6 research outputs found
Image‐Contrast Technology Based on the Electrochemiluminescence of Porous Silicon and Its Application in Fingerprint Visualization
No full textMOAI: Module-Optimizing Architecture for Non-Interactive Secure Transformer Inference
No full textPrivacy concerns have been raised in Large Language Models (LLM) inference when models are deployed in Cloud Service Providers (CSP). Homomorphic encryption (HE) offers a promising solution by enabling secure inference directly over encrypted inputs. However, the high computational overhead of HE remains a major bottleneck. To address this challenge, we propose MOAI, an efficient HE-based, non-interactive framework for secure transformer inference. MOAI gains significant efficiency improvement from: (1) a novel evaluation flow that combines column and diagonal packing with consistent strategies across all layers, eliminating expensive format conversions. (2) rotation-free algorithms for Softmax and LayerNorm that significantly reduce the number of costly HE rotations, removing 2448 HE rotations in BERT-base inference. (3) Column packing removes rotations in plaintext–ciphertext matrix multiplications and interleaved batching further reduces the rotations in ciphertext–ciphertext matrix multiplications. MOAI uses at least 1.7x fewer HE rotations compared to the state-of-the-art works across all matrix multiplications of BERT-base. As a result, We achieve a 52.8% reduction in evaluation time compared to the state-of-the-art HE-based non-interactive secure transformer inference, THOR (Moon et al., CCS’25). We then apply MOAI on the Powerformer’s framework and achieve a 55.7% reduction in evaluation time compared to Powerformer (Park et al., ACL’25), which approximates Softmax and LayerNorm with simpler functions in transformer and proposes HE-based non-interactive transformer inference. We report an amortized time of 2.36 minutes per input on a single GPU environment. We show the extendibility by applying MOAI in LLaMA-3-8B. Our implementation is publicly available as open source
Machine learning integration of single-cell and bulk transcriptomics identifies fibroblast-driven prognostic markers in colorectal cancer
No full textSingle-cell RNA sequencing (scRNA-seq) has significantly advanced our understanding of cellular heterogeneity and the complex interplay within the tumor microenvironment (TME) of colorectal cancer (CRC). However, translating these molecular insights into clinically actionable prognostic biomarkers and therapeutic strategies remains a considerable challenge. In this study, we conducted a comprehensive scRNA-seq analysis of 306 CRC samples comprising 448,255 cells to characterize the TME in depth. By constructing intercellular communication networks based on connection counts and communication probabilities, we identified fibroblasts as central regulatory hubs within the TME. Using Wilcoxon rank-sum tests and univariate survival analyses, we initially identified 23 prognostic fibroblast markers. These were refined to a seven-gene fibroblast-related prognostic signature via an integrated machine learning approach. The signature exhibited robust predictive performance in the The Cancer Genome Atlas - Colon Adenocarcinoma (TCGA-COAD) training cohort (n=351; C-index=0.65) and was successfully validated in the GSE17536 dataset (n=177; C-index=0.63). Functional enrichment analyses revealed that this signature is involved in immune regulation and multiple tumor-associated cellular pathways. Notably, high-risk patients displayed increased macrophage and NK cell infiltration, impaired immune function, and elevated immune rejection scores, while low-risk patients demonstrated heightened sensitivity to camptothecin and irinotecan. Together, our findings underscore the prognostic value of fibroblast-derived signatures in CRC and support their potential utility in risk stratification and the development of personalized therapeutic strategies, contributing to the advancement of precision oncology
Single-Molecule Discrimination of Saccharides Using Carbon Nitride Nanopores
No full textStructural complexity brings a huge challenge to the
analysis of
sugar chains. As a single-molecule sensor, nanopores have the potential
to provide fingerprint information on saccharides. Traditionally,
direct single-molecule saccharide detection with nanopores is hampered
by their small size and weak affinity. Here, a carbon nitride nanopore
device is developed to discern two types of trisaccharide molecules
(LeApN and SLeCpN) with minor structural differences.
The resolution of LeApN and SLeCpN in the mixture
reaches 0.98, which has never been achieved in solid-state nanopores
so far. Monosaccharide (GlcNAcpN) and disaccharide (LacNAcpN) can also be discriminated using this system, indicating
that the versatile carbon nitride nanopores possess a monosaccharide-level
resolution. This study demonstrates that the carbon nitride nanopores
have the potential for conducting structure analysis on single-molecule
saccharides
Lsr2 is a nucleoid-associated protein that targets AT-rich sequences and virulence genes in Mycobacterium tuberculosis
No full textBacterial nucleoid-associated proteins play important roles in chromosome organization and global gene regulation. We find that Lsr2 of Mycobacterium tuberculosis is a unique nucleoid-associated protein that binds AT-rich regions of the genome, including genomic islands acquired by horizontal gene transfer and regions encoding major virulence factors, such as the ESX secretion systems, the lipid virulence factors PDIM and PGL, and the PE/PPE families of antigenic proteins. Comparison of genome-wide binding data with expression data indicates that Lsr2 binding results in transcriptional repression. Domain-swapping experiments demonstrate that Lsr2 has an N-terminal dimerization domain and a C-terminal DNA-binding domain. Nuclear magnetic resonance analysis of the DNA-binding domain of Lsr2 and its interaction with DNA reveals a unique structure and a unique mechanism that enables Lsr2 to discriminately target AT-rich sequences through interactions with the minor groove of DNA. Taken together, we provide evidence that mycobacteria have employed a structurally distinct molecule with an apparently different DNA recognition mechanism to achieve a function similar to the Enterobacteriaceae H-NS, likely coordinating global gene regulation and virulence in this group of medically important bacteria.Multidisciplinary SciencesSCI(E)61ARTICLE115154-515910
