Set Of Codes Research Articles

Syntactic–Semantic Detection of Clone-Caused Vulnerabilities in the IoT Devices

This paper addresses the problem of IoT security caused by code cloning when developing a massive variety of different smart devices. A clone detection method is proposed to identify clone-caused vulnerabilities in IoT software. A hybrid solution combines syntactic and semantic analyses of the code. Based on the recovered code, an attributed abstract syntax tree is constructed for each code fragment. All nodes of the commonly used abstract syntax tree are proposed to be weighted with semantic attribute vectors. Each attributed tree is then encoded as a semantic vector using a Deep Graph Neural Network. Two graph networks are combined into a Siamese neural model, allowing training to generate semantic vectors and compare vector pairs within each training epoch. Semantic analysis is also applied to clones with low similarity metric values. This allows one to correct the similarity decision in the case of incorrect matching of functions at the syntactic level. To automate the search for clones, the BinDiff algorithm is added in the first stage to accurately select clone candidates. This has a positive impact on the ability to apply the proposed method to large sets of binary code. In an experimental study, the developed method—compared to BinDiff, Gemini, and Asteria tools—has demonstrated the highest efficiency.

Empowering Graph Neural Networks with Block-Based Dual Adaptive Deep Adjustment for Drug Resistance-Related NcRNA Discovery.

Drug resistance to chemotherapeutic agents remains a formidable challenge in cancer treatment, significantly impacting treatment efficacy. Extensive research has exposed the intimate involvement of noncoding RNAs (ncRNAs) in conferring resistance to cancer drugs. Understanding the intricate associations between ncRNAs and drug resistance is of pivotal importance in advancing clinical interventions and expediting drug development. However, traditional biological experimental methods are hampered by limitations, such as labor intensiveness, time consumption, and constraints in scalability. Addressing these challenges necessitates the development of efficient computational methods for the accurate prediction of potential ncRNA-drug resistance associations (NDRA). However, most existing predictive models primarily focus on known ncRNA-drug resistance associations, often neglecting the critical aspect of similarity information between ncRNAs and drug resistance. This oversight may hinder the accuracy of characterizing these associations. To overcome the limitations of existing computational models, we proposed B-NDRA, a computational framework designed for the discovery of drug resistance-related ncRNA. Initially, we constructed a heterogeneous graph that integrates ncRNA-drug resistance pairs, leveraging both known associations and similarity fusion information between ncRNAs and drug resistance. Subsequently, we employed an attention mechanism to aggregate local features of graph nodes following a dimensionality reduction of node features. Further, a graph neural network (GNN) facilitated the learning of global node embeddings. Notably, the integration of dual adaptive deep adjustment architectures, encompassing intrablock and interblock methodologies, enabled efficient extraction of global features while balancing local and global features. Finally, B-NDRA employed a multilayer perceptron to predict associations between ncRNAs and drug resistance. Through rigorous 5-fold cross-validation, B-NDRA achieved average AUC, AUPR, Accuracy, Precision, Recall, and F1-score values of 92.2%, 91.9%, 84.88%, 86.9%, 82.37%, and 84.44%, respectively. Furthermore, comparative evaluations were conducted on established models, namely, GAEMDA, GRPAMDA, and LRGCPND. The results, obtained through three distinct 5-fold cross-validation strategies, demonstrated a notable performance improvement across almost all metrics for our B-NDRA. Specific case studies targeting Doxorubicin and Imatinib further validated the practicality of our B-NDRA in discovering potential NDRA. These results confirm the potential of our B-NDRA as a valuable tool in advancing cancer research and therapeutic development. The source code and data set of B-NDRA can be found at https://github.com/XuanLi1145/B-NDRA.

