEctopic pregnancy (EP) is well known for its critical maternal outcome. Early detection could make the difference between life and death in pregnancy. Our aim was to make a prompt diagnosis before the rupture occur. Thus, the predictive analytical models using both conventional statistics and machine learning (ML) methods were studied.Materials and methodsA retrospective cohort study was conducted on 407 pregnancies with unknown location (PULs): 306 PULs for internal validation and 101 PULs for external validation, randomized with a nested cross-validation technique. Using a set of 22 study features based on clinical factors, serum marker and ultrasound findings from electronic medical records, analyzing with neural networks (NNs), decision tree (DT), support vector machines (SVMs), and a statistical logistic regression (LR). Diagnostic performances were compared with the area under the curve (ROC-AUC), including sensitivity and specificity for decisional use.ResultsComparing model performance (internal validation) to predict EP, LR ranked first, with a mean ROC-AUC ± SD of 0.879 ± 0.010. In testing data (external validation), NNs ranked first, followed closely by LR, SVMs, and DT with average ROC-AUC ± SD of 0.898 ± 0.027, 0.896 ± 0.034, 0.882 ± 0.029, and 0.856 ± 0.033, respectively. For clinical aid, we report sensitivity of mean ± SD in LR: 90.20% ± 3.49%; SVM: 89.79% ± 3.66%; DT: 89.22% ± 4.53%; and NNs: 86.92% ± 3.24%, consecutively. However, specificity ± SD was ranked by NNs, followed by SVMs, LR, and DT, which were 82.02 ± 8.34%, 80.37 ± 5.15%, 79.65% ± 6.01%, and 78.97% ± 4.07%, respectively.ConclusionBoth statistics and the ML model could achieve satisfactory predictions for EP. In model learning, the highest ranked model was LR, showing that EP prediction might possess linear or causal data pattern. However, in new testing data, NNs could overcome statistics. This highlights the potency of ML in solving complicated problems with various patterns, while overcoming generalization error of data.

Classification of emotional states via transdermal cardiovascular spatiotemporal facial patterns using multispectral face videos

We describe a new method for remote emotional state assessment using multispectral face videos, and present our findings: unique transdermal, cardiovascular and spatiotemporal facial patterns associated with different emotional states. The method does not rely on stereotypical facial expressions but utilizes different wavelength sensitivities (visible spectrum, near-infrared, and long-wave infrared) to gauge correlates of autonomic nervous system activity spatially and temporally distributed across the human face (e.g., blood flow, hemoglobin concentration, and temperature). We conducted an experiment where 110 participants viewed 150 short emotion-eliciting videos and reported their emotional experience, while three cameras recorded facial videos with multiple wavelengths. Spatiotemporal multispectral features from the multispectral videos were used as inputs to a machine learning model that was able to classify participants’ emotional state (i.e., amusement, disgust, fear, sexual arousal, or no emotion) with satisfactory results (average ROC AUC score of 0.75), while providing feature importance analysis that allows the examination of facial occurrences per emotional state. We discuss findings concerning the different spatiotemporal patterns associated with different emotional states as well as the different advantages of the current method over existing approaches to emotion detection.

Individualized pathway activity algorithm identifies oncogenic pathways in pan-cancer analysis.

SummaryBackgroundAccumulative evidences have shown that dysregulation of biological pathways contributed to the initiation and progression of malignant tumours. Several methods for pathway activity measurement have been proposed, but they are restricted to making comparisons between groups or sensitive to experimental batch effects.MethodsWe introduced a novel method for individualized pathway activity measurement (IPAM) that is based on the ranking of gene expression levels in individual sample. Taking advantage of IPAM, we calculated the pathway activity of 318 pathways from KEGG database in the 10528 tumour/normal samples of 33 cancer types from TCGA to identify characteristic dysregulated pathways among different cancer types.FindingsIPAM precisely quantified the level of activity of each pathway in pan-cancer analysis and exhibited better performance in cancer classification and prognosis prediction over five widely used tools. The average ROC-AUC of cancer diagnostic model using tumour-educated platelets (TEPs) reached 92.84%, suggesting the potential of our algorithm in early diagnosis of cancer. We identified several pathways significantly deregulated and associated with patient survival in a large fraction of cancer types, such as tyrosine metabolism, fatty acid degradation, cell cycle, p53 signalling pathway and DNA replication. We also confirmed the dominant role of metabolic pathways in cancer pathway dysregulation and identified the driving factors of specific pathway dysregulation, such as PPARA for branched-chain amino acid metabolism and NR1I2, NR1I3 for fatty acid metabolism.InterpretationOur study will provide novel clues for understanding the pathological mechanisms of cancer, ultimately paving the way for personalized medicine of cancer.FundingA full list of funding can be found in the Acknowledgements section.

