(De)troubling transparency: artificial intelligence (AI) for clinical applications.

Artificial intelligence (AI) and machine learning (ML) techniques occupy a prominent role in medical research in terms of the innovation and development of new technologies. However, while many perceive AI as a technology of promise and hope-one that is allowing for more early and accurate diagnosis-the acceptance of AI and ML technologies in hospitals remains low. A major reason for this is the lack of transparency associated with these technologies, in particular epistemic transparency, which results in AI disturbing or troubling established knowledge practices in clinical contexts. In this article, we describe the development process of one AI application for a clinical setting. We show how epistemic transparency is negotiated and co-produced in close collaboration between AI developers and clinicians and biomedical scientists, forming the context in which AI is accepted as an epistemic operator. Drawing on qualitative research with collaborative researchers developing an AI technology for the early diagnosis of a rare respiratory disease (pulmonary hypertension/PH), this paper examines how including clinicians and clinical scientists in the collaborative practices of AI developers de-troubles transparency. Our research shows how de-troubling transparency occurs in three dimensions of AI development relating to PH: querying of data sets, building software and training the model The close collaboration results in an AI application that is at once social and technological: it integrates and inscribes into the technology the knowledge processes of the different participants in its development. We suggest that it is a misnomer to call these applications 'artificial' intelligence, and that they would be better developed and implemented if they were reframed as forms of sociotechnical intelligence.

Human-based approaches to pharmacology and cardiology: an interdisciplinary and intersectorial workshop.

Both biomedical research and clinical practice rely on complex datasets for the physiological and genetic characterization of human hearts in health and disease. Given the complexity and variety of approaches and recordings, there is now growing recognition of the need to embed computational methods in cardiovascular medicine and science for analysis, integration and prediction. This paper describes a Workshop on Computational Cardiovascular Science that created an international, interdisciplinary and inter-sectorial forum to define the next steps for a human-based approach to disease supported by computational methodologies. The main ideas highlighted were (i) a shift towards human-based methodologies, spurred by advances in new in silico, in vivo, in vitro, and ex vivo techniques and the increasing acknowledgement of the limitations of animal models. (ii) Computational approaches complement, expand, bridge, and integrate in vitro, in vivo, and ex vivo experimental and clinical data and methods, and as such they are an integral part of human-based methodologies in pharmacology and medicine. (iii) The effective implementation of multi- and interdisciplinary approaches, teams, and training combining and integrating computational methods with experimental and clinical approaches across academia, industry, and healthcare settings is a priority. (iv) The human-based cross-disciplinary approach requires experts in specific methodologies and domains, who also have the capacity to communicate and collaborate across disciplines and cross-sector environments. (v) This new translational domain for human-based cardiology and pharmacology requires new partnerships supported financially and institutionally across sectors. Institutional, organizational, and social barriers must be identified, understood and overcome in each specific setting.

Wellcome's 2019 funding for PhD programme training: An experiment in enhancing research culture.

In September 2019, the results of Wellcome's call for proposals for PhD training programmes were announced. This funding call was notable for making contributions to positive research culture as important a selection criterion as scientific excellence in PhD training. Aware that it's intervention on enhancing research culture was experimental, Wellcome also commissioned work to reflect on its processes. I undertook two studies of the early phases of the funding call, of the Wellcome's internal processes in shaping and implementing the call, and of the experiences and reflections of all applicants to the call. This Open Letter is a summary of the cross cutting themes of these two studies: the difficulties of defining positive research culture and of balancing it with scientific excellence; the expectation that supervision of PhD students is a key way to effect research culture change; the need for coherence between funding calls; issues around evaluating the track records of programmes on research culture, which further feed into defining the criteria of evaluation and selection relating to research culture.

