Selected work

Representative Papers

CRC-HGD: A Histopathological Image Dataset for Grading Colorectal Cancer

Jul 14, 2026

Current histopathological image datasets for colorectal cancer generally lack comprehensive coverage of all WHO differentiation grades and multi-magnification registered samples, limiting the application of AI in automated cancer grading. This study addresses this gap by constructing a high-quality, publicly available dataset comprising 1,914 H&E-stained images from 214 colorectal adenocarcinoma patients, encompassing well-, moderately, and poorly differentiated tumors. For each patient, precisely registered images across four magnifications (4× to 40×) are provided. Annotated according to WHO standards and curated through standardized acquisition and release protocols, the dataset is openly shared via Mendeley Data and databiox.com. It represents the first resource to simultaneously offer multi-grade, multi-magnification, and patient-level registered pathological images for colorectal cancer, substantially enhancing the comprehensiveness and reliability of AI model training and evaluation.

0 citationsRead paper

DrugGen 2: A disease-aware language model for enhancing drug discovery

Jul 09, 2026

This study addresses the frequent oversight in existing drug design approaches of how disease context influences target protein behavior, often resulting in generated molecules with limited therapeutic relevance. To bridge this gap, the authors propose a novel GPT-2–based dual-condition molecular generation framework that explicitly incorporates both disease ontologies and target protein sequences into the generative process. The model is trained using supervised fine-tuning followed by group relative policy optimization (GRPO), a reinforcement learning strategy that jointly optimizes chemical validity, novelty, diversity, and high predicted binding affinity. Experiments on five targets associated with diabetic nephropathy demonstrate that the generated molecules exhibit structural similarity to approved drugs and significantly outperform baseline methods in predicted binding affinity, with several candidates surpassing enalapril—a reference drug—in silico, thereby indicating enhanced clinical potential.

0 citationsRead paper

Small Language Models for Privacy-Preserving Clinical Information Extraction in Low-Resource Languages

Feb 24, 2026

This study addresses the challenges of clinical information extraction from low-resource language medical texts, including data scarcity, high annotation costs, and privacy constraints. The authors propose a lightweight, tuning-free two-stage approach: first, anonymized Persian medical transcripts are translated into English using Aya-expanse-8B; then, thirteen binary clinical attributes are extracted in parallel via few-shot prompting on compact open-source models such as Qwen2.5, Llama-3.1/3.2, and Gemma-3. Experimental results show that Qwen2.5-7B-Instruct achieves the best performance (median macro F1 of 0.899, MCC of 0.797). Larger models significantly outperform smaller ones in sensitivity and MCC. Translating to English improves sensitivity at a slight cost to specificity. Extraction of physiological symptoms proves relatively reliable, whereas psychological and complex signs remain challenging. This work demonstrates the effectiveness of jointly optimizing model scale and input language, offering a practical, privacy-preserving pathway for low-resource clinical settings.

0 citationsRead paper

AAVGen: Precision Engineering of Adeno-associated Viral Capsids for Renal Selective Targeting

Feb 21, 2026

Natural adeno-associated virus (AAV) capsids exhibit significant limitations in renal tropism, immune evasion, and production efficiency. To address these challenges, this study introduces AAVGen, a generative AI framework that integrates the ESM-2 protein language model, supervised fine-tuning, and a novel Group Sequence Policy Optimization reinforcement learning strategy. This approach enables, for the first time, the simultaneous optimization of AAV capsids across three critical objectives: manufacturability, kidney targeting, and thermal stability. The designed VP1 sequences outperform natural capsids in all three key performance metrics while maintaining correct structural folding as validated by AlphaFold3, thereby demonstrating the feasibility and efficacy of de novo multi-trait capsid engineering.

0 citationsRead paper

Hierarchical Generalized Category Discovery for Brain Tumor Classification in Digital Pathology

Oct 03, 2025

Existing brain tumor classification methods rely on predefined categories and thus fail to identify novel tumor types absent from training data; while Generalized Category Discovery (GCD) enables joint partitioning of known and unknown classes, it neglects the inherent hierarchical taxonomic structure of brain tumors. Method: We propose HGCD-BT, the first GCD framework incorporating hierarchical class structure, which integrates contrastive learning with semi-supervised hierarchical clustering and introduces a novel hierarchical clustering loss. This loss leverages labeled-data priors while modeling hierarchical semantic relationships among unlabeled samples. Contribution/Results: Evaluated on the OpenSRH dataset, HGCD-BT achieves a 28% absolute improvement in classification accuracy over state-of-the-art GCD methods. Moreover, it demonstrates strong generalization capability for slice-level classification of multimodal digital pathology images.