Improving Compound Activity Classification via Deep Transfer and Representation Learning.

Recent advances in molecular machine learning, especially deep neural networks such as graph neural networks (GNNs), for predicting structure–activity relationships (SAR) have shown tremendous potential in computer-aided drug discovery. However, the applicability of such deep neural networks is limited by the requirement of large amounts of training data. In order to cope with limited training data for a target task, transfer learning for SAR modeling has been recently adopted to leverage information from data of related tasks. In this work, in contrast to the popular parameter-based transfer learning such as pretraining, we develop novel deep transfer learning methods TAc and TAc-fc to leverage source domain data and transfer useful information to the target domain. TAc learns to generate effective molecular features that can generalize well from one domain to another and increase the classification performance in the target domain. Additionally, TAc-fc extends TAc by incorporating novel components to selectively learn feature-wise and compound-wise transferability. We used the bioassay screening data from PubChem and identified 120 pairs of bioassays such that the active compounds in each pair are more similar to each other compared to their inactive compounds. Overall, TAc achieves the best performance with an average ROC-AUC of 0.801; it significantly improves the ROC-AUC of 83% of target tasks with an average task-wise performance improvement of 7.102%, compared to the best baseline dmpna. Our experiments clearly demonstrate that TAc achieves significant improvement over all baselines across a large number of target tasks. Furthermore, although TAc-fc achieves slightly worse ROC-AUC on average compared to TAc (0.798 vs 0.801), TAc-fc still achieves the best performance on more tasks in terms of PR-AUC and F1 compared to other methods. In summary, TAc-fc is also found to be a strong model with competitive or even better performance than TAc on a notable number of target tasks.

Dynamic Forecast of Desert Locust Presence Using Machine Learning with a Multivariate Time Lag Sliding Window Technique

Desert locust plagues can easily cause a regional food crisis and thus affect social stability. Preventive control of the disaster highlights the early detection of hopper gregarization before they form devastating swarms. However, the response of hopper band emergence to environmental fluctuation exhibits a time lag. To realize the dynamic forecast of band occurrence with optimal temporal predictors, we proposed an SVM-based model with a temporal sliding window technique by coupling multisource time-series imagery with historical locust ground survey observations from between 2000–2020. The sliding window method was based on a lagging variable importance ranking used to analyze the temporal organization of environmental indicators in band-forming sequences and eventually facilitate the early prediction of band emergence. Statistical results show that hopper bands are more likely to occur within 41–64 days after increased rainfall; soil moisture dynamics increasing by approximately 0.05 m³/m³ then decreasing may enhance the chance of observing bands after 73–80 days. While sparse vegetation areas with NDVI increasing from 0.18 to 0.25 tend to witness bands after 17–40 days. The forecast model combining the optimal time lags of these dynamic indicators with other static indicators allows for a 16-day extended outlook of band presence in Somalia, Ethiopia, and Kenya. Monthly predictions from February to December 2020 display an overall accuracy of 77.46%, with an average ROC-AUC of 0.767 and a mean F-score close to 0.772. The multivariate forecast framework based on the lagging effect can realize the early warning of band presence in different spatiotemporal scenarios, supporting early decisions and response strategies for desert locust preventive management.

