In the analysis of current genomic data, application of machine learning and data mining techniques has become more attractive given the rising complexity of the projects. As part of the Genetic Analysis Workshop 19, approaches from this domain were explored, mostly motivated from two starting points. First, assuming an underlying structure in the genomic data, data mining might identify this and thus improve downstream association analyses. Second, computational methods for machine learning need to be developed further to efficiently deal with the current wealth of data. In the course of discussing results and experiences from the machine learning and data mining approaches, six common messages were extracted. These depict the current state of these approaches in the application to complex genomic data. Although some challenges remain for future studies, important forward steps were taken in the integration of different data types and the evaluation of the evidence. Mining the data for underlying genetic or phenotypic structure and using this information in subsequent analyses proved to be extremely helpful and is likely to become of even greater use with more complex data sets.

Complex diseases are defined to be determined by multiple genetic and environmental factors alone as well as in interactions. To analyze interactions in genetic data, many statistical methods have been suggested, with most of them relying on statistical regression models. Given the known limitations of classical methods, approaches from the machine-learning community have also become attractive. From this latter family, a fast-growing collection of methods emerged that are based on the Multifactor Dimensionality Reduction (MDR) approach. Since its first introduction, MDR has enjoyed great popularity in applications and has been extended and modified multiple times. Based on a literature search, we here provide a systematic and comprehensive overview of these suggested methods. The methods are described in detail, and the availability of implementations is listed. Most recent approaches offer to deal with large-scale data sets and rare variants, which is why we expect these methods to even gain in popularity.

Coronary artery disease (CAD) is the leading global cause of mortality and has substantial heritability with a polygenic architecture. Recent approaches of risk prediction were based on polygenic risk scores (PRS) not taking possible nonlinear effects into account and restricted in that they focused on genetic loci associated with CAD, only. We benchmarked PRS, (penalized) logistic regression, naïve Bayes (NB), random forests (RF), support vector machines (SVM), and gradient boosting (GB) on a data set of 7,736 CAD cases and 6,774 controls from Germany to identify the algorithms for most accurate classification of CAD status. The final models were tested on an independent data set from Germany (527 CAD cases and 473 controls). We found PRS to be the best algorithm, yielding an area under the receiver operating curve (AUC) of 0.92 (95% CI [0.90, 0.95], 50,633 loci) in the German test data. NB and SVM (AUC ~ 0.81) performed better than RF and GB (AUC ~ 0.75). We conclude that using PRS to predict CAD is superior to machine learning methods.

In the context of the COVID-19 pandemic, wearable sensors are important for early detection of critical illness especially in COVID-19 outpatients. We sought to determine in this pilot study whether a wearable in-ear sensor for continuous body temperature and heart rate monitoring (Cosinuss company, Munich) is sufficiently accurate for body temperature and heart rate monitoring. Comparing with several anesthesiologic standard of care monitoring devices (urinary bladder and zero-heat flux thermometer and ECG), we evaluated the in-ear sensor during non-cardiac surgery (German Clinical Trials Register Reg.-No: DRKS00012848). Limits of Agreement (LoA) based on Bland–Altman analysis were used to study the agreement between the in-ear sensor and the reference methods. The estimated LoA of the Cosinuss One and bladder temperature monitoring were [-0.79, 0.49] °C (95% confidence intervals [-1.03, -0.65] (lower LoA) and [0.35, 0.73] (upper LoA)), and [-0.78, 0.34] °C (95% confidence intervals [-1.18, -0.59] (lower LoA) and [0.16, 0.74] (upper LoA)) of the Cosinuss One and zero-heat flux temperature monitoring. 89% and 79% of Cosinuss One temperature monitoring were within ± 0.5 °C limit of bladder and zero-heat flux monitoring, respectively. The estimated LoA of Cosinuss One and ECG heart rate monitoring were [-4.81, 4.27] BPM (95% confidence intervals [-5.09, -4.56] (lower LoA) and [4.01, 4.54] (upper LoA)). The proportion of detection differences within ± 2BPM was 84%. Body temperature and heart rate were reliably measured by the wearable in-ear sensor.