For some time, it has been widely acknowledged that the culture of scientific research is negative, some would even say 'toxic'. There are many factors contributing to negative research culture, many of them systemic or structural, rather than having to do with individual scientists. These factors can make it more difficult to behave in ways that get in the way of best practices in research (such as sharing data and collaborating) and easier to behave in ways that are detrimental to research and to the well being of the research community, such as being aggressively competitive. There are a great many issues that fall under research culture, from addressing bullying and harassment, all the way to open science and scientific integrity. In September 2019, 23 Wellcome funded PhD programmes were launched. These programmes had been funded in a call that for the first time put positive research culture on a par with scientific excellence. This was a bold and experimental intervention in the research system. Aware of this fact, Wellcome also commissioned two studies into the shaping and implementation of the call, and into the way it had been received by applicants to the call. This Open Letter summarises the main themes that cut across the two studies: the difficulties of defining positive research culture and of balancing it with scientific excellence; the expectation that supervision of PhD students is a key way to effect research culture change; the need for coherence between funding calls; issues around evaluating the track records of programmes on research culture, which further feed into defining the criteria of evaluation and selection relating to research culture.

Why adverse outcome pathways need to be FAIR.

Adverse outcome pathways (AOPs) provide evidence for demonstrating and assessing causality between measurable toxicological mechanisms and human or environmental adverse effects. AOPs have gained increasing attention over the past decade and are believed to provide the necessary steppingstone for more effective risk assessment of chemicals and materials and moving beyond the need for animal testing. However, as with all types of data and knowledge today, AOPs need to be reusable by machines, i.e., machine-actionable, in order to reach their full impact potential. Machine-actionability is supported by the FAIR principles, which guide findability, accessibility, interoperability, and reusability of data and knowledge. Here, we describe why AOPs need to be FAIR and touch on aspects such as the improved visibility and the increased trust that FAIRification of AOPs provides.

New approach methodologies (NAMs) can detect biological phenomena that occur before they add up to serious problems like cancer, infertility, death, and others. NAMs detect key events (KE) along well-proven and agreed adverse outcome pathways (AOP). If a substance tests positive in a NAM for an upstream KE, this signals an early warning that actual adversity might follow. However, what if the knowledge about these AOPs is a well-kept secret? And what if decision-makers find AOPs too exotic to apply in risk assessment? This is where FAIR comes in! FAIR stands for making information findable, accessible, interoperable and re-useable. It aims to increase availability, usefulness, and trustworthiness of data. Here, we show that by interpreting the FAIR principles beyond a purely technical level, AOPs can ring in a new era of 3Rs applicability ‒ by increasing their visibility and making their creation process more transparent and reproducible.

Enhancing research culture through PhD training: a systems approach to identifying leverage points for policy formation.

This article examines the role of PhD training programmes in identifying and implementing positive interventions in research culture in the biosciences. Using a data set consisting of transcripts from interviews and group discussions with 179 participants from 18 of the current 23 (78%) UK-based Wellcome-funded PhD programmes, we apply a systems theory methodology to the system of higher education and PhD training. Using system mapping as an investigative tool, this approach identifies points of leverage within the system where policy interventions might be best targeted to affect changes to research culture in the global higher education sector. The results of this investigation highlight the student-supervisor relationship as a nexus for these interventions and recommends the programme structure as a global policy for PhD training.

CIAO: a living experiment in interdisciplinary large-scale collaboration facilitated by the Adverse Outcome Pathway framework.