РАЗРАБОТКА АЛГОРИТМА КЛАССИФИКАЦИИ НА МОДЕЛИ АКАДЕМИЧЕСКИХ ДАННЫХ С ПОМОЩЬЮ ГИПЕРПАРАМЕТРИЧЕСКОЙ НАСТРОЙКИ ДЛЯ НАЙМА ПРЕПОДАВАТЕЛЕЙ В ИРАКСКИХ УНИВЕРСИТЕТАХ

In this paper, the researcher proposes a classification method for any employment opportunity in Iraqi universities using complete placement data by category for a dataset of employment in Iraqi educational institutions. The analysis of the scientific literature shows that this problem has not been sufficiently investigated and requires its solution. Developing a classification method based on genetic algorithms to get the best results to solve the problem of hiring new professors in Iraqi universities, getting accurate results and comparing several methods to solve this problem. The researcher tries to study learning classification algorithms with a teacher, such as logistic regression, Support Vector Classifier (SVC), Closest Neighbors KNN, Gaussian Naive Bayes algorithm, decision tree, random forest, gradient boosting and linear factor analysis (LDA). Hyperparameter optimization is also used for controlled algorithms to achieve better results. According to the proposed algorithms, the reliability and accuracy of the results obtained was ensured by a five-fold check. Results. Experimental results have shown that by adjusting hyperparameters in Linear Discriminant Analysis (LDA), it is possible to increase the accuracy of the results, as well as get a better result compared to other algorithms. The analysis criteria such as average accuracy, average score, G-Mean and Average ROC AUC were taken into account. It is according to the first three criteria that linear discriminant analysis demonstrates the highest accuracy of evaluation. It is established that among all the classification algorithms used, the classification of linear discriminant analysis is comparatively more accurate in all other scenarios of classification algorithms. It was found that adding the lsqr parameter solver and 0.81 shrinkage to linear discriminant analysis can improve performance. The article proves that the proposed classification algorithm based on the academic data model is able to solve the problem of hiring teachers at the university.

Convolutional Neural Networks for Fully Automated Diagnosis of Cardiac Amyloidosis by Cardiac Magnetic Resonance Imaging

Aims: We tested the hypothesis that artificial intelligence (AI)-powered algorithms applied to cardiac magnetic resonance (CMR) images could be able to detect the potential patterns of cardiac amyloidosis (CA). Readers in CMR centers with a low volume of referrals for the detection of myocardial storage diseases or a low volume of CMRs, in general, may overlook CA. In light of the growing prevalence of the disease and emerging therapeutic options, there is an urgent need to avoid misdiagnoses. Methods and Results: Using CMR data from 502 patients (CA: n = 82), we trained convolutional neural networks (CNNs) to automatically diagnose patients with CA. We compared the diagnostic accuracy of different state-of-the-art deep learning techniques on common CMR imaging protocols in detecting imaging patterns associated with CA. As a result of a 10-fold cross-validated evaluation, the best-performing fine-tuned CNN achieved an average ROC AUC score of 0.96, resulting in a diagnostic accuracy of 94% sensitivity and 90% specificity. Conclusions: Applying AI to CMR to diagnose CA may set a remarkable milestone in an attempt to establish a fully computational diagnostic path for the diagnosis of CA, in order to support the complex diagnostic work-up requiring a profound knowledge of experts from different disciplines.

Chemical toxicity prediction based on semi-supervised learning and graph convolutional neural network

As safety is one of the most important properties of drugs, chemical toxicology prediction has received increasing attentions in the drug discovery research. Traditionally, researchers rely on in vitro and in vivo experiments to test the toxicity of chemical compounds. However, not only are these experiments time consuming and costly, but experiments that involve animal testing are increasingly subject to ethical concerns. While traditional machine learning (ML) methods have been used in the field with some success, the limited availability of annotated toxicity data is the major hurdle for further improving model performance. Inspired by the success of semi-supervised learning (SSL) algorithms, we propose a Graph Convolution Neural Network (GCN) to predict chemical toxicity and trained the network by the Mean Teacher (MT) SSL algorithm. Using the Tox21 data, our optimal SSL-GCN models for predicting the twelve toxicological endpoints achieve an average ROC-AUC score of 0.757 in the test set, which is a 6% improvement over GCN models trained by supervised learning and conventional ML methods. Our SSL-GCN models also exhibit superior performance when compared to models constructed using the built-in DeepChem ML methods. This study demonstrates that SSL can increase the prediction power of models by learning from unannotated data. The optimal unannotated to annotated data ratio ranges between 1:1 and 4:1. This study demonstrates the success of SSL in chemical toxicity prediction; the same technique is expected to be beneficial to other chemical property prediction tasks by utilizing existing large chemical databases. Our optimal model SSL-GCN is hosted on an online server accessible through: https://app.cbbio.online/ssl-gcn/home.

