TextStyleBrush: Transfer of Text Aesthetics From a Single Example.

We present a novel approach for disentangling the content of a text image from all aspects of its appearance. The appearance representation we derive can then be applied to new content, for one-shot transfer of the source style to new content. We learn this disentanglement in a self-supervised manner. Our method processes entire word boxes, without requiring segmentation of text from background, per-character processing, or making assumptions on string lengths. We show results in different text domains which were previously handled by specialized methods, e.g., scene text, handwritten text. To these ends, we make a number of technical contributions: (1) We disentangle the style and content of a textual image into a non-parametric, fixed-dimensional vector. (2) We propose a novel approach inspired by StyleGAN but conditioned over the example style at different resolution and content. (3) We present novel self-supervised training criteria which preserve both source style and target content using a pre-trained font classifier and text recognizer. Finally, (4) we also introduce Imgur5K, a new challenging dataset for handwritten word images. We offer numerous qualitative photo-realistic results of our method. We further show that our method surpasses previous work in quantitative tests on scene text and handwriting datasets, as well as in a user study.

Language-Agnostic Reproducible Data Analysis Using Literate Programming.

A modern biomedical research project can easily contain hundreds of analysis steps and lack of reproducibility of the analyses has been recognized as a severe issue. While thorough documentation enables reproducibility, the number of analysis programs used can be so large that in reality reproducibility cannot be easily achieved. Literate programming is an approach to present computer programs to human readers. The code is rearranged to follow the logic of the program, and to explain that logic in a natural language. The code executed by the computer is extracted from the literate source code. As such, literate programming is an ideal formalism for systematizing analysis steps in biomedical research. We have developed the reproducible computing tool Lir (literate, reproducible computing) that allows a tool-agnostic approach to biomedical data analysis. We demonstrate the utility of Lir by applying it to a case study. Our aim was to investigate the role of endosomal trafficking regulators to the progression of breast cancer. In this analysis, a variety of tools were combined to interpret the available data: a relational database, standard command-line tools, and a statistical computing environment. The analysis revealed that the lipid transport related genes LAPTM4B and NDRG1 are coamplified in breast cancer patients, and identified genes potentially cooperating with LAPTM4B in breast cancer progression. Our case study demonstrates that with Lir, an array of tools can be combined in the same data analysis to improve efficiency, reproducibility, and ease of understanding. Lir is an open-source software available at github.com/borisvassilev/lir.

Identification of genes that are essential to restrict genome duplication to once per cell division.

Nuclear genome duplication is normally restricted to once per cell division, but aberrant events that allow excess DNA replication (EDR) promote genomic instability and aneuploidy, both of which are characteristics of cancer development. Here we provide the first comprehensive identification of genes that are essential to restrict genome duplication to once per cell division. An siRNA library of 21,584 human genes was screened for those that prevent EDR in cancer cells with undetectable chromosomal instability. Candidates were validated by testing multiple siRNAs and chemical inhibitors on both TP53+ and TP53- cells to reveal the relevance of this ubiquitous tumor suppressor to preventing EDR, and in the presence of an apoptosis inhibitor to reveal the full extent of EDR. The results revealed 42 genes that prevented either DNA re-replication or unscheduled endoreplication. All of them participate in one or more of eight cell cycle events. Seventeen of them have not been identified previously in this capacity. Remarkably, 14 of the 42 genes have been shown to prevent aneuploidy in mice. Moreover, suppressing a gene that prevents EDR increased the ability of the chemotherapeutic drug Paclitaxel to induce EDR, suggesting new opportunities for synthetic lethalities in the treatment of human cancers.

Elevated levels of StAR-related lipid transfer protein 3 alter cholesterol balance and adhesiveness of breast cancer cells: potential mechanisms contributing to progression of HER2-positive breast cancers.

