I am a Ph.D. candidate at the Department of Biomedical Engineering, Rensselaer Polytechnic Institute, USA, co-advised by Prof. Ge Wang and Prof. Pingkun Yan at the Biomedical Imaging Center. My current research focuses on medical imaging and multimodal analytical frameworks that jointly leverage CT imaging, genetic data, and clinical variables to improve CVD risk prediction. Previously, I obtained my B.Sc. in Physics from the University of Science and Technology of China. In my undergraduate studies, I focused on wavefront modulation methods using deep learning, under the supervision of Prof. Xiaoye Xu and Prof. Zhiwei Xiong.
Prof. Ge Wang’s Lab: AI-based X-ray Imaging System (AXIS) lab
Prof. Pingkun Yan’s Lab: Deep Imaging Analytics Lab (DIAL)
Ph.D. Student in Biomedical Engineering, 2022-present
Rensselaer Polytechnic Institute
B.Sc. in Physics, 2022
University of Science and Technology of China

Cardiovascular disease (CVD) modeling has become highly accurate, yet most models remain ``black boxes’’ that prioritize predictive scores over reliable and reusable clinical knowledge. We propose Evidence-Grounded Learning (EGL), which repeatedly tests candidate variable sets under a fixed evaluation budget and records auditable evidence about which factors and pairwise relationships are consistently supported. EGL–LLM adds an interpretable large language model (LLM) control plane that uses a semantic agenda to decide what to test next, while keeping the evidence definitions and budget unchanged. On the COPDGene cohort, EGL shows reliable convergence of core evidence and recovers clinically plausible dominant factors and cohort-relevant interaction patterns. Semantic steering broadens the exploration path while preserving stable dominant factors and redirecting evidence acquisition toward alternative interaction neighborhoods, supporting evidence-grounded discovery that yields auditable cohort-level insights. Code is available at https://github.com/liangyuxuan1/EGL.

Deciphering which transcription factor (TF) motif instances are functionally required for enhancer activity typically demands ChIP-based assays or labor-intensive perturbation experiments. We introduce a virtual motif-perturbation framework that uses AlphaGenome, a large sequence-to-function foundation model, to infer motif-level regulatory contribution directly from DNA sequence. Candidate C/EBP$mathrmβ $ motifs are identified within chromatin-active regions and systematically ablated in silico; the resulting changes in predicted regulatory activity are quantified and assessed against a sham-derived null distribution to establish statistical confidence. To evaluate whether sequence-level perturbations recapitulate biologically meaningful regulatory dependence, we compared in silico predictions with CUT&RUN measurements of H3K27ac following CEBPB knockout in multiple myeloma cells. Key activating motifs exhibited concordant loss of H3K27ac across both settings, whereas loci with apparent discrepancies reflected biologically interpretable mechanisms. Together, these findings suggest that large sequence-based models can approximate the directional consequences of TF perturbation, supporting motif-level functional analysis directly from DNA sequence. This supports sequence-only virtual ablation as a scalable framework for hypothesis generation and regulatory annotation that can be extended to additional TFs, cell types, and chromatin modalities.

Fairness of artificial intelligence and machine learning models, often caused by imbalanced datasets, has long been a concern. While many efforts aim to minimize model bias, this study suggests that traditional fairness evaluation methods may be biased, highlighting the need for a proper evaluation scheme with multiple evaluation metrics due to varying results under different criteria. Moreover, the limited data size of minority groups introduces significant data uncertainty, which can undermine the judgement of fairness. This paper introduces an innovative evaluation approach that estimates data uncertainty in minority groups through bootstrapping from majority groups for a more objective statistical assessment. Extensive experiments reveal that traditional evaluation methods might have drawn inaccurate conclusions about model fairness. The proposed method delivers an unbiased fairness assessment by adeptly addressing the inherent complications of model evaluation on imbalanced datasets. The results show that such comprehensive evaluation can provide more confidence when adopting those models.
Please leave a message if you have any questions.