Interpretable and Lightweight 3-D Deep Learning Model for Automated ACL Diagnosis.

We propose an interpretable and lightweight 3D deep neural network model that diagnoses anterior cruciate ligament(ACL) tears from a knee MRI exam. Previous works focused primarily on achieving better diagnostic accuracy but paid less attention to practical aspects such as explainability and model size. They mainly relied on ImageNet pre-trained 2D deep neural network backbones, such as AlexNet or ResNet, which are computationally expensive. Some of them tried to interpret the models using post-inference visualization tools, such as CAM or Grad-CAM, which lack in generating accurate heatmaps. Our work addresses the two limitations by understanding the characteristics of ACL tear diagnosis. We argue that the semantic features required for classifying ACL tears are locally confined and highly homogeneous. We harness the unique characteristics of the task by incorporating: 1) attention modules and Gaussian positional encoding to reinforce the seeking of local features; 2) squeeze modules and fewer convolutional filters to reflect the homogeneity of the features. As a result, our model is interpretable: our attention modules can precisely highlight the ACL region without any location information given to them. Our model is extremely lightweight: consisting of only 43K trainable parameters and 7.1G of Floating-point operations per second (FLOPs), that is 225 times smaller and 91 times lesser than the previous state-of-the-art, respectively. Our model is accurate: our model outperforms the previous state-of-the-art with the average ROC-AUC of 0.983 and 0.980 on the Chiba and Stanford knee datasets, respectively.

AI-driven deep CNN approach for multi-label pathology classification using chest X-Rays.

Artificial intelligence (AI) has played a significant role in image analysis and feature extraction, applied to detect and diagnose a wide range of chest-related diseases. Although several researchers have used current state-of-the-art approaches and have produced impressive chest-related clinical outcomes, specific techniques may not contribute many advantages if one type of disease is detected without the rest being identified. Those who tried to identify multiple chest-related diseases were ineffective due to insufficient data and the available data not being balanced. This research provides a significant contribution to the healthcare industry and the research community by proposing a synthetic data augmentation in three deep Convolutional Neural Networks (CNNs) architectures for the detection of 14 chest-related diseases. The employed models are DenseNet121, InceptionResNetV2, and ResNet152V2; after training and validation, an average ROC-AUC score of 0.80 was obtained competitive as compared to the previous models that were trained for multi-class classification to detect anomalies in x-ray images. This research illustrates how the proposed model practices state-of-the-art deep neural networks to classify 14 chest-related diseases with better accuracy.

A natural language processing and deep learning approach to identify child abuse from pediatric electronic medical records.

Child physical abuse is a leading cause of traumatic injury and death in children. In 2017, child abuse was responsible for 1688 fatalities in the United States, of 3.5 million children referred to Child Protection Services and 674,000 substantiated victims. While large referral hospitals maintain teams trained in Child Abuse Pediatrics, smaller community hospitals often do not have such dedicated resources to evaluate patients for potential abuse. Moreover, identification of abuse has a low margin of error, as false positive identifications lead to unwarranted separations, while false negatives allow dangerous situations to continue. This context makes the consistent detection of and response to abuse difficult, particularly given subtle signs in young, non-verbal patients. Here, we describe the development of artificial intelligence algorithms that use unstructured free-text in the electronic medical record—including notes from physicians, nurses, and social workers—to identify children who are suspected victims of physical abuse. Importantly, only the notes from time of first encounter (e.g.: birth, routine visit, sickness) to the last record before child protection team involvement were used. This allowed us to develop an algorithm using only information available prior to referral to the specialized child protection team. The study was performed in a multi-center referral pediatric hospital on patients screened for abuse within five different locations between 2015 and 2019. Of 1123 patients, 867 records were available after data cleaning and processing, and 55% were abuse-positive as determined by a multi-disciplinary team of clinical professionals. These electronic medical records were encoded with three natural language processing (NLP) algorithms—Bag of Words (BOW), Word Embeddings (WE), and Rules-Based (RB)—and used to train multiple neural network architectures. The BOW and WE encodings utilize the full free-text, while RB selects crucial phrases as identified by physicians. The best architecture was selected by average classification accuracy for the best performing model from each train-test split of a cross-validation experiment. Natural language processing coupled with neural networks detected cases of likely child abuse using only information available to clinicians prior to child protection team referral with average accuracy of 0.90±0.02 and average area under the receiver operator characteristic curve (ROC-AUC) 0.93±0.02 for the best performing Bag of Words models. The best performing rules-based models achieved average accuracy of 0.77±0.04 and average ROC-AUC 0.81±0.05, while a Word Embeddings strategy was severely limited by lack of representative embeddings. Importantly, the best performing model had a false positive rate of 8%, as compared to rates of 20% or higher in previously reported studies. This artificial intelligence approach can help screen patients for whom an abuse concern exists and streamline the identification of patients who may benefit from referral to a child protection team. Furthermore, this approach could be applied to develop computer-aided-diagnosis platforms for the challenging and often intractable problem of reliably identifying pediatric patients suffering from physical abuse.

