The Switch, the Ladder, and the Matrix: Models for Classifying AI Systems

Organisations that design and deploy artificial intelligence (AI) systems increasingly commit themselves to high-level, ethical principles. However, there still exists a gap between principles and practices in AI ethics. One major obstacle organisations face when attempting to operationalise AI Ethics is the lack of a well-defined material scope. Put differently, the question to which systems and processes AI ethics principles ought to apply remains unanswered. Of course, there exists no universally accepted definition of AI, and different systems pose different ethical challenges. Nevertheless, pragmatic problem-solving demands that things should be sorted so that their grouping will promote successful actions for some specific end. In this article, we review and compare previous attempts to classify AI systems for the purpose of implementing AI governance in practice. We find that attempts to classify AI systems proposed in previous literature use one of three mental models: the Switch, i.e., a binary approach according to which systems either are or are not considered AI systems depending on their characteristics; the Ladder, i.e., a risk-based approach that classifies systems according to the ethical risks they pose; and the Matrix, i.e., a multi-dimensional classification of systems that take various aspects into account, such as context, input data, and decision-model. Each of these models for classifying AI systems comes with its own set of strengths and weaknesses. By conceptualising different ways of classifying AI systems into simple mental models, we hope to provide organisations that design, deploy, or regulate AI systems with the vocabulary needed to demarcate the material scope of their AI governance frameworks.

The US Algorithmic Accountability Act of 2022 vs. The EU Artificial Intelligence Act: what can they learn from each other?

On the whole, the US Algorithmic Accountability Act of 2022 (US AAA) is a pragmatic approach to balancing the benefits and risks of automated decision systems. Yet there is still room for improvement. This commentary highlights how the US AAA can both inform and learn from the European Artificial Intelligence Act (EU AIA).

RA-MAP, molecular immunological landscapes in early rheumatoid arthritis and healthy vaccine recipients

Rheumatoid arthritis (RA) is a chronic inflammatory disorder with poorly defined aetiology characterised by synovial inflammation with variable disease severity and drug responsiveness. To investigate the peripheral blood immune cell landscape of early, drug naive RA, we performed comprehensive clinical and molecular profiling of 267 RA patients and 52 healthy vaccine recipients for up to 18 months to establish a high quality sample biobank including plasma, serum, peripheral blood cells, urine, genomic DNA, RNA from whole blood, lymphocyte and monocyte subsets. We have performed extensive multi-omic immune phenotyping, including genomic, metabolomic, proteomic, transcriptomic and autoantibody profiling. We anticipate that these detailed clinical and molecular data will serve as a fundamental resource offering insights into immune-mediated disease pathogenesis, progression and therapeutic response, ultimately contributing to the development and application of targeted therapies for RA.

Local Explanations via Necessity and Sufficiency: Unifying Theory and Practice

Necessity and sufficiency are the building blocks of all successful explanations. Yet despite their importance, these notions have been conceptually underdeveloped and inconsistently applied in explainable artificial intelligence (XAI), a fast-growing research area that is so far lacking in firm theoretical foundations. In this article, an expanded version of a paper originally presented at the 37th Conference on Uncertainty in Artificial Intelligence (Watson et al., 2021), we attempt to fill this gap. Building on work in logic, probability, and causality, we establish the central role of necessity and sufficiency in XAI, unifying seemingly disparate methods in a single formal framework. We propose a novel formulation of these concepts, and demonstrate its advantages over leading alternatives. We present a sound and complete algorithm for computing explanatory factors with respect to a given context and set of agentive preferences, allowing users to identify necessary and sufficient conditions for desired outcomes at minimal cost. Experiments on real and simulated data confirm our method’s competitive performance against state of the art XAI tools on a diverse array of tasks.

