Validating instructional design and predicting student performance in histology education: Using machine learning via virtual microscopy.

As a part of modern technological environments, virtual microscopy enriches histological learning, with support from large institutional investments. However, existing literature does not supply empirical evidence of its role in improving pedagogy. Virtual microscopy provides fresh opportunities for investigating user behavior during the histology learning process, through digitized histological slides. This study establishes how students' perceptions and user behavior data can be processed and analyzed using machine learning algorithms. These also provide predictive data called learning analytics that enable predicting students' performance and behavior favorable for academic success. This information can be interpreted and used for validating instructional designs. Data on the perceptions, performances, and user behavior of 552 students enrolled in a histology course were collected from the virtual microscope, Cytomine®. These data were analyzed using an ensemble of machine learning algorithms, the extra-tree regression method, and predictive statistics. The predictive algorithms identified the most pertinent histological slides and descriptive tags, alongside 10 types of student behavior conducive to academic success. We used these data to validate our instructional design, and align the educational purpose, learning outcomes, and evaluation methods of digitized histological slides on Cytomine®. This model also predicts students' examination scores, with an error margin of <0.5 out of 20 points. The results empirically demonstrate the value of a digital learning environment for both students and teachers of histology.

Screening of repeated dose toxicity data in safety evaluation reports of cosmetic ingredients issued by the Scientific Committee on Consumer Safety between 2009 and 2019.

A focal point in the safety evaluation of cosmetic ingredients includes oral repeated dose toxicity testing, which is intended to address the most complex human endpoints. Seven years after the full implementation of the animal testing ban for cosmetic ingredients in the EU, there are still no alternative methods available capable of fully replacing oral repeated dose toxicity testing. Until this issue is resolved, the development of new cosmetic ingredients remains seriously hampered. The present paper describes a thorough screening of the oral repeated dose toxicity data included in safety evaluation reports of cosmetic ingredients addressed in the Annexes of the Cosmetics Regulation (EC) No 1223/2009, issued by the Scientific Committee on Consumer Safety between 2009 and 2019. The liver and the haematological system were identified as the potentially most frequently affected organs upon oral administration of cosmetic ingredients to animals. Evaluation of altered biochemical, morphological, and histopathological parameters related to hepatotoxicity indicated that the most recurrent events are liver weight changes, elevated liver enzymes, and alterations in serum cholesterol and bilirubin levels. Combined listing of affected parameters associated with steatosis and cholestasis indicated the possible occurrence of cholestasis, provoked by a limited number of cosmetic ingredients. The most frequently affected parameters related to the haematological system were indicative of anaemia. An in-depth analysis allowed characterisation of both regenerative and non-regenerative anaemia, pointing to direct and indirect haematotoxicity, respectively. The results presented in this study call for prioritisation of research targeted towards the development of new approach methodologies fit for animal-free repeated dose toxicity evaluation of cosmetic ingredients.

"Just-in-time" generation of datasets by considering structured representations of given consent for GDPR compliance.

Data processing is increasingly becoming the subject of various policies and regulations, such as the European General Data Protection Regulation (GDPR) that came into effect in May 2018. One important aspect of GDPR is informed consent, which captures one's permission for using one's personal information for specific data processing purposes. Organizations must demonstrate that they comply with these policies. The fines that come with non-compliance are of such importance that it has driven research in facilitating compliance verification. The state-of-the-art primarily focuses on, for instance, the analysis of prescriptive models and posthoc analysis on logs to check whether data processing is compliant to GDPR. We argue that GDPR compliance can be facilitated by ensuring datasets used in processing activities are compliant with consent from the very start. The problem addressed in this paper is how we can generate datasets that comply with given consent "just-in-time". We propose RDF and OWL ontologies to represent the consent that an organization has collected and its relationship with data processing purposes. We use this ontology to annotate schemas, allowing us to generate declarative mappings that transform (relational) data into RDF driven by the annotations. We furthermore demonstrate how we can create compliant datasets by altering the results of the mapping. The use of RDF and OWL allows us to implement the entire process in a declarative manner using SPARQL. We have integrated all components in a service that furthermore captures provenance information for each step, further contributing to the transparency that is needed towards facilitating compliance verification. We demonstrate the approach with a synthetic dataset simulating users (re-)giving, withdrawing, and rejecting their consent on data processing purposes of systems. In summary, it is argued that the approach facilitates transparency and compliance verification from the start, reducing the need for posthoc compliance analysis common in the state-of-the-art.

Data linkage in medical science using the resource description framework: the AVERT model.

There is an ongoing challenge as to how best manage and understand 'big data' in precision medicine settings. This paper describes the potential for a Linked Data approach, using a Resource Description Framework (RDF) model, to combine multiple datasets with temporal and spatial elements of varying dimensionality. This "AVERT model" provides a framework for converting multiple standalone files of various formats, from both clinical and environmental settings, into a single data source. This data source can thereafter be queried effectively, shared with outside parties, more easily understood by multiple stakeholders using standardized vocabularies, incorporating provenance metadata and supporting temporo-spatial reasoning. The approach has further advantages in terms of data sharing, security and subsequent analysis. We use a case study relating to anti-Glomerular Basement Membrane (GBM) disease, a rare autoimmune condition, to illustrate a technical proof of concept for the AVERT model.