ASFP (Artificial Intelligence based Scoring Function Platform): a web server for the development of customized scoring functions

Virtual screening (VS) based on molecular docking has emerged as one of the mainstream technologies of drug discovery due to its low cost and high efficiency. However, the scoring functions (SFs) implemented in most docking programs are not always accurate enough and how to improve their prediction accuracy is still a big challenge. Here, we propose an integrated platform called ASFP, a web server for the development of customized SFs for structure-based VS. There are three main modules in ASFP: (1) the descriptor generation module that can generate up to 3437 descriptors for the modelling of protein–ligand interactions; (2) the AI-based SF construction module that can establish target-specific SFs based on the pre-generated descriptors through three machine learning (ML) techniques; (3) the online prediction module that provides some well-constructed target-specific SFs for VS and an additional generic SF for binding affinity prediction. Our methodology has been validated on several benchmark datasets. The target-specific SFs can achieve an average ROC AUC of 0.973 towards 32 targets and the generic SF can achieve the Pearson correlation coefficient of 0.81 on the PDBbind version 2016 core set. To sum up, the ASFP server is a powerful tool for structure-based VS.

Efficient Identification of Patients Eligible for Clinical Studies Using Case-Based Reasoning on The Scottish Health Research Register (SHARE)

IntroductionTrials often struggle to achieve their target sample size with only half doing so. Some researchers have turned to Electronic Health Records (EHRs), seeking a more efficient way of recruitment. The Scottish Health Research Register (SHARE) obtained patients’ consent for their EHRs to be used as a searching base from which researchers can find potential participants. However, due to the fact that EHR data is not complete, sufficient or accurate, a database search strategy may not generate the best case-finding result. Objectives and ApproachA retrospective study was conducted to evaluate the performance of a case-based reasoning method in identifying participants for population-based clinical studies which had recruited through SHARE. A case-based reasoning framework was applied to nine studies with 119 total participants using two-fold cross-validation. Records of 30,000 random individuals were also merged with each test set to simulate the real-world recruitment setting. A prediction score for study participation was generated for each one in the test set through comparison of their diagnosis, procedure, pharmaceutical prescription, and laboratory test results attributes and those of the participants of a particular study. Evaluation was conducted by calculating Area Under the ROC Curve and information retrieval metrics for the ranking list of the test set by prediction score. We also compared the most likely participants as identified by searching a database to those ranked highest by our model. ResultsThe average ROCAUC for nine projects was 81% indicating strong predictive ability. However, the derived ranking lists showed lower predictive performance. 21% of the persons ranked within top 50 positions being the same as identified by searching databases. Conclusion / ImplicationsCase-based reasoning may be more effective than database search strategy for participant identification. This hypothesis requires a prospective study for further validation. The lower performance of ranking lists suggests improvements are needed in the collection and curation of EHRs.

Efficient identification of patients eligible for clinical studies using case-based reasoning on Scottish Health Research register (SHARE)

