Presenting artificial intelligence, deep learning, and machine learning studies to clinicians and healthcare stakeholders: an introductory reference with a guideline and a Clinical AI Research (CAIR) checklist proposal.

Background and purpose - Artificial intelligence (AI), deep learning (DL), and machine learning (ML) have become common research fields in orthopedics and medicine in general. Engineers perform much of the work. While they gear the results towards healthcare professionals, the difference in competencies and goals creates challenges for collaboration and knowledge exchange. We aim to provide clinicians with a context and understanding of AI research by facilitating communication between creators, researchers, clinicians, and readers of medical AI and ML research.Methods and results - We present the common tasks, considerations, and pitfalls (both methodological and ethical) that clinicians will encounter in AI research. We discuss the following topics: labeling, missing data, training, testing, and overfitting. Common performance and outcome measures for various AI and ML tasks are presented, including accuracy, precision, recall, F1 score, Dice score, the area under the curve, and ROC curves. We also discuss ethical considerations in terms of privacy, fairness, autonomy, safety, responsibility, and liability regarding data collecting or sharing.Interpretation - We have developed guidelines for reporting medical AI research to clinicians in the run-up to a broader consensus process. The proposed guidelines consist of a Clinical Artificial Intelligence Research (CAIR) checklist and specific performance metrics guidelines to present and evaluate research using AI components. Researchers, engineers, clinicians, and other stakeholders can use these proposal guidelines and the CAIR checklist to read, present, and evaluate AI research geared towards a healthcare setting.

Subdivision of the MDR superfamily of medium-chain dehydrogenases/reductases through iterative hidden Markov model refinement.

BACKGROUND: The Medium-chain Dehydrogenases/Reductases (MDR) form a protein superfamily whose size and complexity defeats traditional means of subclassification; it currently has over 15000 members in the databases, the pairwise sequence identity is typically around 25%, there are members from all kingdoms of life, the chain-lengths vary as does the oligomericity, and the members are partaking in a multitude of biological processes. There are profile hidden Markov models (HMMs) available for detecting MDR superfamily members, but none for determining which MDR family each protein belongs to. The current torrential influx of new sequence data enables elucidation of more and more protein families, and at an increasingly fine granularity. However, gathering good quality training data usually requires manual attention by experts and has therefore been the rate limiting step for expanding the number of available models. RESULTS: We have developed an automated algorithm for HMM refinement that produces stable and reliable models for protein families. This algorithm uses relationships found in data to generate confident seed sets. Using this algorithm we have produced HMMs for 86 distinct MDR families and 34 of their subfamilies which can be used in automated annotation of new sequences. We find that MDR forms with 2 Zn2+ ions in general are dehydrogenases, while MDR forms with no Zn2+ in general are reductases. Furthermore, in Bacteria MDRs without Zn2+ are more frequent than those with Zn2+, while the opposite is true for eukaryotic MDRs, indicating that Zn2+ has been recruited into the MDR superfamily after the initial life kingdom separations. We have also developed a web site http://mdr-enzymes.org that provides textual and numeric search against various characterised MDR family properties, as well as sequence scan functions for reliable classification of novel MDR sequences. CONCLUSIONS: Our method of refinement can be readily applied to create stable and reliable HMMs for both MDR and other protein families, and to confidently subdivide large and complex protein superfamilies. HMMs created using this algorithm correspond to evolutionary entities, making resolution of overlapping models straightforward. The implementation and support scripts for running the algorithm on computer clusters are available as open source software, and the database files underlying the web site are freely downloadable. The web site also makes our findings directly useful also for non-bioinformaticians.

Superfamilies SDR and MDR: from early ancestry to present forms. Emergence of three lines, a Zn-metalloenzyme, and distinct variabilities.