Interpretable machine learning for genomics

High-throughput technologies such as next-generation sequencing allow biologists to observe cell function with unprecedented resolution, but the resulting datasets are too large and complicated for humans to understand without the aid of advanced statistical methods. Machine learning (ML) algorithms, which are designed to automatically find patterns in data, are well suited to this task. Yet these models are often so complex as to be opaque, leaving researchers with few clues about underlying mechanisms. Interpretable machine learning (iML) is a burgeoning subdiscipline of computational statistics devoted to making the predictions of ML models more intelligible to end users. This article is a gentle and critical introduction to iML, with an emphasis on genomic applications. I define relevant concepts, motivate leading methodologies, and provide a simple typology of existing approaches. I survey recent examples of iML in genomics, demonstrating how such techniques are increasingly integrated into research workflows. I argue that iML solutions are required to realize the promise of precision medicine. However, several open challenges remain. I examine the limitations of current state-of-the-art tools and propose a number of directions for future research. While the horizon for iML in genomics is wide and bright, continued progress requires close collaboration across disciplines.

Testing conditional independence in supervised learning algorithms

We propose the conditional predictive impact (CPI), a consistent and unbiased estimator of the association between one or several features and a given outcome, conditional on a reduced feature set. Building on the knockoff framework of Candès et al. (J R Stat Soc Ser B 80:551–577, 2018), we develop a novel testing procedure that works in conjunction with any valid knockoff sampler, supervised learning algorithm, and loss function. The CPI can be efficiently computed for high-dimensional data without any sparsity constraints. We demonstrate convergence criteria for the CPI and develop statistical inference procedures for evaluating its magnitude, significance, and precision. These tests aid in feature and model selection, extending traditional frequentist and Bayesian techniques to general supervised learning tasks. The CPI may also be applied in causal discovery to identify underlying multivariate graph structures. We test our method using various algorithms, including linear regression, neural networks, random forests, and support vector machines. Empirical results show that the CPI compares favorably to alternative variable importance measures and other nonparametric tests of conditional independence on a diverse array of real and synthetic datasets. Simulations confirm that our inference procedures successfully control Type I error with competitive power in a range of settings. Our method has been implemented in an package, , which can be downloaded from https://github.com/dswatson/cpi.

Molecular Portraits of Early Rheumatoid Arthritis Identify Clinical and Treatment Response Phenotypes

There is a current imperative to unravel the hierarchy of molecular pathways that drive the transition of early to established disease in rheumatoid arthritis (RA). Herein, we report a comprehensive RNA sequencing analysis of the molecular pathways that drive early RA progression in the disease tissue (synovium), comparing matched peripheral blood RNA-seq in a large cohort of early treatment-naive patients, namely, the Pathobiology of Early Arthritis Cohort (PEAC). We developed a data exploration website (https://peac.hpc.qmul.ac.uk/) to dissect gene signatures across synovial and blood compartments, integrated with deep phenotypic profiling. We identified transcriptional subgroups in synovium linked to three distinct pathotypes: fibroblastic pauci-immune pathotype, macrophage-rich diffuse-myeloid pathotype, and a lympho-myeloid pathotype characterized by infiltration of lymphocytes and myeloid cells. This is suggestive of divergent pathogenic pathways or activation disease states. Pro-myeloid inflammatory synovial gene signatures correlated with clinical response to initial drug therapy, whereas plasma cell genes identified a poor prognosis subgroup with progressive structural damage.

•

Deep phenotyping and RNA-seq of early rheumatoid arthritis individuals pre-treatment

•

Synovial plasma cell gene expression predicts future progressive joint damage on X-ray

•

Blood interferon gene signature associates with synovial B and plasma cell infiltration

•

Interactive website enables RNA-seq and clinical data to be fully explored

Reaching the End-Game for GWAS: Machine Learning Approaches for the Prioritization of Complex Disease Loci

