Phylogenetic inference, grounded in molecular evolution models, is essential for understanding the evolutionary relationships in biological data. Accounting for the uncertainty of phylogenetic tree variables, which include tree topologies and evolutionary distances on branches, is crucial for accurately inferring species relationships from molecular data and tasks requiring variable marginalization. Variational Bayesian methods are key to developing scalable, practical models; however, it remains challenging to conduct phylogenetic inference without restricting the combinatorially vast number of possible tree topologies. In this work, we introduce a novel, fully differentiable formulation of phylogenetic inference that leverages a unique representation of topological distributions in continuous geometric spaces. Through practical considerations on design spaces and control variates for gradient estimations, our approach, GeoPhy, enables variational inference without limiting the topological candidates. In experiments using real benchmark datasets, GeoPhy significantly outperformed other approximate Bayesian methods that considered whole topologies.

We propose an evaluation framework for class probability estimates (CPEs) in the presence of label uncertainty, which is commonly observed as diagnosis disagreement between experts in the medical domain. We also formalize evaluation metrics for higher-order statistics, including inter-rater disagreement, to assess predictions on label uncertainty. Moreover, we propose a novel post-hoc method called alpha-calibration, that equips neural network classifiers with calibrated distributions over CPEs. Using synthetic experiments and a large-scale medical imaging application, we show that our approach significantly enhances the reliability of uncertainty estimates: disagreement probabilities and posterior CPEs.

BACKGROUND: With the recent development of microarray and high-throughput sequencing (HTS) technologies, a number of studies have revealed catalogs of copy number variants (CNVs) and their association with phenotypes and complex traits. In parallel, a number of approaches to predict CNV regions and genotypes are proposed for both microarray and HTS data. However, only a few approaches focus on haplotyping of CNV loci. RESULTS: We propose a novel approach to infer copy unit alleles and their numbers in each sample simultaneously from population-scale HTS data by variational Bayesian inference on a generative probabilistic model inspired by latent Dirichlet allocation, which is a well studied model for document classification problems. In simulation studies, we evaluated concordance between inferred and true copy unit alleles for lower-, middle-, and higher-copy number dataset, in which precision and recall were ≥ 0.9 for data with mean coverage ≥ 10× per copy unit. We also applied the approach to HTS data of 1123 samples at highly variable salivary amylase gene locus and a pseudogene locus, and confirmed consistency of the estimated alleles within samples belonging to a trio of CEPH/Utah pedigree 1463 with 11 offspring. CONCLUSIONS: Our proposed approach enables detailed analysis of copy number variations, such as association study between copy unit alleles and phenotypes or biological features including human diseases.

MOTIVATION: Variant calling from genome-wide sequencing data is essential for the analysis of disease-causing mutations and elucidation of disease mechanisms. However, variant calling in low coverage regions is difficult due to sequence read errors and mapping errors. Hence, variant calling approaches that are robust to low coverage data are demanded. RESULTS: We propose a new variant calling approach that considers pedigree information and haplotyping based on sequence reads spanning two or more heterozygous positions termed phase informative reads. In our approach, genotyping and haplotyping by the assignment of each read to a haplotype based on phase informative reads are simultaneously performed. Therefore, positions with low evidence for heterozygosity are rescued by phase informative reads, and such rescued positions contribute to haplotyping in a synergistic way. In addition, pedigree information supports more accurate haplotyping as well as genotyping, especially in low coverage regions. Although heterozygous positions are useful for haplotyping, homozygous positions are not informative and weaken the information from heterozygous positions, as majority of positions are homozygous. Thus, we introduce latent variables that determine zygosity at each position to filter out homozygous positions for haplotyping. In performance evaluation with a parent-offspring trio sequencing data, our approach outperforms existing approaches in accuracy on the agreement with single nucleotide polymorphism array genotyping results. Also, performance analysis considering distance between variants showed that the use of phase informative reads is effective for accurate variant calling, and further performance improvement is expected with longer sequencing data. CONTACT: [email protected] .