Two large gene and protein superfamilies, SDR and MDR (short- and medium-chain dehydrogenases/reductases), were originally defined from analysis of alcohol and polyol dehydrogenases. The superfamilies contain minimally 82 and 25 genes, respectively, in humans, minimally 324 and 86 enzyme families when known lines in other organisms are also included, and over 47,000 and 15,000 variants in existing sequence data bank entries. SDR enzymes have one-domain subunits without metal and MDR two-domain subunits without or with zinc, and these three lines appear to have emerged in that order from the universal cellular ancestor. This is compatible with their molecular architectures, present multiplicity, and overall distribution in the kingdoms of life, with SDR also of viral occurrence. An MDR-zinc, when present, is often, but not always, catalytic. It appears also to have a structural role in inter-domain interactions, coenzyme binding and substrate pocket formation, as supported by domain variability ratios and ligand positions. Differences among structural and catalytic zinc ions may be relative and involve several states. Combined, the comparisons trace evolutionary properties of huge superfamilies, with partially redundant enzymes in cellular redox functions.

Quantitative membrane proteomics applying narrow range peptide isoelectric focusing for studies of small cell lung cancer resistance mechanisms.

Drug resistance is often associated with upregulation of membrane-associated drug-efflux systems, and thus global membrane proteomics methods are valuable tools in the search for novel components of drug resistance phenotypes. Herein we have compared the microsomal proteome from the lung cancer cell line H69 and its isogenic Doxorubicin-resistant subcell line H69AR. The method used includes microsome preparation, iTRAQ labeling followed by narrow range peptide IEF in an immobilized pH-gradient (IPG-IEF) and LC-MS/MS analysis. We demonstrate that the microsomal preparation and iTRAQ labeling is reproducible regarding protein content and composition. The rationale using narrow range peptide IPG-IEF separation is demonstrated by its ability to: (i) lowering the complexity of the sample by two-thirds while keeping high proteome coverage (96%), (ii) providing high separation efficiency, and (iii) allowing for peptide validation and possibly identifications of post-transcriptional modifications. After analyzing one-fifth of the IEF fractions (effective pH range of 4.0-4.5), a total of 3704 proteins were identified, among which 527 were predicted to be membrane proteins. One of the proteins found to be differentially expressed was Serca 2, a calcium pump located in the ER membrane that potentially could result in changes of apoptotic response toward Doxorubicin.

Analysis of ancient sequence motifs in the H-PPase family.

The unique family of membrane-bound proton-pumping inorganic pyrophosphatases, involving pyrophosphate as the alternative to ATP, was investigated by characterizing 166 members of the UniProtKB/Swiss-Prot + UniProtKB/TrEMBL databases and available completed genomes, using sequence comparisons and a hidden Markov model based upon a conserved 57-residue region in the loop between transmembrane segments 5 and 6. The hidden Markov model was also used to search the approximately one million sequences recently reported from a large-scale sequencing project of organisms in the Sargasso Sea, resulting in additional 164 partial pyrophosphatase sequences. The strongly conserved 57-residue region was found to contain two nonapeptidyl sequences, mainly consisting of the four 'very early' proteinaceous amino acid residues Gly, Ala, Val and Asp, compatible with an ancient origin of the inorganic pyrophosphatases. The nonapeptide patterns have charged amino acid residues at positions 1, 5 and 9, are apparent binding sites for the substrate and parts of the active site, and were shown to be so specific for these enzymes that they can be used for functional assignments of unannotated genomes.

Origin and evolution of medium chain alcohol dehydrogenases.

Different lines of alcohol dehydrogenases (ADHs) have separate superfamily origins, already recognized but now extended and re-evaluated by re-screening of the latest databank update. The short-chain form (SDR) is still the superfamily with most abundant occurrence, most multiple divergence, most prokaryotic emphasis, and most non-complicated architecture. This pattern is compatible with an early appearance at the time of the emergence of prokaryotic cellular life. The medium-chain form (MDR) is also old but second in terms of all the parameters above, and therefore compatible with a second emergence. However, this step appears seemingly earlier than previously considered, and may indicate sub-stages of early emergences at the increased resolution available from the now greater number of data entries. The Zn-MDR origin constitutes a third stage, possibly compatible with the transition to oxidative conditions on earth. Within all these three lines, repeated enzymogeneses gave the present divergence. MDR-ADH origin(s), at a fourth stage, may also be further resolved in multiple or extended modes, but the classical liver MDR-ADH of the liver type can still be traced to a gene duplication ~550 MYA (million years ago), at the early vertebrate radiation, compatible with the post-eon-shift, "Cambrian explosion". Classes and isozymes correspond to subsequent and recent duplicatory events, respectively. They illustrate a peculiar pattern with functional and emerging evolutionary distinctions between parent and emerging lines, suggesting a parallelism between duplicatory and mutational events, now also visible at separate sub-stages. Combined, all forms show distinctive patterns at different levels and illustrate correlations with global events. They further show that simple molecular observations on patterns, multiplicities and occurrence give much information, suggesting common divergence rules not much disturbed by horizontal gene transfers after the initial origins.