Genome-wide association studies (GWAS) have revealed thousands of genetic loci that underpin the complex biology of many human traits. However, the strength of GWAS - the ability to detect genetic association by linkage disequilibrium (LD) - is also its limitation. Whilst the ever-increasing study size and improved design have augmented the power of GWAS to detect effects, differentiation of causal variants or genes from other highly correlated genes associated by LD remains the real challenge. This has severely hindered the biological insights and clinical translation of GWAS findings. Although thousands of disease susceptibility loci have been reported, causal genes at these loci remain elusive. Machine learning (ML) techniques offer an opportunity to dissect the heterogeneity of variant and gene signals in the post-GWAS analysis phase. ML models for GWAS prioritization vary greatly in their complexity, ranging from relatively simple logistic regression approaches to more complex ensemble models such as random forests and gradient boosting, as well as deep learning models, i.e., neural networks. Paired with functional validation, these methods show important promise for clinical translation, providing a strong evidence-based approach to direct post-GWAS research. However, as ML approaches continue to evolve to meet the challenge of causal gene identification, a critical assessment of the underlying methodologies and their applicability to the GWAS prioritization problem is needed. This review investigates the landscape of ML applications in three parts: selected models, input features, and output model performance, with a focus on prioritizations of complex disease associated loci. Overall, we explore the contributions ML has made towards reaching the GWAS end-game with consequent wide-ranging translational impact.

Oncometabolite induced primary cilia loss in pheochromocytoma

Primary cilia are sensory organelles involved in regulation of cellular signaling. Cilia loss is frequently observed in tumors; yet, the responsible mechanisms and consequences for tumorigenesis remain unclear. We demonstrate that cilia structure and function is disrupted in human pheochromocytomas - endocrine tumors of the adrenal medulla. This is concomitant with transcriptional changes within cilia-mediated signaling pathways that are associated with tumorigenesis generally and pheochromocytomas specifically. Importantly, cilia loss was most dramatic in patients with germline mutations in the pseudohypoxia-linked genes SDHx and VHL. Using a pheochromocytoma cell line derived from rat, we show that hypoxia and oncometabolite-induced pseudohypoxia are key drivers of cilia loss and identify that this is dependent on activation of an Aurora-A/HDAC6 cilia resorption pathway. We also show cilia loss drives dramatic transcriptional changes associated with proliferation and tumorigenesis. Our data provide evidence for primary cilia dysfunction contributing to pathogenesis of pheochromocytoma by a hypoxic/pseudohypoxic mechanism and implicates oncometabolites as ciliary regulators. This is important as pheochromocytomas can cause mortality by mechanisms including catecholamine production and malignant transformation, while hypoxia is a general feature of solid tumors. Moreover, pseudohypoxia-induced cilia resorption can be pharmacologically inhibited, suggesting potential for therapeutic intervention.

Sex differences in the nitrate-nitrite-NO• pathway: Role of oral nitrate-reducing bacteria

Oral reduction of nitrate to nitrite is dependent on the oral microbiome and is the first step of an alternative mammalian pathway to produce nitric oxide in humans. Preliminary evidence suggests important sex differences in this pathway. We prospectively investigated sex-differences following inorganic nitrate supplementation on nitrate/nitrite levels and vascular function, and separately examined sex differences in oral nitrate reduction, and oral microbiota by 16S rRNA profiling. At baseline, females exhibit higher nitrite levels in all biological matrices despite similar nitrate levels to males. Following inorganic nitrate supplementation, plasma nitrite was increased to a significantly greater extent in females than in males and pulse wave velocity was only reduced in females. Females exhibited higher oral bacterial nitrate-reducing activity at baseline and after nitrate supplementation. Despite these differences, there were no differences in the composition of either the total salivary microbiota or those oral taxa with nitrate reductase genes. Our results demonstrate that females have augmented oral nitrate reduction that contributes to higher nitrite levels at baseline and also after inorganic nitrate supplementation, however this was not associated with differences in microbial composition (clinicaltrials.gov: NCT01583803).

A Framework for Multi-Omic Prediction of Treatment Response to Biologic Therapy for Psoriasis