BackgroundTrials often struggle to achieve their target sample size with only half doing so. Some researchers have turned to Electronic Health Records (EHRs), seeking a more efficient way of recruitment. The Scottish Health Research Register (SHARE) obtained patients’ consent for their EHRs to be used as a searching base from which researchers can find potential participants. However, due to the fact that EHR data is not complete, sufficient or accurate, a database search strategy may not generate the best case-finding result. The current study aims to evaluate the performance of a case-based reasoning method in identifying participants for population-based clinical studies recruiting through SHARE, and assess the difference between its resultant cohort and the original one deriving from searching EHRs.MethodsA case-based reasoning framework was applied to 119 participants in nine projects using two-fold cross-validation, with records from a further 86,292 individuals used for testing. A prediction score for study participation was derived from the diagnosis, procedure, pharmaceutical prescription, and laboratory test results attributes of each participant. Evaluation was conducted by calculating Area Under the ROC Curve and information retrieval metrics for the ranking list of the test set by prediction score. We compared the most likely participants as identified by searching a database to those ranked highest by our model.ResultsThe average ROCAUC for nine projects was 81% indicating strong predictive ability for these data. However, the derived ranking lists showed lower predictive performance, with only 21% of the persons ranked within top 50 positions being the same as identified by searching databases.ConclusionsCase-based reasoning is may be more effective than a database search strategy for participant identification for clinical studies using population EHRs. The lower performance of ranking lists derived from case-based reasoning means that patients identified as highly suitable for study participation may still not be recruited. This suggests that further study is needed into improvements in the collection and curation of population EHRs, such as use of free text data to aid reliable identification of people more likely to be recruited to clinical trials.

Improving (Q)SAR predictions by examining bias in the selection of compounds for experimental testing

ABSTRACTExisting data on structures and biological activities are limited and distributed unevenly across distinct molecular targets and chemical compounds. The question arises if these data represent an unbiased sample of the general population of chemical-biological interactions. To answer this question, we analyzed ChEMBL data for 87,583 molecules tested against 919 protein targets using supervised and unsupervised approaches. Hierarchical clustering of the Murcko frameworks generated using Chemistry Development Toolkit showed that the available data form a big diffuse cloud without apparent structure. In contrast hereto, PASS-based classifiers allowed prediction whether the compound had been tested against the particular molecular target, despite whether it was active or not. Thus, one may conclude that the selection of chemical compounds for testing against specific targets is biased, probably due to the influence of prior knowledge. We assessed the possibility to improve (Q)SAR predictions using this fact: PASS prediction of the interaction with the particular target for compounds predicted as tested against the target has significantly higher accuracy than for those predicted as untested (average ROC AUC are about 0.87 and 0.75, respectively). Thus, considering the existing bias in the data of the training set may increase the performance of virtual screening.

Application of likelihood ratios and optimal decision thresholds in fire debris analysis based on a partial least squares discriminant analysis (PLS-DA) model

A set of 9000 computationally generated (in-silico) total ion spectra comprising samples containing ignitable liquids from ASTM E1618-14 classes mixed with pyrolyzed substrates (class IL) and samples containing a mixture of pyrolyzed substrates without ignitable liquid contribution (class SUB) was used to generate a partial least squares discriminant analysis (PLS-DA) model. The training data set was balanced in IL and SUB class composition and the IL class samples were composed of a uniform distribution of ignitable liquids from the ASTM E1618-14 defined classes. The number of components in the optimized PLS-DA model, 13, was determined by optimizing the area under the receiver operating characteristic curve (ROC AUC) as a function of the number of components. Ten cross validation folds had an average ROC AUC of 0.930 ± 0.006. The PLS-DA model was further challenged with a uniformly distributed and balanced IL/SUB validation set of 1000 computationally mixed total ion spectra, giving an ROC AUC 0f 0.93. The model which had been validated on computationally generated total ion spectra was tested on a set of “real” fire debris samples from large-scale burn tests that had been independently evaluated by an “informed analyst”. The results from this study demonstrate the application of a chemometric model to the forensic analysis of fire debris data and emphasize the connection between quantified strength of the evidence and categorical decisions based on a defined operational decision point on the ROC curve. The success of the PLS-DA model demonstrates the feasibility of applying chemometric methods to the analysis of fire debris as a means of introducing probabilistic statements of evidentiary value to the court.

