跳到主要內容區

115.11.26 演講公告

資工系專題演講(Lecture)
日期(date)/時間(time):115年11月26日(四/Thu)13:10~15:00
地點(location):電綜大樓B101演講廳
演講者(speaker):林峻宇 副教授
服務單位(job):國立陽明交通大學 生物資訊及系統生物研究所/數據科學與工程研究所
講題(topic):Patient-Specific Network Modeling for Cancer Treatment Response Prediction: From Graph-Based Proximity to Interpretable AI
摘要(summary):
Predicting individual responses to cancer therapies remains a major computational challenge due to substantial patient-to-patient molecular heterogeneity. Conventional network-based approaches often rely on aggregate protein–protein interaction (PPI) networks, which may overlook patient-specific molecular interactions. In this talk, I will present our recent advances in patient-specific network modeling, integrating graph-based analysis and interpretable machine learning for cancer treatment response prediction.
First, I will introduce SWEET, a method for inferring single-sample networks (SINs) from transcriptomic data, followed by SNUG, a network-based drug efficacy prediction framework. SNUG integrates patient-specific networks, topology-derived hub signatures, and proximity-based hypergeometric test (PHT) scoring to evaluate drug–patient associations. In cancer cell lines, SNUG achieved an AUROC of 0.71, outperforming conventional network proximity methods while reducing computational time.
Next, I will present SNIPE, an interpretable machine-learning framework for predicting immune checkpoint inhibitor responses. SNIPE leverages patient-specific networks and Gscore-based pathway analysis to derive personalized pathway features for machine-learning models. SHAP analysis further provides insights into the biological pathways contributing to model predictions. Across five independent melanoma cohorts comprising more than 300 patients, SNIPE demonstrated improved predictive performance over conventional approaches.
Together, these studies highlight how network inference, graph topology, statistical scoring, and interpretable AI can be integrated to advance personalized cancer treatment response prediction.

與工程認證核心能力關聯性:
☑ 認識時事議題,瞭解資訊工程技術對環境、社會及全球的影響
瀏覽數:
登入成功