Mental Health–Related Outpatient Visits Among Adolescents and Young Adults, 2006-2019

Concerns over the mental health of young people have been increasing over the past decade, especially with the rise in mental health burden seen during the COVID-19 pandemic. Examining trends in mental health-related outpatient visits provides critical information to elucidate contributing factors, identify vulnerable populations, and inform strategies to address the mental health crisis. To examine characteristics and trends in mental health-related outpatient visits and psychotropic medication use among US adolescents and young adults. A retrospective cross-sectional analysis of nationally representative data from the National Ambulatory Medical Care Survey, an annual probability sample survey, was conducted from January 2006 to December 2019. Participants included adolescents (age 12-17 years) and young adults (age 18-24 years) with office-based outpatient visits in the US. Data were analyzed from March 1, 2023, to September 15, 2023. Mental health-related outpatient visits were identified based on established sets of diagnostic codes for psychiatric disorders. Temporal trends in the annual proportion of mental health-related outpatient visits were assessed, including visits associated with use of psychotropic medications. Analyses were stratified by age and sex. From 2006 to 2019, there were an estimated 1.1 billion outpatient visits by adolescents and young adults, of which 145.0 million (13.1%) were associated with a mental health condition (mean [SD] age, 18.4 [3.5] years; 74.0 million females [51.0%]). Mental health-related diagnoses were more prevalent among visits by male (16.8%) compared with female (10.9%) patients (P < .001). This difference was most pronounced among young adults, with 20.1% of visits associated with a psychiatric diagnosis among males vs 10.1% among females (P < .001). The proportion of mental health-related visits nearly doubled, from 8.9% in 2006 to 16.9% in 2019 (P < .001). Among all outpatient visits, 17.2% were associated with the prescription of at least 1 psychotropic medication, with significant increases from 12.8% to 22.4% by 2019 (P < .001). In this cross-sectional study, there were substantial increases in mental health-related outpatient visits and use of psychotropic medications, with greater overall burden among male patients. These findings provide a baseline for understanding post-pandemic shifts and suggest that current treatment and prevention strategies will need to address preexisting psychiatric needs in addition to the effects of the COVID-19 pandemic.

The performance of large language models in intercollegiate Membership of the Royal College of Surgeons examination.

Large language models (LLM), such as Chat Generative Pre-trained Transformer (ChatGPT) and Bard utilise deep learning algorithms that have been trained on a massive data set of text and code to generate human-like responses. Several studies have demonstrated satisfactory performance on postgraduate examinations, including the United States Medical Licensing Examination. We aimed to evaluate artificial intelligence performance in Part A of the intercollegiate Membership of the Royal College of Surgeons (MRCS) examination. The MRCS mock examination from Pastest, a commonly used question bank for examinees, was used to assess the performance of three LLMs: GPT-3.5, GPT 4.0 and Bard. Three hundred mock questions were input into the three LLMs, and the responses provided by the LLMs were recorded and analysed. The pass mark was set at 70%. The overall accuracies for GPT-3.5, GPT 4.0 and Bard were 67.33%, 71.67% and 65.67%, respectively (p = 0.27). The performances of GPT-3.5, GPT 4.0 and Bard in Applied Basic Sciences were 68.89%, 72.78% and 63.33% (p = 0.15), respectively. Furthermore, the three LLMs obtained correct answers in 65.00%, 70.00% and 69.17% of the Principles of Surgery in General questions (p = 0.67). There were no differences in performance in the overall and subcategories among the three LLMs. Our findings demonstrated satisfactory performance for all three LLMs in the MRCS Part A examination, with GPT 4.0 the only LLM that achieved the pass mark set.

On the computation of some code sets of the added Sierpinski triangle

In recent years, the intrinsic metrics have been formulated on the classical fractals. In particular, Sierpinski-like triangles such as equilateral, isosceles, scalene, added and mod$-3$ Sierpinski triangle have been considered in many different studies. The intrinsic metrics can be defined in different ways. One of the methods applied to obtain the intrinsic metric formulas is to use the code representations of the points on these self-similar sets. To define the intrinsic metrics via the code representations of the points on fractals makes also possible to investigate different geometrical, topological properties and geodesics of these sets. In this paper, we investigate some circles and closed sets of the added Sierpinski triangle and express them as the code sets by using its intrinsic metric.