Biologic therapies have shown high efficacy in psoriasis, but individual response varies and is poorly understood. To inform biomarker discovery in the Psoriasis Stratification to Optimise Relevant Therapy (i.e., PSORT) study, we evaluated a comprehensive array of omics platforms across three time points and multiple tissues in a pilot investigation of 10 patients with severe psoriasis, treated with the tumor necrosis factor (TNF) inhibitor, etanercept. We used RNA sequencing to analyze mRNA and small RNA transcriptome in blood, lesional and nonlesional skin, and the SOMAscan platform to investigate the serum proteome. Using an integrative systems biology approach, we identified signals of treatment response in genes and pathways associated with TNF signaling, psoriasis pathology, and the major histocompatibility complex region. We found association between clinical response and TNF-regulated genes in blood and skin. Using a combination of differential expression testing, upstream regulator analysis, clustering techniques, and predictive modeling, we show that baseline samples are indicative of patient response to biologic therapies, including signals in blood, which have traditionally been considered unreliable for inference in dermatology. In conclusion, our pilot study provides both an analytical framework and empirical basis to estimate power for larger studies, specifically the ongoing PSORT study, which we show as powered for biomarker discovery and patient stratification.

Research Techniques Made Simple: Bioinformatics for Genome-Scale Biology

High-throughput biology presents unique opportunities and challenges for dermatological research. Drawing on a small handful of exemplary studies, we review some of the major lessons of these new technologies. We caution against several common errors and introduce helpful statistical concepts that may be unfamiliar to researchers without experience in bioinformatics. We recommend specific software tools that can aid dermatologists at varying levels of computational literacy, including platforms with command line and graphical user interfaces. The future of dermatology lies in integrative research, in which clinicians, laboratory scientists, and data analysts come together to plan, execute, and publish their work in open forums that promote critical discussion and reproducibility. In this article, we offer guidelines that we hope will steer researchers toward best practices for this new and dynamic era of data intensive dermatology.

Pre-operative oral carbohydrate loading in colorectal surgery: a randomized controlled trial

OBJECTIVE

Surgery induces a catabolic response with stress hormone release and insulin resistance. The aim of this study was to assess the effect of pre-operative carbohydrate administration on grip strength, gastrointestinal function and hospital stay following elective colorectal surgery.

METHODS

Thirty-six patients undergoing elective colonic resection were randomized into one of three groups. Group 1 were fasted; Group 2 were given pre-operative oral water, Group 3 received equivalent volumes of a Maltodextrin drink. Time to first flatus, first bowel movement and hospital stay were recorded. Muscle strength was measured pre-operatively, and on alternate days thereafter until discharge using a grip strength dynamometer.

RESULTS

Patients in the carbohydrate group had a median postoperative hospital stay of 7.5 days compared with 13 days in the water group (P > 0.01) and 10 days in the fasted group (P = 0.06). The median time postsurgery to first flatus was 3 days for both the fasted and water groups compared with 1.5 days in the carbohydrate group (P = 0.13). First bowel movement occurred on day 3 in the carbohydrate group, day 4 in the fasting group and day 5 in the water group. The fasted group showed a significant reduction in postoperative grip strength (P < 0.05) with a median drop of 10% at discharge. Neither the water nor the carbohydrate groups showed significant reductions in muscle strength.

CONCLUSION

We found that pre-operative administration of oral carbohydrate leads to a significantly reduced postoperative hospital stay, and a trend towards earlier return of gut function when compared with fasting or supplementary water.

Cyclin D3 expression, cell proliferation and pathological stage of human primary colorectal cancer

Cyclin D3 promotes cell cycle progression but its expression and prognostic significance in human colorectal cancer is unknown. This study assayed cyclin D3 expression against cell cycle phase fraction and Duke's stage in 35 fresh human primary colorectal cancers. DNA content, cell cycle phase fraction and cyclin D3 expression were assessed by flow cytometry in disaggregated tumors. Cyclin D3 expression and S-phase fraction were independently related to Duke's stage. In Duke's stage C tumors, a higher proportion of cells expressed cyclin D3 (14.4 vs. 8.8%, mean; p < 0.05 by Mann-Whitney U test) and were in DNA synthesis (S) phase (21.1 vs. 9.7%, mean; p < 0.05 by Mann-Whitney U test). Neoplastic deregulation of cyclin D3 expression may provide a selective growth advantage which is related to stage in human colorectal cancer.

