Biology subjects

Fujimoto, H.

Publications and source records attributed to Fujimoto, H..

4 recordsLinked to original sources

bioRxiv2025

Designing Cell Cell Relation Extraction Models: A Systematic Evaluation of Entity Representation and Pre-training Strategies

Extracting cell-cell relations from biomedical literature is essential for understanding intercellular communication in immunity, inflammation, and tissue biology. However, cell-cell relation extraction has not been established as a standalone biomedical relation extraction task, and no benchmark corpus or systematic evaluation framework currently exists. Fully manual corpus construction is costly and difficult to scale, limiting literature-based analyses of cell-cell communication. Here, we define a sentence-level cell-cell relation extraction task and construct complementary manually annotated corpora under realistic annotation constraints. To enable scalable annotation, rule-based literature mining is used solely as an annotation accelerator to identify candidate sentences, while all relation labels are assigned manually. In addition, an independently annotated PubMed corpus without rule-based filtering is constructed to evaluate robustness on natural sentence distributions. Using these resources, we evaluate representative model configurations involving entity indication strategies, classification architectures, and continued pre-training. Our results show that cell-cell relation extraction remains challenging under realistic conditions. Increasing training data size yields consistent performance gains, and specific combinations of entity-aware architectures and continued pre-training provide modest robustness improvements. Nevertheless, performance on unfiltered PubMed sentences remains in the 70% accuracy range, and error analyses indicate that failures cannot be readily explained by simple surface-level factors. Comparisons with general-purpose large language models further suggest that task complexity, rather than model class, is the primary limiting factor. Together, these findings establish a practical foundation for literature-scale cell-cell relation extraction while clarifying its intrinsic limitations.

bioinformatics↗

bioRxiv2025

N-cadherin in osteolineage cells restrains breast cancer cell growth via inhibition of a PI3K-dependent, Tgf-β1-driven feed-forward loop

Tumor growth and metastases are affected by interactions between tumor and microenvironment cells. We have reported the presence of Sp7-positive cells with an osteogenic signature in primary mouse breast cancer, where they stimulate tumor growth; and genetic ablation of Cdh2 (encoding N-cadherin) in these cells enhances their pro-tumorigenic action. To study the molecular mechanisms of this biologic system, we used MC3T3 cells, phenotypically similar to tumor associated osteolineage cells. Ablation of Cdh2 in MC3T3 cells enhances PI3K-Akt-{beta}-catenin signaling in response to transforming growth factor-{beta}1 (Tgf-{beta}1), resulting in increased production of Tgf-{beta}1. Interference with PI3K activity is mediated by N-cadherin binding to PI3K components, p85 and p100, resulting in reduced activation of PI3K-Akt-{beta}-catenin signaling. Downstream, Cdh2 ablation enhances Tgf-{beta}1-induced binding of Sp1 and Lef-1 to the Tgfb1 promoter, leading to increased promoter activity and enhanced Tgf-{beta}1 production. This is associated with miR-21 up-regulation and decreased expression of Pten, a PI3K inhibitor. MC3T3 cells promote growth of breast cancer cells (BCC) when co-cultured in vitro or co-injected in mouse mammary fat pad, and Cdh2-deficiency enhances this pro-tumorigenic effect. Notably, genetic ablation of the Tgf-{beta}1 receptor subunit, Tgfbr1, in BCC abrogates the pro-tumorigenic action of MC3T3 cells and its enhancement by Cdh2 ablation. Finally, Sp7-driven Tgfbr1 ablation in mice also reduces the growth of BCC in the mammary fat pad. Thus, autocrine Tgf-{beta}1 production via PI3K-Akt-{beta}-catenin signaling is a key mechanism by which osteolineage cells promote BCC growth, an action restrained by Ncad via interference with PI3K components. HighlightsO_LIIn tumor-associated osteolineage cells, N-cadherin reduces the growth of breast tumors; it also inhibits Tgf-{beta}1-activated PI3K/AKT/{beta}-catenin signaling C_LIO_LITgf-{beta}1 produced by tumor microenvironment cells stimulates breast cancer cell growth and further autocrine production of Tgf-{beta}1 C_LIO_LIThe anti-tumorigenic action of N-cadherin is mediated by a braking effect on a Tgf-{beta}1-driven, pro-tumorigenic, feed-forward cycle between microenvironment and tumor cells. C_LI

cancer biology↗

bioRxiv2022

Localized assembly for long reads enables genome-wide analysis of repetitive regions at single-base resolution in human genomes

BackgroundLong-read sequencing technologies have the potential to overcome the limitations of short reads and provide a comprehensive picture of the human genome. However, it remains hard to characterize repetitive sequences by reconstructing genomic structures at high resolution solely from long reads. Here, we developed a localized assembly method (LoMA) that constructs highly accurate consensus sequences (CSs) from long reads. MethodsWe first developed LoMA, by combining minimap2, MAFFT, and our algorithm, which classifies diploid haplotypes based on structural variants and constructs CSs. Using this tool, we analyzed two human samples (NA18943 and NA19240) sequenced with the Oxford Nanopore sequencer. We defined target regions in each genome based on mapping patterns and then constructed a high-quality catalog of the human insertion solely from the long-read data. ResultsThe assessment of LoMA showed high accuracy of CSs (error rate < 0.3%) compared with raw data (error rate > 8%) and superiority to the previous study. The genome-wide analysis of NA18943 and NA19240 identified 5,516 and 6,542 insertions ({zeta} 100 bp) respectively. Most insertions ([~]80%) were derived from the tandem repeat and transposable elements. We also detected processed pseudogenes, insertions in transposable elements, and long insertions (> 10 kbp). Further, our analysis suggested that short tandem duplications were association with gene expression and transposons. ConclusionsOur analysis showed that LoMA constructs high-quality sequences from long reads with substantial errors. This study revealed the true structures of insertions with high accuracy and inferred mechanisms for the insertions. Our approach contributes to the future human genome studies. LoMA is available at our GitHub page: https://github.com/kolikem/loma.

genomics↗

bioRxiv2018

Acetylation of nuclear localization signal controls importin-mediated nuclear transport of Ku70

Ku70 participates in various intra-and extra-nucleic processes. For multifunctional control, machinery that precisely regulates the intracellular localization of Ku70 is essential. Recently, it was reported that acetylation of Ku70 regulates its function. Here, we demonstrate that specific lysine residues in Ku70 that are targets of acetylation are critical for regulating nuclear transport in vivo. Ku70-GFP fusion proteins transiently expressed in cultured cells localized in the nucleus, whereas mimicking acetylation of K553 or K556 in the Ku70 nuclear localization signal (NLS) by substituting these lysine residues with glutamine markedly decreased the nuclear localization of Ku70. Moreover, the Ku70-importin interaction was suppressed in the K553Q and K556Q mutants. Theoretical estimations indicated that the binding energy between the Ku70 NLS and importin- decreases with acetylation of lysine residues in the Ku70 NLS, similar to the case when these lysine residues are substituted with glutamine. These results suggest that acetylation of specific lysine residues in the Ku70 NLS is a key switch that controls the localization of Ku70 by modulating interactions between Ku70 and nuclear transport factors.

cell biology↗