No position-specific interference from prior lists in cued recognition: A challenge for position coding (and other) theories of serial memory

Position-specific intrusions of items from prior lists are rare but important phenomena that distinguish broad classes of theory in serial memory. They are uniquely predicted by position coding theories, which assume items on all lists are associated with the same set of codes representing their positions. Activating a position code activates items associated with it in current and prior lists in proportion to their distance from the activated position. Thus, prior list intrusions are most likely to come from the coded position. Alternative “item dependent” theories based on associations between items and contexts built from items have difficulty accounting for the position specificity of prior list intrusions. We tested the position coding account with a position-cued recognition task designed to produce prior list interference. Cuing a position should activate a position code, which should activate items in nearby positions in the current and prior lists. We presented lures from the prior list to test for position-specific activation in response time and error rate; lures from nearby positions should interfere more. We found no evidence for such interference in 10 experiments, falsifying the position coding prediction. We ran two serial recall experiments with the same materials and found position-specific prior list intrusions. These results challenge all theories of serial memory: Position coding theories can explain the prior list intrusions in serial recall and but not the absence of prior list interference in cued recognition. Item dependent theories can explain the absence of prior list interference in cued recognition but cannot explain the occurrence of prior list intrusions in serial recall.

FEA-Based Stress-Strain Barometers as Forecasters for Corneal Refractive Power Change in Orthokeratology.

To improve the effectivity of patient-specific finite element analysis (FEA) to predict refractive power change (RPC) in rigid Ortho-K contact lens fitting. Novel eyelid boundary detection is introduced to the FEA model to better model the effects of the lid on lens performance, and stress and strain outcomes are investigated to identify the most effective FEA components to use in modelling. The current study utilises fully anonymised records of 249 eyes, 132 right eyes, and 117 left eyes from subjects aged 14.1 ± 4.0 years on average (range 9 to 38 years), which were selected for secondary analysis processing. A set of custom-built MATLAB codes was built to automate the process from reading Medmont E300 height and distance files to processing and displaying FEA stress and strain outcomes. Measurements from before and after contact lens wear were handled to obtain the corneal surface change in shape and power. Tangential refractive power maps were constructed from which changes in refractive power pre- and post-Ortho-K wear were determined as the refractive power change (RPC). A total of 249 patient-specific FEA with innovative eyelid boundary detection and 3D construction analyses were automatically built and run for every anterior eye and lens combination while the lens was located in its clinically detected position. Maps of four stress components: contact pressure, Mises stress, pressure, and maximum principal stress were created in addition to maximum principal logarithmic strain maps. Stress and strain components were compared to the clinical RPC maps using the two-dimensional (2D) normalised cross-correlation and structural similarity (SSIM) index measure. On the one hand, the maximum principal logarithmic strain recorded the highest moderate 2D cross-correlation area of 8.6 ± 10.3%, and contact pressure recorded the lowest area of 6.6 ± 9%. Mises stress recorded the second highest moderate 2D cross-correlation area with 8.3 ± 10.4%. On the other hand, when the SSIM index was used to compare the areas that were most similar to the clinical RPC, maximum principal stress was the most similar, with an average strong similarity percentage area of 26.5 ± 3.3%, and contact pressure was the least strong similarity area of 10.3 ± 7.3%. Regarding the moderate similarity areas, all components were recorded at around 34.4% similarity area except the contact pressure, which was down to 32.7 ± 5.8%. FEA is an increasingly effective tool in being able to predict the refractive outcome of Ortho-K treatment. Its accuracy depends on identifying which clinical and modelling metrics contribute to the most accurate prediction of RPC with minimal ocular complications. In terms of clinical metrics, age, Intra-ocular pressure (IOP), central corneal thickness (CCT), surface topography, lens decentration and the 3D eyelid effect are all important for effective modelling. In terms of FEA components, maximum principal stress was found to be the best FEA barometer that can be used to predict the performance of Ortho-K lenses. In contrast, contact pressure provided the worst stress performance. In terms of strain, the maximum principal logarithmic strain was an effective strain barometer.