0 citationsRead paper

Recent publications

Latest Papers

CRC-HGD: A Histopathological Image Dataset for Grading Colorectal Cancer

Jul 14, 2026

Current histopathological image datasets for colorectal cancer generally lack comprehensive coverage of all WHO differentiation grades and multi-magnification registered samples, limiting the application of AI in automated cancer grading. This study addresses this gap by constructing a high-quality, publicly available dataset comprising 1,914 H&E-stained images from 214 colorectal adenocarcinoma patients, encompassing well-, moderately, and poorly differentiated tumors. For each patient, precisely registered images across four magnifications (4× to 40×) are provided. Annotated according to WHO standards and curated through standardized acquisition and release protocols, the dataset is openly shared via Mendeley Data and databiox.com. It represents the first resource to simultaneously offer multi-grade, multi-magnification, and patient-level registered pathological images for colorectal cancer, substantially enhancing the comprehensiveness and reliability of AI model training and evaluation.

0 citationsRead paper

DrugGen 2: A disease-aware language model for enhancing drug discovery

Jul 09, 2026

This study addresses the frequent oversight in existing drug design approaches of how disease context influences target protein behavior, often resulting in generated molecules with limited therapeutic relevance. To bridge this gap, the authors propose a novel GPT-2–based dual-condition molecular generation framework that explicitly incorporates both disease ontologies and target protein sequences into the generative process. The model is trained using supervised fine-tuning followed by group relative policy optimization (GRPO), a reinforcement learning strategy that jointly optimizes chemical validity, novelty, diversity, and high predicted binding affinity. Experiments on five targets associated with diabetic nephropathy demonstrate that the generated molecules exhibit structural similarity to approved drugs and significantly outperform baseline methods in predicted binding affinity, with several candidates surpassing enalapril—a reference drug—in silico, thereby indicating enhanced clinical potential.

0 citationsRead paper

Small Language Models for Privacy-Preserving Clinical Information Extraction in Low-Resource Languages

Feb 24, 2026

This study addresses the challenges of clinical information extraction from low-resource language medical texts, including data scarcity, high annotation costs, and privacy constraints. The authors propose a lightweight, tuning-free two-stage approach: first, anonymized Persian medical transcripts are translated into English using Aya-expanse-8B; then, thirteen binary clinical attributes are extracted in parallel via few-shot prompting on compact open-source models such as Qwen2.5, Llama-3.1/3.2, and Gemma-3. Experimental results show that Qwen2.5-7B-Instruct achieves the best performance (median macro F1 of 0.899, MCC of 0.797). Larger models significantly outperform smaller ones in sensitivity and MCC. Translating to English improves sensitivity at a slight cost to specificity. Extraction of physiological symptoms proves relatively reliable, whereas psychological and complex signs remain challenging. This work demonstrates the effectiveness of jointly optimizing model scale and input language, offering a practical, privacy-preserving pathway for low-resource clinical settings.

0 citationsRead paper

AAVGen: Precision Engineering of Adeno-associated Viral Capsids for Renal Selective Targeting

Feb 21, 2026

Natural adeno-associated virus (AAV) capsids exhibit significant limitations in renal tropism, immune evasion, and production efficiency. To address these challenges, this study introduces AAVGen, a generative AI framework that integrates the ESM-2 protein language model, supervised fine-tuning, and a novel Group Sequence Policy Optimization reinforcement learning strategy. This approach enables, for the first time, the simultaneous optimization of AAV capsids across three critical objectives: manufacturability, kidney targeting, and thermal stability. The designed VP1 sequences outperform natural capsids in all three key performance metrics while maintaining correct structural folding as validated by AlphaFold3, thereby demonstrating the feasibility and efficacy of de novo multi-trait capsid engineering.

0 citationsRead paper

Hierarchical Generalized Category Discovery for Brain Tumor Classification in Digital Pathology

Oct 03, 2025

Existing brain tumor classification methods rely on predefined categories and thus fail to identify novel tumor types absent from training data; while Generalized Category Discovery (GCD) enables joint partitioning of known and unknown classes, it neglects the inherent hierarchical taxonomic structure of brain tumors. Method: We propose HGCD-BT, the first GCD framework incorporating hierarchical class structure, which integrates contrastive learning with semi-supervised hierarchical clustering and introduces a novel hierarchical clustering loss. This loss leverages labeled-data priors while modeling hierarchical semantic relationships among unlabeled samples. Contribution/Results: Evaluated on the OpenSRH dataset, HGCD-BT achieves a 28% absolute improvement in classification accuracy over state-of-the-art GCD methods. Moreover, it demonstrates strong generalization capability for slice-level classification of multimodal digital pathology images.

0 citationsRead paper