Biology subjects

Kodera, S.

Publications and source records attributed to Kodera, S..

3 recordsLinked to original sources

bioRxiv2026

REN-former prioritizes candidate regulators of kidney disease-state transitions through single-cell foundation modeling and human genetics

Background Acute kidney injury (AKI)-to-chronic kidney disease (CKD) transition is associated with dynamic changes in tubular cell state. However, conventional single-cell transcriptomic analyses primarily identify genes differentially expressed between disease states and do not directly evaluate genes associated with directional transitions between cellular states. Methods We developed REN-former by fine-tuning Geneformer using the GSE183276 single-cell RNA-sequencing dataset from 45 participants representing Normal Reference, AKI, and CKD states. In silico gene deletion and overexpression analyses were used to estimate directional transcriptomic shifts, with a focus on proximal tubular cells. Selected genes were evaluated using summary-data-based Mendelian randomization (SMR), colocalization and expression analysis in additional KPMP participants not included in GSE183276. Results REN-former achieved recall values of 0.99, 0.80, and 0.79 for Normal Reference, AKI, and CKD, respectively. In silico perturbation analyses identified distinct gene programs associated with transitions from Normal Reference to AKI, from Normal Reference to CKD, from AKI to CKD, and from CKD to Normal Reference. Conventional analysis showed metabolic suppression and increased inflammatory and stress-response activation. SMR identified IFITM3, CALR, TTR, CALM1, MUC13, and RPL13, and colocalization supported IFITM3, CALR, TTR, and CALM1. In additional KPMP data, IFITM3 was higher, whereas TTR and CALM1 were lower, in CKD proximal tubules; CALR did not differ significantly. The observed expression changes were concordant with the REN-former-predicted directions for TTR and CALM1 but discordant for IFITM3. Conclusion REN-former provides a framework for prioritizing candidate regulators of kidney disease-associated cell states by integrating predicted perturbation effects with human genetic and transcriptomic evidence.

bioinformatics↗

bioRxiv2025

PAH-former: Transfer Learning for Efficient Discovery of Pulmonary Arterial Hypertension-Associated Genes

Single-cell RNA sequencing (scRNA-seq) of patient samples holds promise for understanding disease mechanisms, but faces the challenge of excessive cost and effort in acquisition, processing, and data analysis, making it essential to leverage existing data. Pulmonary artery hypertension (PAH) is a refractory disease characterized by pulmonary vascular remodeling, and access to patient specimens is limited due to difficulties in tissue collection. In this study, we employed transfer learning with Geneformer, a deep learning algorithm pre-trained with scRNA-seq datasets and fine-tuned it with public PAH lung tissue data to identify the disease-relevant genes. The resulting algorithm, which we named PAH- former, demonstrated that its prediction accuracy varied significantly depending on the dataset used for fine-tuning. PAH-former enabled us to perform in silico perturbation analysis and identified PAH related genes. Loss-of-function PAH related genes in human pulmonary artery endothelial cells increased the expression of SOX18, a signature gene of PAH. This integration of artificial intelligence and biological experiments can significantly advance our understanding of molecular mechanisms of PAH.

bioinformatics↗

bioRxiv2021

Deep learning-based chest X-ray age serves as a novel biomarker for cardiovascular aging

Chest X-ray (CXR) is one of the most commonly performed medical imaging tests. Although aging, sex and disease status have been known to cause changes in CXR findings, the extent of these effects has not been fully characterized. Here, we present a deep neural network (DNN) model trained using more than 100,000 CXRs to estimate the patients age and sex solely from CXRs. Our DNN exhibited high performance in terms of estimating age and sex, with Pearsons correlation coefficient between the actual and estimated age of above 0.9 and an area under the ROC curve of 0.98 for sex estimation. The difference between the actual and estimated age is large in CXRs with abnormal findings, suggesting that the estimated age ("CXR age") can be a biomarker for disease status. Furthermore, by applying our DNN to CXRs of consecutive 1,562 hospitalized heart failure patients, we demonstrated that an elevated CXR age is not only associated with aging-related diseases, such as hypertension and atrial fibrillation, but also a worse outcome of heart failure. Given these results, our new concept "CXR age" serves as a novel biomarker for cardiovascular aging and can help clinicians to predict, prevent, and manage cardiovascular diseases.

bioinformatics↗