Verification and Processing of Academic Certificates Using Web Services

A web service is a set of code of a server application that performs a specific task or service whenever it is invoked through a Remote Procedure Call (RPC) by a client application that is connected to the server over the Internet. Web services are one of the digital strategies that can be used by the government for curtailing academic certificate scandals being perpetrated by fraudulent individuals in various sectors of the economy. The traditional method of checking academic records and certificates manually is inefficient and prone to forgery by unscrupulous elements. Similarly, the e-transcript system being adopted by some organizations to authenticate the academic qualifications of individuals for possible admission or employment still involves a manual process. The objective of this work therefore is to show how web services for certificate verification of all tiers of the educational system in Nigeria can be maintained by one central body which is the Federal Ministry of Education in Nigeria to enable various end-users (such as educational institutions, commercial organizations, etc) use the web services to authenticate and process the academic records of individuals effortlessly. A simulated data of 20 SSCE results of students who completed their secondary school education at various institutions was used to illustrate the application of a web service. The web service program was written in Java. A client application which was also written in Java was used to query the web service through RPC to authenticate and process the certificate claims of a sample of five students for admission into a tertiary institution. Results showed that the web service excellently ascertained the eligible candidates for admission and those that were not. The web service technology is therefore highly recommended for use by the federal ministry of education in Nigeria to eradicate the academic certificate scandals amongst the citizenry.

A construction of multiple Z-complementary code sets with inter-set low correlation

This paper presents a construction of multiple Z-complementary code sets (ZCCSs) of prime power length with low inter-set cross-correlation values using extended generalized Boolean functions (EGBFs). By collecting the proposed multiple Z-complementary code sets (ZCCSs), a low correlation zone complementary sequence set (LCZ-CSS) with new parameters is also achieved as a byproduct of this construction. Furthermore, we propose a new multi-cell multi-carrier code-division multiple-access (MC-CDMA) system for the downlink scenario, which has the ability to reduce multiple access interference (MAI) completely and decrease inter-cell interference (ICI) to a very low level in quasi-synchronous (QS) scenarios. Similarly, the proposed code can also be used for uplink MC-CDMA system. We provide the bit error rate (BER) performance of the proposed system and compared with the traditional system model. Moreover, we demonstrate that some of the existing constructions are special cases of the proposed construction.

ACPScanner: Prediction of Anticancer Peptides by Integrated Machine Learning Methodologies.

Novel therapeutic alternatives for cancer treatment are increasingly attracting global research attention. Although chemotherapy remains a primary clinical solution, it often results in significant side effects for patients. In recent years, anticancer peptides (ACPs) have emerged as promising candidates for highly specific anticancer drugs, and a number of computational approaches have been developed to identify ACPs. However, existing methods do not recognize specific types of anticancer function. In this article, we propose ACPScanner, an integrated approach to predict ACPs and non-ACPs at first and then predict several specific activity types for potential ACPs. We incorporate sequential, physicochemical properties, secondary structural information, and deep representation learning embeddings which are generated from artificial intelligence methods to build feature space. Customized deep learning and statistical learning methods are combined to form an integral architecture for the comprehensive two-level prediction task. To the best of our knowledge, ACPScanner is the first approach for specific ACP activity prediction. The comparative evaluation illustrates that ACPScanner achieves competitive prediction performance in both prediction phases in independent testings. We establish a web server at http://acpscanner.denglab.org to provide convenient usage of ACPScanner and make the predictive framework, source code, and data sets publicly available.