The STARD3 gene belongs to the minimal amplicon in HER2-positive breast cancers and encodes a cholesterol-binding membrane protein. To study how elevated StAR-related lipid transfer protein 3 (StARD3) expression affects breast cancer cells, we generated MCF-7 cells stably overexpressing StARD3-green fluorescent protein. We found that StARD3-overexpressing cells exhibited nonadherent morphological features, had increased Src levels, and had altered cholesterol balance, as evidenced by elevated mRNA levels of the cholesterol biosynthesis rate-limiting enzyme 3-hydroxy-3-methylglutaryl-coenzyme A reductase, and increased plasma membrane cholesterol content. On removal of serum and insulin from the culture medium, the morphological characteristics of the StARD3-overexpressing cells changed, the cells became adherent, and they developed enlarged focal adhesions. Under these conditions, the StARD3-overexpressing cells maintained elevated Src and plasma membrane cholesterol content and showed increased phosphorylation of focal adhesion kinase. In two Finnish nationwide patient cohorts, approximately 10% (212/2220) breast cancers exhibited high StARD3 protein levels, which was strongly associated with HER2 amplification; several factors related to poor disease outcome and poor breast cancer-specific survival. In addition, high StARD3 levels in breast cancers were associated with elevated 3-hydroxy-3-methylglutaryl-coenzyme A reductase mRNA levels and anti-Src-Tyr416 immunoreactivity. These results provide evidence that StARD3 overexpression results in increased cholesterol biosynthesis and Src kinase activity in breast cancer cells and suggest that elevated StARD3 expression may contribute to breast cancer aggressiveness by increasing membrane cholesterol and enhancing oncogenic signaling.

NDRG1 functions in LDL receptor trafficking by regulating endosomal recycling and degradation.

N-myc downstream-regulated gene 1 (NDRG1) mutations cause Charcot-Marie-Tooth disease type 4D (CMT4D). However, the cellular function of NDRG1 and how it causes CMT4D are poorly understood. We report that NDRG1 silencing in epithelial cells results in decreased uptake of low-density lipoprotein (LDL) due to reduced LDL receptor (LDLR) abundance at the plasma membrane. This is accompanied by the accumulation of LDLR in enlarged EEA1-positive endosomes that contain numerous intraluminal vesicles and sequester ceramide. Concomitantly, LDLR ubiquitylation is increased but its degradation is reduced and ESCRT (endosomal sorting complex required for transport) proteins are downregulated. Co-depletion of IDOL (inducible degrader of the LDLR), which ubiquitylates the LDLR and promotes its degradation, rescues plasma membrane LDLR levels and LDL uptake. In murine oligodendrocytes, Ndrg1 silencing not only results in reduced LDL uptake but also in downregulation of the oligodendrocyte differentiation factor Olig2. Both phenotypes are rescued by co-silencing of Idol, suggesting that ligand uptake through LDLR family members controls oligodendrocyte differentiation. These findings identify NDRG1 as a novel regulator of multivesicular body formation and endosomal LDLR trafficking. The deficiency of functional NDRG1 in CMT4D might impair lipid processing and differentiation of myelinating cells.

Structural fragment clustering reveals novel structural and functional motifs in alpha-helical transmembrane proteins.

BACKGROUND: A large proportion of an organism's genome encodes for membrane proteins. Membrane proteins are important for many cellular processes, and several diseases can be linked to mutations in them. With the tremendous growth of sequence data, there is an increasing need to reliably identify membrane proteins from sequence, to functionally annotate them, and to correctly predict their topology. RESULTS: We introduce a technique called structural fragment clustering, which learns sequential motifs from 3D structural fragments. From over 500,000 fragments, we obtain 213 statistically significant, non-redundant, and novel motifs that are highly specific to alpha-helical transmembrane proteins. From these 213 motifs, 58 of them were assigned to function and checked in the scientific literature for a biological assessment. Seventy percent of the motifs are found in co-factor, ligand, and ion binding sites, 30% at protein interaction interfaces, and 12% bind specific lipids such as glycerol or cardiolipins. The vast majority of motifs (94%) appear across evolutionarily unrelated families, highlighting the modularity of functional design in membrane proteins. We describe three novel motifs in detail: (1) a dimer interface motif found in voltage-gated chloride channels, (2) a proton transfer motif found in heme-copper oxidases, and (3) a convergently evolved interface helix motif found in an aspartate symporter, a serine protease, and cytochrome b. CONCLUSIONS: Our findings suggest that functional modules exist in membrane proteins, and that they occur in completely different evolutionary contexts and cover different binding sites. Structural fragment clustering allows us to link sequence motifs to function through clusters of structural fragments. The sequence motifs can be applied to identify and characterize membrane proteins in novel genomes.