One of the first steps in the analysis of single-cell RNA sequencing (scRNA-seq) data is the assignment of cell types. Although a number of supervised methods have been developed for this, in most cases such assignment is performed by first clustering cells in low-dimensional space and then assigning cell types to different clusters. To overcome noise and to improve cell type assignments, we developed UNIFAN, a neural network method that simultaneously clusters and annotates cells using known gene sets. UNIFAN combines both low-dimensional representation for all genes and cell-specific gene set activity scores to determine the clustering. We applied UNIFAN to human and mouse scRNA-seq data sets from several different organs. We show, by using knowledge about gene sets, that UNIFAN greatly outperforms prior methods developed for clustering scRNA-seq data. The gene sets assigned by UNIFAN to different clusters provide strong evidence for the cell type that is represented by this cluster, making annotations easier.

Variation within human genomes is unevenly distributed, and variants show spatial clustering. DNA-replication-related template switching is a poorly known mutational mechanism capable of causing major chromosomal rearrangements as well as creating short inverted sequence copies that appear as local mutation clusters in sequence comparisons. I reanalyzed haplotype-resolved genome assemblies representing 25 human populations and multinucleotide variants aggregated from 140,000 human sequencing experiments. Local template switching could explain thousands of complex mutation clusters across the human genome, the loci segregating within and between populations. I developed computational tools for identification of template switch events using both short-read sequencing data and genotype data, and for genotyping candidate loci using short-read data. The characteristics of template-switch mutations complicate their detection, and widely used analysis pipelines for short-read sequencing data, normally capable of identifying single nucleotide changes, were found to miss template-switch mutations of tens of base pairs, potentially invalidating medical genetic studies searching for a causative allele behind genetic diseases. Combined with the massive sequencing data now available for humans, the novel tools described here enable building catalogs of affected loci and studying the cellular mechanisms behind template switching in both healthy organisms and disease.

The advent of massively parallel sequencing revealed extensive transcription beyond protein-coding genes, identifying tens of thousands of long noncoding RNAs (lncRNAs). Selected functional examples raised the possibility that lncRNAs, as a class, may maintain broad regulatory roles. Expression of lncRNAs is strongly linked with adjacent protein-coding gene expression, suggesting potential cis-regulatory functions. A more detailed understanding of these regulatory roles may be obtained through careful examination of the precise timing of lncRNA expression relative to adjacent protein-coding genes. Despite the diversity of reported lncRNA regulatory mechanisms, where causal cis-regulatory relationships exist, lncRNA transcription is expected to precede changes in target gene expression. Using a high temporal resolution RNA-seq time course, we profiled the expression dynamics of several thousand lncRNAs and protein-coding genes in synchronized, transitioning human cells. Our findings reveal that lncRNAs are expressed synchronously with adjacent protein-coding genes. Analysis of lipopolysaccharide-activated mouse dendritic cells revealed the same temporal relationship observed in transitioning human cells. Our findings suggest broad-scale cis-regulatory roles for lncRNAs are not common. The strong association between lncRNAs and adjacent genes may instead indicate an origin as transcriptional by-products from active protein-coding gene promoters and enhancers.

Clinical exome sequencing has yielded extensive disease-related missense single-nucleotide variants (SNVs) of uncertain significance, leading to diagnostic uncertainty. KCNQ4 is one of the most commonly responsible genes for autosomal dominant nonsyndromic hearing loss. According to the gnomAD cohort, approximately one in 100 people harbors missense variants in KCNQ4 (missense variants with minor allele frequency > 0.1% were excluded), but most are of unknown consequence. To prospectively characterize the function of all 4085 possible missense SNVs of human KCNQ4, we recorded the whole-cell currents using the patch-clamp technique and categorized 1068 missense SNVs as loss of function, as well as 728 loss-of-function SNVs located in the transmembrane domains. Further, to mimic the heterozygous condition in Deafness nonsyndromic autosomal dominant 2 (DFNA2) patients caused by KCNQ4 variants, we coexpressed loss-of-function variants with wild-type KCNQ4 and found 516 variants showed impaired or only partially rescued heterogeneous channel function. Overall, our functional classification is highly concordant with the auditory phenotypes in Kcnq4 mutant mice and the assessments of pathogenicity in clinical variant interpretations. Taken together, our results provide strong functional evidence to support the pathogenicity classification of newly discovered KCNQ4 missense variants in clinical genetic testing.

The successful discovery of novel biological therapeutics by selection requires highly diverse libraries of candidate sequences that contain a high proportion of desirable candidates. Here we propose the use of computationally designed factorizable libraries made of concatenated segment libraries as a method of creating large libraries that meet an objective function at low cost. We show that factorizable libraries can be designed efficiently by representing objective functions that describe sequence optimality as an inner product of feature vectors, which we use to design an optimization method we call stochastically annealed product spaces (SAPS). We then use this approach to design diverse and efficient libraries of antibody CDR-H3 sequences with various optimized characteristics.

Mediator is a conserved coregulator playing a key role in RNA polymerase (Pol) II transcription. Mediator also links transcription and nucleotide excision repair (NER) via a direct contact with Rad2/ERCC5(XPG) endonuclease. In this work, we analyzed the genome-wide distribution of Rad26/ERCC6(CSB) and Rad1-Rad10/ERCC4(XPF)-ERCC1, addressing the question of a potential link of these proteins with Mediator and Pol II in yeast Saccharomyces cerevisiae Our genomic analyses reveal that Rad1-Rad10 and Rad26 are present on the yeast genome in the absence of genotoxic stress, especially at highly transcribed regions, with Rad26 binding strongly correlating with that of Pol II. Moreover, we show that Rad1-Rad10 and Rad26 colocalize with Mediator at intergenic regions and physically interact with this complex. Using kin28 TFIIH mutant, we found that Mediator stabilization on core promoters leads to an increase in Rad1-Rad10 chromatin binding, whereas Rad26 occupancy follows mainly a decrease in Pol II transcription. Combined with multivariate analyses, our results show the relationships between Rad1-Rad10, Rad26, Mediator, and Pol II, modulated by the changes in binding dynamics of Mediator and Pol II transcription. In conclusion, we extend the Mediator link to Rad1-Rad10 and Rad26 NER proteins and reveal important differences in their dependence on Mediator and Pol II. Rad2 is the most dependent on Mediator, followed by Rad1-Rad10, whereas Rad26 is the most closely related to Pol II. Our work thus contributes to new concepts of the functional interplay between transcription and DNA repair machineries, which are relevant for human diseases including cancer and XP/CS syndromes.