Visual Analysis Method for Traffic Trajectory with Dynamic Topic Movement Patterns Based on the Improved Markov Decision Process

The visual analysis of trajectory topics is helpful for mining potential trajectory patterns, but the traditional visual analysis method ignores the evolution of the temporal coherence of the topic. In this paper, a novel visual analysis method for dynamic topic analysis of traffic trajectory is proposed, which is used to explore and analyze the traffic trajectory topic and evolution. Firstly, the spatial information is integrated into trajectory words, calculating the dynamic trajectory topic model based on dynamic analysis modeling and, consequently, correlating the evolution of the trajectory topic between adjacent time slices. Secondly, in the trajectory topic, a representative trajectory sequence is generated to overcome the problem of the trajectory topic model not considering the word order, based on the improved Markov Decision Process. Subsequently, a set of meaningful visual codes is designed to analyze the trajectory topic and its evolution through the parallel window visual model from a spatial-temporal perspective. Finally, a case evaluation shows that the proposed method is effective in analyzing potential trajectory movement patterns.

How to Find the Equivalence Classes in a Set of Linear Codes in Practice?

An algorithm for equivalence of linear codes over finite fields is presented. Its main advantage is that it can extract exactly one representative from each equivalence class among a large number of linear codes. It can also be used as a test for isomorphism of binary matrices. The algorithm is implemented in the program LCequivalence, which is designed to obtain the inequivalent codes in a set of linear codes over a finite field with q&lt;64 elements. This program is a module of the free software package QextNewEdition for constructing, classifying and studying linear codes.

Defining childhood disability: ICF-CY developmental code sets

Purpose This study describes the development of four age-based item code sets from WHO’s International classification of functioning disability and health, children and youth version (ICF-CY). Given the continuing goal of universal implementation of the ICF-CY, a reduced set of codes was identified from more than 1600 codes to facilitate the use of the classification for clinical, research and policy applications of the ICF-CY Methods The ICF-CY developmental code sets were developed by consensus using the Delphi method. A multi-disciplinary group of international experts representing 27 countries from 5 WHO world regions completed a series of iterative online surveys to rate categories of child functioning essential for inclusion in the respective age-based code set. Results Four age-based code sets covering 37, 52, 60 and 57 codes across four domains of the ICF-CY were successfully derived with a high level of participant consensus. The code sets align with developmental theory and represent essential indicators of functioning defining key stages of child development. Conclusion The ICF-CY developmental code sets offer a common, universal language of childhood functioning and disability with global application for multidisciplinary research, clinical practice and policy.

Identification of heart failure hospitalization from NHS Digital data: comparison with expert adjudication.

Population-wide, person-level, linked electronic health record data are increasingly used to estimate epidemiology, guide resource allocation, and identify events in clinical trials. The accuracy of data from NHS Digital (now part of NHS England) for identifying hospitalization for heart failure (HHF), a key HF standard, is not clear. This study aimed to evaluate the accuracy of NHS Digital data for identifying HHF. Patients experiencing at least one HHF, as determined by NHS Digital data, and age- and sex-matched patients not experiencing HHF, were identified from a prospective cohort study and underwent expert adjudication. Three code sets commonly used to identify HHF were applied to the data and compared with expert adjudication (I50: International Classification of Diseases-10 codes beginning I50; OIS: Clinical Commissioning Groups Outcomes Indicator Set; and NICOR: National Institute for Cardiovascular Outcomes Research, used as the basis for the National Heart Failure Audit in England and Wales). Five hundred four patients underwent expert adjudication, of which 10 (2%) were adjudicated to have experienced HHF. Specificity was high across all three code sets in the first diagnosis position {I50: 96.2% [95% confidence interval (CI) 94.1-97.7%]; NICOR: 93.3% [CI 90.8-95.4%]; OIS: 95.6% [CI 93.3-97.2%]} but decreased substantially as the number of diagnosis positions expanded. Sensitivity [40.0% (CI 12.2-73.8%)] and positive predictive value (PPV) [highest with I50: 17.4% (CI 8.1-33.6%)] were low in the first diagnosis position for all coding sets. PPV was higher for the National Heart Failure Audit criteria, albeit modestly [36.4% (CI 16.6-62.2%)]. NHS Digital data were not able to accurately identify HHF and should not be used in isolation for this purpose.

