LAUSR

When someone is thinking of contextual data focuses on data that gives context to a person, entity or event. Thus, contextual data is collected from various sources of information, may have several applications and be used by multiple stakeholders. Indicatively, academic institutions may use contextual data to determine or predict the academic potential of applicants (taking into account their socioeconomic and health background, family or educational history), whereas business organizations may use contextual datasets for market research and analysis, considering various factors, such as geography, season, a specific date or even, the weather conditions. In informatics, a contextual bandit algorithm can adapt to the user-click feedback as the former progresses and exploits pre-existing information about the user’s (and similar users) browsing patterns to select which content to display. In such a case, a contextual bandit algorithm collects additional data (warm start) to help predict which content to display during the bandit test, instead of ‘a cold start’ (starting with no prediction with what the user (s) will click). Indeed, no matter the context, contextual data yields highly accurate predictions, as it is based on several sources of information – adding more contextual data to a prediction, the more precise the latter becomes. Nowadays, human thinking and acting are associated with large-scale, ever-increasing amounts of multi-faceted data derived from diverse and distributed sources and tools. The advent of technology and education has empowered data generation further in almost passionately rapid rhythms. Notwithstanding, there are vast discrepancies or even limitations when population-specific thinking is applied, calling for cost-effectiveness and sustainability in diverse settings. The above remarks advocate the exploitation of evidence-based synergies between a system and its components as well as among multiple systems if present. This is a holistic view of evidence-based practices [1]. Indicatively, we may consider an individual as a citizen (or a patient) in synergy with the social or health system the individual is part of, accompanied by the plethora of data that relate to that individual. This very same logic holds true for drug discovery if individuals are to be considered as biochemical entities. A biochemical entity is unique, even when compared to its analoges, yet it may represent the center of a system or multiple ones if a series of interactions (e.g. pharmacokinetic or pharmacodynamic) or topologies (e.g. cell or tissue or organism) are considered. At its node or level or key-point of such a consideration, multi-faceted data exist that may be assembled, mined and analyzed (Figure 1). In the context of drug discovery, the biological context adds up. The latter may also be the net context of several others, such as the organism-, tissue-, diseaseand cell linecontext [2]. We may consider the context of drug discovery per se as the net context of the rational drug design-context, the small molecule-context, the biologics-context, the experimental-context and the computer-aided-context. Overall, such contexts are paired with large and highly variable datasets that may be informative, only if paired to a context. Contextualization, the process of identifying data that relate to an entity based on that entity’s contextual information, is the means forward. In light of such a recalcitrant complexity, one must caution against the continuing divide between datasets, sources of information and even stakeholders and empower data integration. Indeed, data integration is revolutionizing many industries, but not as much as in healthcare, where data collection, integration and interpretation from many different sources, including the human body, becomes of benefit. One may think of the digital health data integration, which enables the collection of core data (drug response/ toxicity), metadata (patients’ medical history) and contextual data in a clinical trial setting of a new chemical entity. In this scenario, one gets a better understanding of this new chemical entity faster than in traditional trials and with reduced costs. Moving from the age of classical chemical synthesis to combinatorial chemistry and then, computer-aided drug discovery, technological advances, and proficiency in chemistry, biology, atomic theory, mathematics and computing promised rational thinking and design. Today, we are still in need of data integration, data interpretation, opinion-mining and decision-making devoid of biases. In the era of big data, datasets need to be reliable and reproducible to facilitate accurate predictions and turn information growth to knowledge growth. Data needs to be curated and cross-referenced, while noise should be distinguished from the analytic signal. Currently, drug discovery is not devoid of biases.