Assessing AI Awareness and Identifying Essential Competencies: Insights From Key Stakeholders in Integrating AI Into Medical Education.

Background: The increasing importance of artificial intelligence (AI) in health care has generated a growing need for health care professionals to possess a comprehensive understanding of AI technologies, requiring an adaptation in medical education. Objective: This paper explores stakeholder perceptions and expectations regarding AI in medicine and examines their potential impact on the medical curriculum. This study project aims to assess the AI experiences and awareness of different stakeholders and identify essential AI-related topics in medical education to define necessary competencies for students. Methods: The empirical data were collected as part of the TüKITZMed project between August 2022 and March 2023, using a semistructured qualitative interview. These interviews were administered to a diverse group of stakeholders to explore their experiences and perspectives of AI in medicine. A qualitative content analysis of the collected data was conducted using MAXQDA software. Results: Semistructured interviews were conducted with 38 participants (6 lecturers, 9 clinicians, 10 students, 6 AI experts, and 7 institutional stakeholders). The qualitative content analysis revealed 6 primary categories with a total of 24 subcategories to answer the research questions. The evaluation of the stakeholders' statements revealed several commonalities and differences regarding their understanding of AI. Crucial identified AI themes based on the main categories were as follows: possible curriculum contents, skills, and competencies; programming skills; curriculum scope; and curriculum structure. Conclusions: The analysis emphasizes integrating AI into medical curricula to ensure students' proficiency in clinical applications. Standardized AI comprehension is crucial for defining and teaching relevant content. Considering diverse perspectives in implementation is essential to comprehensively define AI in the medical context, addressing gaps and facilitating effective solutions for future AI use in medical studies. The results provide insights into potential curriculum content and structure, including aspects of AI in medicine.

VIPurPCA: Visualizing and Propagating Uncertainty in Principal Component Analysis.

Variables obtained by experimental measurements or statistical inference typically carry uncertainties. When an algorithm uses such quantities as input variables, this uncertainty should propagate to the algorithm's output. Concretely, we consider the classic notion of principal component analysis (PCA): If it is applied to a finite data matrix containing imperfect (i.e., uncertain) multidimensional measurements, its output-a lower-dimensional representation-is itself subject to uncertainty. We demonstrate that this uncertainty can be approximated by appropriate linearization of the algorithm's nonlinear functionality, using automatic differentiation. By itself, however, this structured, uncertain output is difficult to interpret for users. We provide an animation method that effectively visualizes the uncertainty of the lower dimensional map. Implemented as an open-source software package, it allows researchers to assess the reliability of PCA embeddings.

Improving taxonomic classification with feature space balancing.

Summary: Modern high-throughput sequencing technologies, such as metagenomic sequencing, generate millions of sequences that need to be assigned to their taxonomic rank. Modern approaches either apply local alignment to existing databases, such as MMseqs2, or use deep neural networks, as in DeepMicrobes and BERTax. Due to the increasing size of datasets and databases, alignment-based approaches are expensive in terms of runtime. Deep learning-based approaches can require specialized hardware and consume large amounts of energy. In this article, we propose to use k-mer profiles of DNA sequences as features for taxonomic classification. Although k-mer profiles have been used before, we were able to significantly increase their predictive power significantly by applying a feature space balancing approach to the training data. This greatly improved the generalization quality of the classifiers. We have implemented different pipelines using our proposed feature extraction and dataset balancing in combination with different simple classifiers, such as bagged decision trees or feature subspace KNNs. By comparing the performance of our pipelines with state-of-the-art algorithms, such as BERTax and MMseqs2 on two different datasets, we show that our pipelines outperform these in almost all classification tasks. In particular, sequences from organisms that were not part of the training were classified with high precision. Availability and implementation: The open-source code and the code to reproduce the results is available in Seafile, at https://tinyurl.com/ysk47fmr. Supplementary information: Supplementary data are available at Bioinformatics Advances online.

BLASTphylo: An Interactive Web Tool for Taxonomic and Phylogenetic Analysis of Prokaryotic Genes.

Identifying a protein's function is crucial to reveal its role in the cellular complex. Computationally, the most common approach is to search for homologous proteins in a large database of proteins of known function using BLAST. One goal of such an analysis is the identification and visualization of the protein in the taxonomy of interest. Another goal is the reconstruction of the phylogenetic history of the protein. However, the BLAST result provides information about the occurrence of the protein in the taxonomy and its putative function mainly in a tabular format. This requires manual interventions and makes the taxonomic identification laborious. Although various tools exist to visualize and annotate large-scale trees, none of them intuitively and interactively visualizes the protein's occurrence in the taxonomy for different taxonomic ranks. To target this gap, we developed BLASTphylo, a web tool that combines BLAST with automatic taxonomic mapping and phylogenetic analysis and provides the results in interactive visualizations. We demonstrate the functionalities of BLASTphylo in two case studies.

The Neuromodulator-Encoding sadA Gene Is Widely Distributed in the Human Skin Microbiome.