Multi-sensor data fusion framework for energy optimization in smart homes

Advancements in Internet of Energy (IoE) technologies drive the development of several energy efficient frameworks for better energy optimization, economic savings, safety, and security in smart homes. However, certain challenges such as real-time operational data for each micro-moment, proper application of data fusion techniques, and end-to-end computing and deployment architecture prevent the establishment of an effective energy-efficient framework to provide personalized energy-saving recommendations. This work presents energy management for smart spaces (EMSS), the proposed energy efficiency framework implemented in an edge-cloud computing platform that fuses data from heterogeneous data sources (environmental sensors, camera, plug data, etc.) at appropriate data fusion levels and process them to generate actionable, explainable, personalized, and persuasive recommendations at the right moment. The user response to the generated recommendations triggers the actuators to perform respective energy-saving actions and provide more personalized future recommendations. Further, SMARTHome - a data generation framework based on configurable scenario files and a set of software codes was proposed to generate synthetic data with respect to different building types and micro-moments. The functionalities of the EMSS (device and user registration), user dashboard, analytics, and energy-saving recommendations were made accessible to the user through web and mobile applications. The validation analysis of the EMSS was performed by (i) comparative analysis of the machine learning and deep learning algorithms used by the decision engine to generate energy-saving recommendations and (ii) benchmarking of EMSS based on the taxonomy of data fusion-based energy efficiency frameworks for smart homes.

Decentralized Blockchain-Based and Trust-Aware Task Offloading Strategy for Healthcare IoT

Although smart healthcare has achieved rapid development and has gained considerable attention from academia and industry, pressing challenges exist such as resource limitation and stringent security demand. To tackle the issues, a blockchain-based and trust-aware task-offloading strategy (BBTAS) is proposed in this work, whereby task offloading requests are published as blockchain transactions that are automatically yet securely performed by smart contracts (SCs). A SC is a set of codes implementing pre-defined rules agreed by the participants involved. When a rule meets certain criteria or triggered by some events or transactions, a SC can perform by itself and generate verifiable results, which can be validated and securely stored in the blockchain. The proposed recommendation filtering method (RFM) and trust penalty measure (TPM)-based trust mechanism is powerful in resisting network insider attacks, and the inherent time-inefficiency problem of blockchain can be alleviated by integrating trust factor, which is theoretically analyzed in this work. To optimize the selection of service node, a utility-based decision-making method (UB-DMM) is proposed, which can maximize the system utility on the basis of fully considering multiple significant performance metrics of the system. Extensive simulations have been carried out to validate the effectiveness and superiority of the proposed measures.

A study of deep active learning methods to reduce labelling efforts in biomedical relation extraction

Automatic biomedical relation extraction (bioRE) is an essential task in biomedical research in order to generate high-quality labelled data that can be used for the development of innovative predictive methods. However, building such fully labelled, high quality bioRE data sets of adequate size for the training of state-of-the-art relation extraction models is hindered by an annotation bottleneck due to limitations on time and expertise of researchers and curators. We show here how Active Learning (AL) plays an important role in resolving this issue and positively improve bioRE tasks, effectively overcoming the labelling limits inherent to a data set. Six different AL strategies are benchmarked on seven bioRE data sets, using PubMedBERT as the base model, evaluating their area under the learning curve (AULC) as well as intermediate results measurements. The results demonstrate that uncertainty-based strategies, such as Least-Confident or Margin Sampling, are statistically performing better in terms of F1-score, accuracy and precision, than other types of AL strategies. However, in terms of recall, a diversity-based strategy, called Core-set, outperforms all strategies. AL strategies are shown to reduce the annotation need (in order to reach a performance at par with training on all data), from 6% to 38%, depending on the data set; with Margin Sampling and Least-Confident Sampling strategies moreover obtaining the best AULCs compared to the Random Sampling baseline. We show through the experiments the importance of using AL methods to reduce the amount of labelling needed to construct high-quality data sets leading to optimal performance of deep learning models. The code and data sets to reproduce all the results presented in the article are available at https://github.com/oligogenic/Deep_active_learning_bioRE.

