Biology subjects

Sakashita, S.

Publications and source records attributed to Sakashita, S..

4 recordsLinked to original sources

bioRxiv2026

SPATNIC enables H&E-grounded cancer-cell identification across spatial and single-cell transcriptomic modalities in colorectal cancer

Tumors contain diverse non-cancer cells that constitute the tumor microenvironment; thus, identifying which cells are cancerous is a foundational and critical first step in cancer research. However, to our knowledge, no transcriptome-based classifier has been trained using pathologist-defined annotations from matched H&E images at single-cell resolution. Moreover, few tools can be shared across diverse platforms. Here, we present SPATNIC, a variational autoencoder that classifies individual cells as cancer or normal, trained on Xenium cells labeled by a pathologist review of matched H&E images. It matches or exceeds the performance of existing tools in primary colorectal tumors and liver and lung metastases. SPATNIC generalizes zero-shot across Visium, Visium HD, CosMx, Xenium, MERSCOPE, and single-cell RNA sequencing. In public single-cell atlases, it identified a rare POU2F3 tuft-like cancer stem cell population missed by existing annotations and associated with worse survival. In rectal cancer profiled before and after chemoradiotherapy, it separated cancer from normal epithelium in tissue, thus enabling per-patient screening of antibody-drug conjugate (ADC) targets.

cancer biology↗

bioRxiv2026

Single-cell spatial multiomics identifies POSTN+ CAFs mediating chemoradiotherapy resistance in rectal cancer

Neoadjuvant chemoradiotherapy (CRT) is standard for locally advanced rectal cancer (LARC), yet many patients retain residual disease. To resolve CRT-associated remodeling of the tumor microenvironment, we generated a multimodal spatial atlas from serial sections of paired pretreatment and post-treatment specimens from 24 patients using Xenium single-cell spatial transcriptomics and PhenoCycler multiplex proteomics, profiling 2.8 million cells; matched Visium HD datasets were generated on adjacent serial sections. Resistance was most strongly associated with fibroblast and myeloid programs adjacent to residual tumor. We identify a periostin (POSTN)-expressing CAF subset selectively enriched around residual tumor cells in non-responders, displaying a myofibroblastic phenotype and activating extracellular matrix remodeling, noncanonical WNT signaling, and immunosuppressive pathways. Tumor cells neighboring POSTN+ CAFs show consistent epithelial-mesenchymal transition signatures. Together, this atlas enables interrogation of CRT-induced spatial remodeling and nominates POSTN+ CAFs as key mediators and targets of CRT resistance, with direct relevance to CRT-based combination strategies.

cancer biology↗

bioRxiv2025

Label Noise in Pathological Segmentation is Overlooked, Leading to Potential Overestimation of Artificial Intelligence Models

Artificial intelligence (AI) has transformed medical imaging, driving advancements in radiology and endoscopy. Semantic segmentation, a pixel-level technique crucial for delineating pathological features, has become a cornerstone of digital pathology. Pathology segmentation AI models are often trained using annotations generated by pathologists. Despite the meticulous care typically exercised, these annotations frequently contain empirical label noise. However, the specific types of label noise in pathology data and their impact on AI model training remain inadequately explored. This study systematically investigated the effects of label noise on the performance of pathology segmentation models. Using publicly available datasets and a breast cancer semantic segmentation dataset, modules were developed to simulate four types of artificial label noise at varying intensity levels. These datasets were used to train deep learning models with encoder-decoder architectures, and their performance was evaluated using metrics such as the Dice coefficient, precision, recall, and intersection over union. The results indicated that models were highly susceptible to overfitting label noise, particularly boundary-dependent noise such as dilation and shrinkage. Discrepancies were identified between apparent performance scores obtained under real-world conditions and true performance scores derived using clean test data. This overestimation risk was most pronounced for datasets containing boundary-altering noise. Furthermore, random combinations of noise types and levels significantly impaired model generalization. This study underscores the critical importance of addressing label noise in pathology datasets. It is proposed that future efforts focus on developing standardized methods for quantifying and mitigating label noise, along with creating robust benchmarks using noise-inclusive datasets. Enhancing annotation quality and addressing label noise can improve the reliability and generalizability of AI in pathology, facilitating broader clinical adoption.

bioinformatics↗

bioRxiv2020

KuLGaP: A Selective Measure for Assessing Therapy Response in Patient-Derived Xenografts

Quantifying response to drug treatment in mouse models of human cancer is important for treatment development and assignment, and yet remains a challenging task. A preferred measure to quantify this response should take into account as much of the experimental data as possible, i.e. both tumor size over time and the variation among replicates. We propose a theoretically grounded measure, KuLGaP, to compute the difference between the treatment and control arms. KuLGaP is more selective than currently existing measures, reduces the risk of false positive calls and improves translation of the lab results to clinical practice.

bioinformatics↗