Research articles

James Liley, John Todd and Chris Wallace present a statistical method for determining whether disease-associated variants have different effect sizes in phenotypically defined subgroups of disease cases. The test can be combined with existing methods to determine whether genetic heterogeneity is driven by population stratification or by different mechanisms of disease pathology.

Yun Song and colleagues present SMC++, a statistical method for population history inference capable of analyzing unphased whole genomes and sample sizes much larger than can be analyzed by current methods. The authors apply SMC++ to sequence data from human, Drosophila and finch populations.

John Storey, David Blei and colleagues present a method, TeraStructure, for estimating population structure from human genomic data sets on a scale not possible with current methods. TeraStructure is able to analyze data from the Human Genome Diversity Panel and the 1000 Genomes Project in less than three hours.

Gill Bejerano and colleagues present M-CAP, a classifier that estimates variant pathogenicity in clinical exome data sets. They show that M-CAP outperforms other existing methods at all thresholds and correctly dismisses 60% of rare missense variants of uncertain significance at 95% sensitivity.

Po-Ru Loh, Alkes Price and colleagues present Eagle2, a reference-based phasing algorithm that allows for highly accurate and efficient phasing of genotypes across a broad range of cohort sizes. They demonstrate an approximately 10% improvement in accuracy and 20% improvement in speed compared to a competing method, SHAPEIT2.

Runjun Kumar, S. Joshua Swamidass and Ron Bose present an unsupervised parsimony-guided method, ParsSNP, for prioritizing candidate cancer driver mutations. They apply ParsSNP to a gastric cancer data set and predict potential driver mutations not detected by other methods, including truncations in known tumor-suppressor genes and previously confirmed drivers.

Victoria Hore, Jonathan Marchini and colleagues present a method for multiple-tissue gene expression studies aimed at uncovering gene networks linked to genetic variation. They apply their method to RNA sequencing data from adipose, skin and lymphoblastoid cell lines and identify several biologically relevant gene networks with a genetic basis.

Richard Mott, Simon Myers and colleagues present a new imputation method, STITCH, which does not require genotyping arrays or high-quality reference panels. They use STITCH to accurately impute genotypes in both outbred laboratory mice and a sample human population directly from low-coverage (<2×) sequencing data.

Po-Ru Loh, Pier Francesco Palamara and Alkes Price develop a new long-range phasing method, Eagle, that harnesses long, shared identical-by-descent tracts and can be applied to large outbred populations. They use Eagle to phase samples from the UK Biobank and find that it is faster and has better accuracy than existing methods.

Jonathan Marchini and colleagues develop a new method for haplotype phasing, SHAPEIT3, capable of handling large data sets from biobanks containing >100,000 genotyped samples. They find that their method is fast and accurate, with a low switch error rate, and can be scaled to data sets from increasingly larger cohorts.

Soumya Raychaudhuri, Buhm Han and colleagues present a statistical method to distinguish whether shared genetic risk variants among complex traits are driven by whole-group pleiotropy or a subset of individuals who constitute a genetically heterogeneous subgroup. They use the method to examine genetic sharing among autoimmune diseases and between major depressive disorder and schizophrenia and find that most genetic sharing cannot be explained by subgroup heterogeneity but that, in contrast, seronegative rheumatoid arthritis is a heterogeneous condition.

Andy Dahl and colleagues present a method for imputing missing phenotype data in genetic studies with multiple correlated phenotypes where samples can have any level of relatedness. They apply their method to simulated and real data sets and show that it improves the sensitivity to detect association signals.

Iuliana Ionita-Laza, Kenneth McCallum and colleagues developed an unsupervised statistical approach, Eigen, that integrates different functional annotations into a single measure of functional importance for coding and noncoding variants. Their meta-score can outperform the recently proposed CADD score and can be applied to fine-mapping studies.