Introduction: The CIAO project was launched in Spring 2020 to address the need to make sense of the numerous and disparate data available on COVID-19 pathogenesis. Based on a crowdsourcing model of large-scale collaboration, the project has exploited the Adverse Outcome Pathway (AOP) knowledge management framework built to support chemical risk assessment driven by mechanistic understanding of the biological perturbations at the different organizational levels. Hence the AOPs might have real potential to integrate data produced through different approaches and from different disciplines as experienced in the context of COVID-19. In this study, we aim to address the effectiveness of the AOP framework (i) in supporting an interdisciplinary collaboration for a viral disease and (ii) in working as the conceptual mediator of a crowdsourcing model of collaboration. Methods: We used a survey disseminated among the CIAO participants, a workshop open to all interested CIAO contributors, a series of interviews with some participants and a self-reflection on the processes. Results: The project has supported genuine interdisciplinarity with exchange of knowledge. The framework provided a common reference point for discussion and collaboration. The diagram used in the AOPs assisted with making explicit what are the different perspectives brought to the knowledge about the pathways. The AOP-Wiki showed up many aspects about its usability for those not already in the world of AOPs. Meanwhile their use in CIAO highlighted needed adaptations. Introduction of new Wiki elements for modulating factors was potentially the most disruptive one. Regarding how well AOPs support a crowdsourcing model of large-scale collaboration, the CIAO project showed that this is successful when there is a strong central organizational impetus and when clarity about the terms of the collaboration is brought as early as possible. Discussion: Extrapolate the successful CIAO approach and related processes to other areas of science where the AOP could foster interdisciplinary and systematic organization of the knowledge is an exciting perspective.

Bridging experiments, models and simulations: an integrative approach to validation in computational cardiac electrophysiology.

Computational models in physiology often integrate functional and structural information from a large range of spatiotemporal scales from the ionic to the whole organ level. Their sophistication raises both expectations and skepticism concerning how computational methods can improve our understanding of living organisms and also how they can reduce, replace, and refine animal experiments. A fundamental requirement to fulfill these expectations and achieve the full potential of computational physiology is a clear understanding of what models represent and how they can be validated. The present study aims at informing strategies for validation by elucidating the complex interrelations among experiments, models, and simulations in cardiac electrophysiology. We describe the processes, data, and knowledge involved in the construction of whole ventricular multiscale models of cardiac electrophysiology. Our analysis reveals that models, simulations, and experiments are intertwined, in an assemblage that is a system itself, namely the model-simulation-experiment (MSE) system. We argue that validation is part of the whole MSE system and is contingent upon 1) understanding and coping with sources of biovariability; 2) testing and developing robust techniques and tools as a prerequisite to conducting physiological investigations; 3) defining and adopting standards to facilitate the interoperability of experiments, models, and simulations; 4) and understanding physiological validation as an iterative process that contributes to defining the specific aspects of cardiac electrophysiology the MSE system targets, rather than being only an external test, and that this is driven by advances in experimental and computational methods and the combination of both.

Professional expectations and patient expectations concerning the development of Artificial Intelligence (AI) for the early diagnosis of Pulmonary Hypertension (PH).

The expectations of professionals working on the development of healthcare Artificial Intelligence (AI) technologies and the patients who will be affected by them have received limited attention. This paper reports on a Foresight Workshop with professionals involved with pulmonary hypertension (PH) and a Focus Group with members of a PH patient group, to discuss expectations of AI development and implementation. We show that while professionals and patients had similar expectations of AI, with respect to the priority of early diagnosis; data risks of privacy and reuse; and responsibility, other expectations differed. One important point of difference was in the attitude toward using AI to point up other potential health problems (in addition to PH). A second difference was in the expectations regarding how much clinical professionals should know about the role of AI in diagnosis. These findings allow us to better prepare for the future by providing a frank appraisal of the complexities of AI development with foresight, and the anxieties of key stakeholders.

Plastic diagnostics: The remaking of disease and evidence in personalized medicine.

Politically authorized reports on personalized and precision medicine stress an urgent need for finer-grained disease categories and faster taxonomic revision, through integration of genomic and phenotypic data. Developing a data-driven taxonomy is, however, not as simple as it sounds. It is often assumed that an integrated data infrastructure is relatively easy to implement in countries that already have highly centralized and digitalized health care systems. Our analysis of initiatives associated with the Danish National Genome Center, recently launched to bring Denmark to the forefront of personalized medicine, tells a different story. Through a "meta-taxonomy" of taxonomic revisions, we discuss what a genomics-based disease taxonomy entails, epistemically as well as organizationally. Whereas policy reports promote a vision of seamless data integration and standardization, we highlight how the envisioned strategy imposes significant changes on the organization of health care systems. Our analysis shows how persistent tensions in medicine between variation and standardization, and between change and continuity, remain obstacles for the production as well as the evaluation of genomics-based taxonomies of difference. We identify inherent conflicts between the ideal of dynamic revision and existing regulatory functions of disease categories in, for example, the organization and management of health care systems. Moreover, we raise concerns about shifts in the regulatory regime of evidence standards, where clinical care increasingly becomes a vehicle for biomedical research.