Characterization of new medium-chain alcohol dehydrogenases adds resolution to duplications of the class I/III and the sub-class I genes.

Four additional variants of alcohol and aldehyde dehydrogenases have been purified and functionally characterized, and their primary structures have been determined. The results allow conclusions about the structural and evolutionary relationships within the large family of MDR alcohol dehydrogenases from characterizations of the pigeon (Columba livia) and dogfish (Scyliorhinus canicula) major liver alcohol dehydrogenases. The pigeon enzyme turns out to be of class I type and the dogfish enzyme of class III type. This result gives a third type of evidence, based on purifications and enzyme characterization in lower vertebrates, that the classical liver alcohol dehydrogenase originated by a gene duplication early in the evolution of vertebrates. It is discernable as the major liver form at about the level in-between cartilaginous and osseous fish. The results also show early divergence within the avian orders. Structures were determined by Edman degradations, making it appropriate to acknowledge the methodological contributions of Pehr Edman during the 65 years since his thesis at Karolinska Institutet, where also the present analyses were performed.

BRICHOS - a superfamily of multidomain proteins with diverse functions.

BACKGROUND: The BRICHOS domain has been found in 8 protein families with a wide range of functions and a variety of disease associations, such as respiratory distress syndrome, dementia and cancer. The domain itself is thought to have a chaperone function, and indeed three of the families are associated with amyloid formation, but its structure and many of its functional properties are still unknown. FINDINGS: The proteins in the BRICHOS superfamily have four regions with distinct properties. We have analysed the BRICHOS proteins focusing on sequence conservation, amino acid residue properties, native disorder and secondary structure predictions. Residue conservation shows large variations between the regions, and the spread of residue conservation between different families can vary greatly within the regions. The secondary structure predictions for the BRICHOS proteins show remarkable coherence even where sequence conservation is low, and there seems to be little native disorder. CONCLUSIONS: The greatly variant rates of conservation indicates different functional constraints among the regions and among the families. We present three previously unknown BRICHOS families; group A, which may be ancestral to the ITM2 families; group B, which is a close relative to the gastrokine families, and group C, which appears to be a truly novel, disjoint BRICHOS family. The C-terminal region of group C has nearly identical sequences in all species ranging from fish to man and is seemingly unique to this family, indicating critical functional or structural properties.

Organelle transport in melanophores analyzed by white light image correlation spectroscopy.

Intracellular transport of organelles, vesicles and proteins is crucial in all eukaryotic cells, and is accomplished by motor proteins that move along cytoskeletal filaments. A widely used model of intracellular transport is Xenopus laevis melanophores. These cells help the frog to change color by redistributing melanin-containing organelles in the cytoplasm. The high contrast of the pigment organelles permits changes in distribution to be observed by ordinary light microscopy; other intracellular transport systems often require fluorescence labeling. Here we have developed white light Image Correlation Spectroscopy (ICS) to monitor aggregation and dispersion of pigment. Hitherto in ICS, images of fluorescent particles from Confocal Laser Scanning Microscopy (CLSM) have been used to calculate autocorrelation functions from which the density can be obtained. In the present study we show that ICS can be modified to enable analysis of light-microscopy images; it can be used to monitor pigment aggregation and dispersion, and distinguish between different stimuli. This new approach makes ICS applicable not only to fluorescent but also to black-and-white images from light or electron microscopy, and is thus very versatile in different studies of movement of particles on the membrane or in the cytoplasm of cells without potentially harmful fluorescence labeling and activation.