Spurred by advances in processing power, memory, storage, and an unprecedented wealth of data, computers are being asked to tackle increasingly complex learning tasks, often with astonishing success. Computers have now mastered a popular variant of poker, learned the laws of physics from experimental data, and become experts in video games - tasks that would have been deemed impossible not too long ago. In parallel, the number of companies centered on applying complex data analysis to varying industries has exploded, and it is thus unsurprising that some analytic companies are turning attention to problems in health care. The purpose of this review is to explore what problems in medicine might benefit from such learning approaches and use examples from the literature to introduce basic concepts in machine learning. It is important to note that seemingly large enough medical data sets and adequate learning algorithms have been available for many decades, and yet, although there are thousands of papers applying machine learning algorithms to medical data, very few have contributed meaningfully to clinical care. This lack of impact stands in stark contrast to the enormous relevance of machine learning to many other industries. Thus, part of my effort will be to identify what obstacles there may be to changing the practice of medicine through statistical learning approaches, and discuss how these might be overcome.

Background

Truncating mutations in the giant sarcomeric gene Titin are the most common type of genetic alteration in dilated cardiomyopathy. Detailed studies have amassed a wealth of information about truncating variant position in cases and controls. Nonetheless, considerable confusion exists as to how to interpret the pathogenicity of these variants, hindering our ability to make useful recommendations to patients.

Methods and results

Building on our recent discovery of a conserved internal promoter within the Titin gene, we sought to develop an integrative statistical model to explain the observed pattern of Titin truncation variants in patients with dilated cardiomyopathy and population controls. We amassed Titin truncation mutation information from 1714 human dilated cardiomyopathy cases and >69 000 controls and found 3 factors explaining the distribution of Titin mutations: (1) alternative splicing, (2) whether the internal promoter Cronos isoform was disrupted, and (3) whether the distal C terminus was targeted (in keeping with the observation that truncation variants in this region escape nonsense-mediated decay and continue to be incorporated in the sarcomere). A model using these 3 factors had strong predictive performance with an area under the receiver operating characteristic curve of 0.81. Accordingly, individuals with either the most severe form of dilated cardiomyopathy or whose mutations demonstrated clear family segregation experienced the highest risk profile across all 3 components.

Conclusions

We conclude that quantitative models derived from large-scale human genetic and phenotypic data can be applied to help overcome the ever-growing challenges of genetic data interpretation. Results of our approach can be found at http://cvri.ucsf.edu/~deo/TTNtruncationvariant.html.

The full range of cell functions is under-determined in most human diseases. The evidence that somatic cell competition and clonal imbalance play a role in non-neoplastic chronic disease reveal a need for a dedicated effort to explore single cell function if we are to understand the mechanisms by which cell population behaviors influence disease. It will be vital to document not only the prevalent pathologic behaviors but also those beneficial functions eliminated or suppressed by competition. An improved mechanistic understanding of the role of somatic cell biology will help to stratify chronic disease, define more precisely at an individual level the role of environmental factors and establish principles for prevention and potential intervention throughout the life course and across the trajectory from wellness to disease.

Background

The ECG remains the most widely used diagnostic test for characterization of cardiac structure and electrical activity. We hypothesized that parallel advances in computing power, machine learning algorithms, and availability of large-scale data could substantially expand the clinical inferences derived from the ECG while at the same time preserving interpretability for medical decision-making.

Methods and results

We identified 36 186 ECGs from the University of California, San Francisco database that would enable training of models for estimation of cardiac structure or function or detection of disease. We segmented the ECG into standard component waveforms and intervals using a novel combination of convolutional neural networks and hidden Markov models and evaluated this segmentation by comparing resulting electrical intervals against 141 864 measurements produced during the clinical workflow. We then built a patient-level ECG profile, a 725-element feature vector and used this profile to train and interpret machine learning models for examples of cardiac structure (left ventricular mass, left atrial volume, and mitral annulus e-prime) and disease (pulmonary arterial hypertension, hypertrophic cardiomyopathy, cardiac amyloid, and mitral valve prolapse). ECG measurements derived from the convolutional neural network-hidden Markov model segmentation agreed with clinical estimates, with median absolute deviations as a fraction of observed value of 0.6% for heart rate and 4% for QT interval. Models trained using patient-level ECG profiles enabled surprising quantitative estimates of left ventricular mass and mitral annulus e' velocity (median absolute deviation of 16% and 19%, respectively) with good discrimination for left ventricular hypertrophy and diastolic dysfunction as binary traits. Model performance using our approach for disease detection demonstrated areas under the receiver operating characteristic curve of 0.94 for pulmonary arterial hypertension, 0.91 for hypertrophic cardiomyopathy, 0.86 for cardiac amyloid, and 0.77 for mitral valve prolapse.

Conclusions

Modern machine learning methods can extend the 12-lead ECG to quantitative applications well beyond its current uses while preserving the transparency that is so fundamental to clinical care.

Rationale

Treatment of sinus node disease with regenerative or cell-based therapies will require a detailed understanding of gene regulatory networks in cardiac pacemaker cells (PCs).

Objective

To characterize the transcriptome of PCs using RNA sequencing and to identify transcriptional networks responsible for PC gene expression.

Methods and results

We used laser capture microdissection on a sinus node reporter mouse line to isolate RNA from PCs for RNA sequencing. Differential expression and network analysis identified novel sinoatrial node-enriched genes and predicted that the transcription factor Islet-1 is active in developing PCs. RNA sequencing on sinoatrial node tissue lacking Islet-1 established that Islet-1 is an important transcriptional regulator within the developing sinoatrial node.