Growth dysregulation and p53 accumulation in human primary colorectal cancer

p53 accumulation is common in colorectal cancer, but effects on growth homeostasis are unclear. In this study, DNA content, cell cycle phase fractions and DNA strand-breaks consistent with apoptosis were assessed by flow cytometry in 42 fresh primary colorectal tumours and matched normal mucosa. p53 accumulation was assessed in 37 fixed tumour sections, by immunohistochemistry. In normal mucosa, 10.3 +/- 6.6% (mean +/- s.d.) cells were in DNA synthesis phase while 28.7 +/- 17.9% showed apoptosis. A relationship suggestive of growth homeostasis, was observed between these parameters (r = 0.8; P < 0.05). In cancers, a greater number of cells were in DNA synthesis phase (15.6 +/- 12.9% tumour vs mucosa 10.3 +/- 6.6%; P < 0.02) while fewer showed apoptosis than normal mucosa (18.5 +/- 17.0% tumour vs mucosa 28.7 +/- 17.9%; P < 0.01). DNA synthesis and apoptosis fractions were unrelated in cancers, suggesting growth dysequilibrium. p53 accumulation was detected in 59% (22/37) tumours and was associated with reduced apoptosis compared to p53-negative tumours or mucosa (14.8 +/- 15% p53 accumulation vs 26.3 +/- 18% p53-negative; P < 0.05; vs 28.7 +/- 17.9% mucosa; P < 0.05). p53 accumulation was unrelated to DNA synthesis phase fractions. p53 accumulation is accompanied by reduced apoptosis which may accentuate growth dysequilibrium in colorectal cancer.

Pain control after thoracotomy: bupivacaine versus lidocaine in continuous extrapleural intercostal nerve blockade

BACKGROUND

The use of a continuous bupivacaine extrapleural intercostal nerve block after posterolateral thoracotomy has been shown in randomized controlled studies to be effective in reducing postoperative pain and restoring pulmonary function. It is our hypothesis that when using a continuous infusion for nerve block, a long-acting agent (bupivacaine) is unnecessary and a shorter-acting agent (lidocaine) would offer equivalent results with less systemic toxicity. This study was designed to determine whether lidocaine was as effective as bupivacaine in a continuous extrapleural intercostal nerve block after posterolateral thoracotomy because lidocaine is a less toxic analgesic agent. The study was prospectively randomized and double-blinded.

METHODS

Forty-six patients undergoing elective posterolateral thoracotomy were randomized to blindly receive bupivacaine (n = 23) or lidocaine (n = 23) by continuous infusion pump through an intraoperatively placed indwelling extrapleural catheter. Postoperative pain was assessed for 48 hours by patient-controlled morphine consumption and by linear visual analog scale. There was no statistically significant difference in age, sex, or type of operation between the two groups.

RESULTS

There was no statistically significant difference between the bupivacaine and lidocaine groups in patient-controlled morphine use or in visual analog scale scores.

CONCLUSIONS

Lidocaine offers equivalent pain control to bupivacaine when administered for continuous extrapleural intercostal nerve block after posterolateral thoracotomy, with less risk of systemic toxicity.

Incidence of inguinal hernias diagnosed during laparoscopy

This prospective study was done to identify the incidence of asymptomatic hernias diagnosed during laparoscopic surgery. Previously unidentified inguinal hernias were discovered in 13 of 100 consecutive patients undergoing laparoscopic procedures (13%). Nine of the patients were male, four were female. The average age was 43.3 years. Four defects were direct and nine were indirect (patent processus vaginalis). Eleven defects were small (< 3 cm) but two were large (> 3 cm). All four direct defects occurred in male patients. We conclude that asymptomatic defects are surprisingly common in our surgical population (13%) and that laparoscopic examination of the pelvis is a sensitive technique for identifying these defects. To date, no patient with laparoscopically diagnosed hernia has had symptoms or complications related to the hernia. We recommend that the surgeon should note in the medical record the presence of a defect, but not proceed with prophylactic repair of asymptomatic defects.

Developing a competency-based education program for nurse-monitored sedation

Because of technological innovations and changes in the methods of health care delivery in the ambulatory setting, a new role for the perioperative nurse has emerged, that of monitoring patients who receive intravenous (IV)-conscious sedation during surgery and other invasive procedures. This article describes how to develop a competency-based education program for an ambulatory surgery center advocating the role of both a circulating nurse and monitoring nurse for the care of patients receiving IV-conscious sedation.