Trace amines (TA) are endogenously produced in mammals, have a low concentration in the central nervous system (CNS), but trigger a variety of neurological effects and intervene in host cell communication. It emerged that neurotransmitters and TA are produced also by the microbiota. As it has been shown that TA contribute to wound healing, we examined the skin microbiome of probands using shotgun metagenomics. The phyla Actinobacteria, Proteobacteria, Firmicutes, and Bacteroidetes were predominant. Since SadA is a highly promiscuous TA-producing decarboxylase in Firmicutes, the skin microbiome was specifically examined for the presence of sadA-homologous genes. By mapping the reads of certain genes, we found that, although there were less reads mapping to sadA than to ubiquitous housekeeping genes (arcC and mutS), normalized reads counts were still >1000 times higher than those of rare control genes (icaA, icaB, and epiA). At protein sequence level SadA homologs were found in at least 7 phyla: Firmicutes, Actinobacteria, Proteobacteria, Bacteroidetes, Acidobacteria, Chloroflexi, and Cyanobacteria, and in 23 genera of the phylum Firmicutes. A high proportion of the genera that have a SadA homolog belong to the classical skin and intestinal microbiota. The distribution of sadA in so many different phyla illustrates the importance of horizontal gene transfer (HGT). We show that the sadA gene is widely distributed in the human skin microbiome. When comparing the sadA read counts in the probands, there was no correlation between age and gender, but an enormous difference in the sadA read counts in the microbiome of the individuals. Since sadA is involved in TA synthesis, it is likely that the TA content of the skin is correlated with the amount of TA producing bacteria in the microbiome. In this way, the microbiome-generated TA could influence signal transmission in the epithelial and nervous system.

The Genome of Staphylococcus epidermidis O47.

The skin colonizing coagulase-negative Staphylococcus epidermidis causes nosocomial infections and is an important opportunistic and highly adaptable pathogen. To gain more insight into this species, we sequenced the genome of the biofilm positive, methicillin susceptible S. epidermidis O47 strain (hereafter O47). This strain belongs to the most frequently isolated sequence type 2. In comparison to the RP62A strain, O47 can be transformed, which makes it a preferred strain for molecular studies. S. epidermidis O47's genome has a single chromosome of about 2.5 million base pairs and no plasmid. Its oriC sequence has the same directionality as S. epidermidis RP62A, S. carnosus, S. haemolyticus, S. saprophyticus and is inverted in comparison to Staphylococcus aureus and S. epidermidis ATCC 12228. A phylogenetic analysis based on all S. epidermidis genomes currently available at GenBank revealed that O47 is closest related to DAR1907. The genome of O47 contains genes for the typical global regulatory systems known in staphylococci. In addition, it contains most of the genes encoding for the typical virulence factors for S. epidermidis but not for S. aureus with the exception of a putative hemolysin III. O47 has the typical S. epidermidis genetic islands and several mobile genetic elements, which include staphylococcal cassette chromosome (SCC) of about 54 kb length and two prophages φO47A and φO47B. However, its genome has no transposons and the smallest number of insertion sequence (IS) elements compared to the other known S. epidermidis genomes. By sequencing and analyzing the genome of O47, we provide the basis for its utilization in genetic and molecular studies of biofilm formation.

Trace amines produced by skin bacteria accelerate wound healing in mice.

Certain skin bacteria are able to convert aromatic amino acids (AAA) into trace amines (TA) that act as neuromodulators. Since the human skin and sweat contain a comparatively high content of AAA one can expect that such bacteria are able to produce TA on our skin. Here we show that TA-producing Staphylococcus epidermidis strains expressing SadA are predominant on human skin and that TA accelerate wound healing. In wounded skin, keratinocytes produce epinephrine (EPI) that leads to cell motility inhibition by ß2-adrenergic receptor (ß2-AR) activation thus delay wound healing. As ß2-AR antagonists, TA and dopamine (DOP) abrogate the effect of EPI thus accelerating wound healing both in vitro and in a mouse model. In the mouse model, the S. epidermidis wild type strain accelerates wound healing compared to its ΔsadA mutant. Our study demonstrates that TA-producing S. epidermidis strains present on our skin might be beneficial for wound healing.

MpsAB is important for Staphylococcus aureus virulence and growth at atmospheric CO2 levels.

The mechanisms behind carbon dioxide (CO2) dependency in non-autotrophic bacterial isolates are unclear. Here we show that the Staphylococcus aureus mpsAB operon, known to play a role in membrane potential generation, is crucial for growth at atmospheric CO2 levels. The genes mpsAB can complement an Escherichia coli carbonic anhydrase (CA) mutant, and CA from E. coli can complement the S. aureus delta-mpsABC mutant. In comparison with the wild type, S. aureus mps mutants produce less hemolytic toxin and are less virulent in animal models of infection. Homologs of mpsA and mpsB are widespread among bacteria and are often found adjacent to each other on the genome. We propose that MpsAB represents a dissolved inorganic carbon transporter, or bicarbonate concentrating system, possibly acting as a sodium bicarbonate cotransporter.

The GET pathway can increase the risk of mitochondrial outer membrane proteins to be mistargeted to the ER.

Tail-anchored (TA) proteins are anchored to their corresponding membrane via a single transmembrane segment (TMS) at their C-terminus. In yeast, the targeting of TA proteins to the endoplasmic reticulum (ER) can be mediated by the guided entry of TA proteins (GET) pathway, whereas it is not yet clear how mitochondrial TA proteins are targeted to their destination. It has been widely observed that some mitochondrial outer membrane (MOM) proteins are mistargeted to the ER when overexpressed or when their targeting signal is masked. However, the mechanism of this erroneous sorting is currently unknown. In this study, we demonstrate the involvement of the GET machinery in the mistargeting of suboptimal MOM proteins to the ER. These findings suggest that the GET machinery can, in principle, recognize and guide mitochondrial and non-canonical TA proteins. Hence, under normal conditions, an active mitochondrial targeting pathway must exist that dominates the kinetic competition against other pathways.