Conclusions

(1) The PC transcriptome diverges sharply from other cardiomyocytes; (2) Islet-1 is a positive transcriptional regulator of the PC gene expression program.

The striking rise of obesity-related metabolic disorders has focused attention on adipocytes as critical mediators of disease phenotypes. To better understand the role played by excess adipose in metabolic dysfunction it is crucial to decipher the transcriptional underpinnings of the low-grade adipose inflammation characteristic of diseases such as type 2 diabetes. Through employing a comparative transcriptomics approach, we identified IRF1 as differentially regulated between primary and in vitro-derived genetically matched adipocytes. This suggests a role as a mediator of adipocyte inflammatory phenotypes, similar to its function in other tissues. Utilizing adipose-derived mesenchymal progenitors we subsequently demonstrated that expression of IRF1 in adipocytes indeed contributes to upregulation of inflammatory processes, both in vitro and in vivo. This highlights IRF1's relevance to obesity-related inflammation and the resultant metabolic dysregulation.

Background

Researchers routinely evaluate novel biomarkers for incorporation into clinical risk models, weighing tradeoffs between cost, availability, and ease of deployment. For risk assessment in population health initiatives, ideal inputs would be those already available for most patients. We hypothesized that common hematologic markers (eg, hematocrit), available in an outpatient complete blood count without differential, would be useful to develop risk models for cardiovascular events.

Methods

We developed Cox proportional hazards models for predicting heart attack, ischemic stroke, heart failure hospitalization, revascularization, and all-cause mortality. For predictors, we used 10 hematologic indices (eg, hematocrit) from routine laboratory measurements, collected March 2016 to May 2017 along with demographic data and diagnostic codes. As outcomes, we used neural network-based automated event adjudication of 1 028 294 discharge summaries. We trained models on 23 238 patients from one hospital in Boston and evaluated them on 29 671 patients from a second one. We assessed calibration using Brier score and discrimination using Harrell's concordance index. In addition, to determine the utility of high-dimensional interactions, we compared our proportional hazards models to random survival forest models.

Results

Event rates in our cohort ranged from 0.0067 to 0.075 per person-year. Models using only hematology indices had concordance index ranging from 0.60 to 0.80 on an external validation set and showed the best discrimination when predicting heart failure (0.80 [95% CI, 0.79-0.82]) and all-cause mortality (0.78 [0.77-0.80]). Compared with models trained only on demographic data and diagnostic codes, models that also used hematology indices had better discrimination and calibration. The concordance index of the resulting models ranged from 0.75 to 0.85 and the improvement in concordance index ranged up to 0.072. Random survival forests had minimal improvement over proportional hazards models.

Conclusions

We conclude that low-cost, ubiquitous inputs, if biologically informative, can provide population-level readouts of risk.

In vertebrates, activation of innate immunity is an early response to injury, implicating it in the regenerative process. However, the mechanisms by which innate signals might regulate stem cell functionality are unknown. Here, we demonstrate that type 2 innate immunity is required for regeneration of skeletal muscle after injury. Muscle damage results in rapid recruitment of eosinophils, which secrete IL-4 to activate the regenerative actions of muscle resident fibro/adipocyte progenitors (FAPs). In FAPs, IL-4/IL-13 signaling serves as a key switch to control their fate and functions. Activation of IL-4/IL-13 signaling promotes proliferation of FAPs to support myogenesis while inhibiting their differentiation into adipocytes. Surprisingly, type 2 cytokine signaling is also required in FAPs, but not in myeloid cells, for rapid clearance of necrotic debris, a process that is necessary for timely and complete regeneration of tissues.

Mutations in the calcium-binding protein calsequestrin cause the highly lethal familial arrhythmia catecholaminergic polymorphic ventricular tachycardia (CPVT). In vivo, calsequestrin multimerizes into filaments, but there is not yet an atomic-resolution structure of a calsequestrin filament. We report a crystal structure of a human cardiac calsequestrin filament with supporting mutational analysis and in vitro filamentation assays. We identify and characterize a new disease-associated calsequestrin mutation, S173I, that is located at the filament-forming interface, and further show that a previously reported dominant disease mutation, K180R, maps to the same surface. Both mutations disrupt filamentation, suggesting that disease pathology is due to defects in multimer formation. An ytterbium-derivatized structure pinpoints multiple credible calcium sites at filament-forming interfaces, explaining the atomic basis of calsequestrin filamentation in the presence of calcium. Our study thus provides a unifying molecular mechanism through which dominant-acting calsequestrin mutations provoke lethal arrhythmias.

For placental mammals, the transition from the in utero maternal environment to postnatal life requires the activation of thermogenesis to maintain their core temperature. This is primarily accomplished by induction of uncoupling protein 1 (UCP1) in brown and beige adipocytes, the principal sites for uncoupled respiration. Despite its importance, how placental mammals license their thermogenic adipocytes to participate in postnatal uncoupled respiration is not known. Here, we provide evidence that the "alarmin" IL-33, a nuclear cytokine that activates type 2 immune responses, licenses brown and beige adipocytes for uncoupled respiration. We find that, in absence of IL-33 or ST2, beige and brown adipocytes develop normally but fail to express an appropriately spliced form of Ucp1 mRNA, resulting in absence of UCP1 protein and impairment in uncoupled respiration and thermoregulation. Together, these data suggest that IL-33 and ST2 function as a developmental switch to license thermogenesis during the perinatal period. PAPERCLIP.