Ear health of aboriginal primary school children in the Eastern Goldfields Region of Western Australia

The ear health and state of hearing of 642 Aboriginal children living in the Eastern Goldfields of Western Australia in 1989 was assessed by audiometry, otoscopy and tympanometry in order to establish the efficacy of the existing ear health program. A heavy burden of ear disease and hearing loss is reported and the findings are compared with previous surveys. A conclusion is drawn that the ear health of this population has not improved in the last decade.

Petrol abuse at Elcho Island: an attempted intervention

An attempt to contain juvenile petrol abuse at Elcho Island in the Northern Territory of Australia is described. Ethyl mercaptan (skunk juice) was introduced as an additive to the petrol to discourage petrol sniffing. Reasons for the failure of the programme are discussed.

Respiratory function in aboriginal children

One hundred and fifty-one Aboriginal school children and 246 non-Aboriginal children, all aged between 11 and 15 years, performed lung function tests with a Vitalograph spirometer. The results showed that the mean forced vital capacity was 20% larger in non-Aboriginal children of both sexes. Similar differences were found for the one-second forced expiratory volume, which was 15% larger for the non-Aboriginal boys and 17% larger for the non-Aboriginal girls. Comment is made on their smoking practices.

Anaemia in Yirrkala

Yirrkala is an Aboriginal community in north-east Arnhem Land in the Northern Territory. Residents aged five years and over participated in a survey to establish the prevalence and causes of anaemia. Eleven per cent were anaemic (haemoglobin level less than 110 g/L). Most of these were iron deficient, and this deficiency was attributed, at least in part, to hookworm infestation; 15% were folate deficient; none was vitamin B12 deficient. There was no haemoglobinopathy, thalassaemia trait or glucose-6-phosphate dehydrogenase (G6PD) deficiency detected.

The use of ultrasound scanning by aboriginal health workers in antenatal care in a remote area of Australia

The value of routine real-time scanning by Aboriginal health workers at antenatal clinics in a remote area of Australia was assessed using data derived from 206 women. The accuracy of the health workers' reports, the detection of high risk factors, the selection of a place for the delivery and the degree of antenatal attendance are discussed.

Obstetrics at Galiwin'ku

A study is presented of the outcome of the 683 deliveries of the Aboriginal women resident on Elcho Island between 1971 and 1982. Comment is made on birth rates, fertility rates, antenatal attendance, risk factors and the outcome of the preterm labours. Current strategies for improving antenatal care are outlined including the use of real time scanning.

A controlled trial of health education in the physician's office

Screening of 6,144 patients in a general practice clinic to assist physician case-finding uncovered 983 (16%) who were uncontrolled hypertensives. Following physician recommendation, 115 patients volunteered for a controlled trial to test the effectiveness of supplementary strategies to the pharmaceutical management of high blood pressure. A study of nonparticipants indicated that about 7% of the practice population was eligible for cardiovascular health education. One group received a health education program, a second was allocated to self-monitor their blood pressure for 6 months, a third group was allocated to both strategies, and the final group, acting as a control, continued to receive their usual care. Physician monitoring of patients continued for the duration of the study and blood pressures decreased in all patients. The study's most important outcome was the joint reduction of blood pressure and medication strength. These were assessed by a "blind" clinician before and after the interventions according to criteria set out in the "stepped-care" approach to management of high blood pressure. People allocated to a health education program conducted in the doctor's common room did twice as well on this measure as those who were not so educated. Daily self-monitoring of blood pressure for 6 months proved to be too much for the majority of those so instructed. It is concluded that the general practice setting remains an important place for health education to prevent cardiac disease and suggestions are made for incorporating this into everyday practice.

General practitioner prescribing habits; The Western Australian experience, 1972-1973

A research project to assess the possibility of modifying general practitioner prescribing habits in Western Australia is described. The results suggest a high level of acceptance of the drug information provided, and a significant alteration in prescribing habits is reported along the lines which might have been predicted.