COVID-19 through Adverse Outcome Pathways: Building networks to better understand the disease - 3rd CIAO AOP Design Workshop.

On April 28-29, 2021, 50 scientists from different fields of expertise met for the 3rd online CIAO workshop. The CIAO project "Modelling the Pathogenesis of COVID-19 using the Adverse Outcome Pathway (AOP) framework" aims at building a holistic assembly of the available scientific knowledge on COVID-19 using the AOP framework. An individual AOP depicts the disease progression from the initial contact with the SARS-CoV-2 virus through biological key events (KE) toward an adverse outcome such as respiratory distress, anosmia or multiorgan failure. Assembling the individual AOPs into a network highlights shared KEs as central biological nodes involved in multiple outcomes observed in COVID-19 patients. During the workshop, the KEs and AOPs established so far by the CIAO members were presented and posi­tioned on a timeline of the disease course. Modulating factors influencing the progression and severity of the disease were also addressed as well as factors beyond purely biological phenomena. CIAO relies on an interdisciplinary crowd­sourcing effort, therefore, approaches to expand the CIAO network by widening the crowd and reaching stakeholders were also discussed. To conclude the workshop, it was decided that the AOPs/KEs will be further consolidated, inte­grating virus variants and long COVID when relevant, while an outreach campaign will be launched to broaden the CIAO scientific crowd.

Harvesting the promise of AOPs: An assessment and recommendations.

The Adverse Outcome Pathway (AOP) concept is a knowledge assembly and communication tool to facilitate the transparent translation of mechanistic information into outcomes meaningful to the regulatory assessment of chemicals. The AOP framework and associated knowledgebases (KBs) have received significant attention and use in the regulatory toxicology community. However, it is increasingly apparent that the potential stakeholder community for the AOP concept and AOP KBs is broader than scientists and regulators directly involved in chemical safety assessment. In this paper we identify and describe those stakeholders who currently-or in the future-could benefit from the application of the AOP framework and knowledge to specific problems. We also summarize the challenges faced in implementing pathway-based approaches such as the AOP framework in biological sciences, and provide a series of recommendations to meet critical needs to ensure further progression of the framework as a useful, sustainable and dependable tool supporting assessments of both human health and the environment. Although the AOP concept has the potential to significantly impact the organization and interpretation of biological information in a variety of disciplines/applications, this promise can only be fully realized through the active engagement of, and input from multiple stakeholders, requiring multi-pronged substantive long-term planning and strategies.

A social-technological epistemology of clinical decision-making as mediated by imaging.

In recent years there has been growing attention to the epistemology of clinical decision-making, but most studies have taken the individual physicians as the central object of analysis. In this paper we argue that knowing in current medical practice has an inherently social character and that imaging plays a mediating role in these practices. We have analyzed clinical decision-making within a medical expert team involved in diagnosis and treatment of patients with pulmonary hypertension (PH), a rare disease requiring multidisciplinary team involvement in diagnosis and management. Within our field study, we conducted observations, interviews, video tasks, and a panel discussion. Decision-making in the PH clinic involves combining evidence from heterogeneous sources into a cohesive framing of a patient, in which interpretations of the different sources can be made consistent with each other. Because pieces of evidence are generated by people with different expertise and interpretation and adjustments take place in interaction between different experts, we argue that this process is socially distributed. Multidisciplinary team meetings are an important place where information is shared, discussed, interpreted, and adjusted, allowing for a collective way of seeing and a shared language to be developed. We demonstrate this with an example of image processing in the PH service, an instance in which knowledge is distributed over multiple people who play a crucial role in generating an evaluation of right heart function. Finally, we argue that images fulfill a mediating role in distributed knowing in 3 ways: first, as enablers or tools in acquiring information; second, as communication facilitators; and third, as pervasively framing the epistemic domain. With this study of clinical decision-making in diagnosis and treatment of PH, we have shown that clinical decision-making is highly social and mediated by technologies. The epistemology of clinical decision-making needs to take social and technological mediation into account.