Improving the Virtual Screening Ability of Target-Specific Scoring Functions Using Deep Learning Methods

Scoring functions play an important role in structure-based virtual screening. It has been widely accepted that target-specific scoring functions (TSSFs) may achieve better performance compared with universal scoring functions in actual drug research and development processes. A method that can effectively construct TSSFs will be of great value to drug design and discovery. In this work, we proposed a deep learning–based model named DeepScore to achieve this goal. DeepScore adopted the form of PMF scoring function to calculate protein–ligand binding affinity. However, different from PMF scoring function, in DeepScore, the score for each protein–ligand atom pair was calculated using a feedforward neural network. Our model significantly outperformed Glide Gscore on validation data set DUD-E. The average ROC-AUC on 102 targets was 0.98. We also combined Gscore and DeepScore together using a consensus method and put forward a consensus model named DeepScoreCS. The comparison results showed that DeepScore outperformed other machine learning–based TSSFs building methods. Furthermore, we presented a strategy to visualize the prediction of DeepScore. All of these results clearly demonstrated that DeepScore would be a useful model in constructing TSSFs and represented a novel way incorporating deep learning and drug design.

A Prediction Model Using Machine Learning Algorithm for Assessing Stone-Free Status after Single Session Shock Wave Lithotripsy to Treat Ureteral Stones

The aim of this study was to develop and validate a decision support model using a machine learning algorithm to predict treatment success after single session shock wave lithotripsy in ureteral stone cases. Of the 1,803 patients treated with shock wave lithotripsy we selected those with ureteral stones who had preoperative computerized tomography available. Treatment success after single session shock wave lithotripsy was defined as freedom from stones or residual stone fragments less than 2 mm long on computerized tomography or plain x-ray of the kidneys, ureters and bladder 2 weeks later. Decision tree analysis was done using a machine learning algorithm to identify relevant parameters. A decision support model was developed to calculate the probability of treatment success. A total of 791 patients were enrolled in study. Mean ± SD stone length was 5.9 ± 2.3 mm and mean stone volume was 89.3 ± 140.0 mm3. The overall treatment success rate after SWL was 64.4% (509 cases). The rate for upper, middle and lower ureter stones was 59.8%, 65.5% and 69.6%, respectively. On decision tree analysis the top 3 performance criteria factors were volume, length and HU. Decision models were constructed with all possible combinations of factors. The model with 15 factors had greater than 92% accuracy and an average ROC AUC of 0.951. We applied a machine learning algorithm, a subfield of artificial intelligence, to predict the outcome after single session shock wave lithotripsy for ureteral stones. A 92.29% accurate decision model was developed with 15 factors and an average ROC AUC of 0.951.

Three-level learning for improving cross-project logging prediction for if-blocks

Log statements present in the source code provide important execution information to the software developers while debugging. Predicting source code constructs that must be logged is a technically challenging task. Machine learning models are popularly used in literature for logging prediction. These models may not be suitable for small or new projects as these projects do not have sufficient prior data to train the prediction model. For such scenarios, cross-project logging prediction can be useful, in which knowledge gained from the standard project(s) is used to build the model. This paper proposes a three-level model, CLIF, for Cross-project if-blocks logging prediction. CLIF first converts textual features into numeric-textual features(Level-1). CLIF then combines numeric-textual features with Boolean and numeric features(Level-2). Since only few code constructs are logged, CLIF finally learns an effective imbalance decision threshold boundary to predict logged/non-logged code constructs(Level-3). We use 2 text mining classifiers(Level-1) and 8 machine learning classifiers(Level-2) and generate a total of 16 CLIF classifiers. We evaluate performance of CLIF classifiers on three open source projects(six source & target project pair). Experimental results show that CLIF outperforms the existing LogOpt-Plus model and give an improvement of up to 8.21% in average F1-score, and 4.89% in average ROC-AUC.

Computer Screen Use Detection Using Smart Eyeglasses

Screen use can influence the circadian phase and cause eye strain. Smart eyeglasses with an integrated colour light sensor can detect screen use. We present a screen use detection approach based on a light sensor embedded into the bridge of smart eyeglasses. By calculating the light intensity at the user’s eyes for different screens and content types we found only computer screens to have significant impact on the circadian phase. Our screen use detection is based on ratios between colour channels and used a linear support vector machine to detect screen use. We validated our detection approach in three studies. A Test bench was built to detect screen use under different ambient light sources and intensities in a controlled environment. In a Lab study, we evaluated recognition performance for different ambient light intensities. Using participant-independent models we achieved a ROC AUC above 0.9 for ambient light intensities below 200 lux. In a study of typical ADLs screen use was detected with an average ROC AUC of 0.83 assuming screen use for 30 % of the time.