Evaluating the effect of organic matter contained in shale on hydraulic fracturing of infill-well

Organic matter which is scattered uniformly in shale can respond to the applied stress and result in the variation of stress field. However, the effects of organic matter content in organic-rich shale on stress interference have not been well considered during developing infill-wells. A fully coupled numerical model is proposed in this paper to consider the whole flow spectrum of shale gas and investigate the effect of organic matter content on stress variation and fracture propagation in infill-well. Through simulating the production and fracturing process with only one set of code, some conclusions can be drawn that the alteration angle of the maximum horizontal principal stress increases and then decreases with the production time. Furthermore, the shrinkage of organic matter enlarges the alteration angle and the magnitude of the maximum horizontal principal stress. Certainly, the optimal fracturing effect in the infill-wells vary due to the different mass fraction of organic matter. This study not only helps to understand the effect of mass fraction of organic matter on stress variation and fracture propagation, but also provides theoretical support for increasing production from shale gas reservoirs.

Microservices performance forecast using dynamic Multiple Predictor Systems

Time series forecasting has been applied to predict performance degradation in Microservice-Based Applications (MBAs). The prediction enables MBA adaptation to avoid performance degradation and maintain the customer experience. The approaches in the literature commonly perform the forecast using a single monolithic model. However, since no model is better than others for all possible scenarios, using only one increases the risk of inaccurate estimates. Thus, an alternative for improving the accuracy and robustness of the performance degradation of MBAs is to adopt an ensemble, i.e., a Multiple Predictor Systems (MPS). This paper proposes an MPS methodology that selects and combines the most suitable models to forecast test patterns from a pool of models. Experiments were carried out with 32 time series containing performance metrics commonly used for MBA adaptation. Different scenarios concerning the pool (homogeneous or heterogeneous) and the selection phase (static or dynamic) are considered. Likewise, six widely used models in the literature were employed for pool generation: ARIMA, Multilayer Perceptron, Support Vector Regression, Random Forest, Long Short-Term Memory, and eXtreme Gradient Boosting. The evaluation shows that the proposed solution maintains or improves the forecast accuracy on 26 out of 32 datasets (81.25%) compared to monolithic models. In particular, the homogeneous pool and the dynamic selection of predictors obtained very satisfactory results. Therefore, MPS provide more accurate decision-making for MBA interventions, avoiding incorrect adjustments that could degrade the customer experience. Source code, figures, and data sets are publicly available at https://github.com/gfads/mps-methodology.

SIMPD: an algorithm for generating simulated time splits for validating machine learning approaches

Time-split cross-validation is broadly recognized as the gold standard for validating predictive models intended for use in medicinal chemistry projects. Unfortunately this type of data is not broadly available outside of large pharmaceutical research organizations. Here we introduce the SIMPD (simulated medicinal chemistry project data) algorithm to split public data sets into training and test sets that mimic the differences observed in real-world medicinal chemistry project data sets. SIMPD uses a multi-objective genetic algorithm with objectives derived from an extensive analysis of the differences between early and late compounds in more than 130 lead-optimization projects run within the Novartis Institutes for BioMedical Research. Applying SIMPD to the real-world data sets produced training/test splits which more accurately reflect the differences in properties and machine-learning performance observed for temporal splits than other standard approaches like random or neighbor splits. We applied the SIMPD algorithm to bioactivity data extracted from ChEMBL and created 99 public data sets which can be used for validating machine-learning models intended for use in the setting of a medicinal chemistry project. The SIMPD code and simulated data sets are available under open-source/open-data licenses at github.com/rinikerlab/molecular_time_series.