Variability in cardiac electrophysiology: Using experimentally-calibrated populations of models to move beyond the single virtual physiological human paradigm.

Physiological variability manifests itself via differences in physiological function between individuals of the same species, and has crucial implications in disease progression and treatment. Despite its importance, physiological variability has traditionally been ignored in experimental and computational investigations due to averaging over samples from multiple individuals. Recently, modelling frameworks have been devised for studying mechanisms underlying physiological variability in cardiac electrophysiology and pro-arrhythmic risk under a variety of conditions and for several animal species as well as human. One such methodology exploits populations of cardiac cell models constrained with experimental data, or experimentally-calibrated populations of models. In this review, we outline the considerations behind constructing an experimentally-calibrated population of models and review the studies that have employed this approach to investigate variability in cardiac electrophysiology in physiological and pathological conditions, as well as under drug action. We also describe the methodology and compare it with alternative approaches for studying variability in cardiac electrophysiology, including cell-specific modelling approaches, sensitivity-analysis based methods, and populations-of-models frameworks that do not consider the experimental calibration step. We conclude with an outlook for the future, predicting the potential of new methodologies for patient-specific modelling extending beyond the single virtual physiological human paradigm.

Validation and variability: dual challenges on the path from systems biology to systems medicine.

Systems biology is currently making a bid to show that it is able to make an important contribution to personalised or precision medicine. In order to do so, systems biologists need to find a way of tackling the pervasive variability of biological systems that is manifested in the medical domain as inter-subject variability. This need is simultaneously social and epistemic: social as systems biologists attempt to engage with the interests and concerns of clinicians and others in applied medical research; epistemic as they attempt to develop new strategies to cope with variability in the validation of the computational models typical of systems biology. This paper describes one attempt to develop such a strategy: a trial with a population-of-models approach in the context of cardiac electrophysiology. I discuss the development of this approach against the background of ongoing tensions between mathematically and experimentally inclined modellers on the one hand, and attempts to forge new collaborations with medical scientists on the other. Apart from the scientific interest of the population-of-models approach for tackling variability, the trial also offers a good illustration of the epistemology of experiment-facing modelling. I claim that it shows the extent to which experiment-facing modelling and validation require the establishment of criteria for comparing models and experiments that enable them to be linked together. These 'grounds of comparability' are the broad framework in which validation experiments are interpreted and evaluated by all the disciplines in the collaboration, or being persuaded to participate in it. I claim that following the process of construction of the grounds of comparability allows us to see the establishment of epistemic norms for judging validation results, through a process of 'normative intra-action' (Rouse, 2002) that shape the social and epistemic evolution of systems approaches to biomedicine.

Retaining volunteers in volunteer computing projects.

Volunteer computing projects (VCPs) have been set up by groups of scientists to recruit members of the public who are asked to donate spare capacity on their personal computers to the processing of scientific data or computationally intensive models. VCPs serve two purposes: to acquire significant computing capacity and to educate the public about science. A particular challenge for these scientists is the retention of volunteers as there is a very high drop-out rate. This paper develops recommendations for scientists and software engineers setting up or running VCPs regarding which strategies to pursue in order to improve volunteer retention rates. These recommendations are based on a qualitative study of volunteers in a VCP (climateprediction.net). A typology of volunteers has been developed, and three particularly important classes of volunteers are presented in this paper: for each type of volunteer, the particular benefits they offer to a project are described, and their motivations for continued participation in a VCP are identified and linked to particular strategies. In this way, those setting up a VCP can identify which types of volunteers they should be particularly keen to retain, and can then find recommendations to increase the retention rates of their target volunteers.