OncoLogic: Một chuyên gia dựa trên cơ chế hệ thống dự đoán tiềm năng gây ung thư của Hóa chất Khả năng để dự đoán khả năng gây độc của một hóa chất chưa được kiểm tra dựa trên các mối quan hệ cấu trúc hoạt động (SAR) phân tích phụ thuộc vào các kiến thức về thông tin độc tính trên các hợp chất cấu trúc hoặc chức năng liên quan và các cơ chế có thể (s) của hành động góp phần cụ thể độc tính thiết bị đầu cuối quan tâm | 10 OncoLogic A Mechanism-Based Expert System for Predicting the Carcinogenic Potential of Chemicals YIN-TAK WOO and DAVID Y. LAI Risk Assessment Division Office of Pollution Prevention and Toxics . Environmental Protection Agency Washington . . 1. INTRODUCTION The ability to predict the potential toxicity of an untested chemical based on structure-activity relationships SAR analysis is dependent on the knowledge of toxicological information on structurally or functionally related compounds and of the possible mechanism s of action that contribute to the specific toxicity endpoint of interest. Among all the toxicity endpoints carcinogenicity is undoubtedly one of the most difficult endpoints to predict because of the complexity 385 2005 by Taylor Francis Group LLC 386 Woo and Lai of its mechanisms of action see Sec. 2 below and the difficulty of obtaining a robust well-balanced database needed for effective SAR analysis. Beyond that cancer data are often difficult to interpret or model because of variability associated with long-term studies differences in the species strains of testing animals used the route of administration and the specific testing protocol. Prospective validation and hypothesis testing are also difficult because of the high cost and the long duration of time needed for carcinogenesis bioassays. Despite the difficulties many different SAR QSAR and data mining methods have been developed to predict carcinogenic potential of untested new and existing chemicals. Several review articles 1-8 have evaluated compared and contrasted many of these methods which include a qualitative expert judgment b classical QSAR studies using regression analysis principal component and factor analysis discriminant and pattern recognition analysis similarity analysis neuronal nets etc. c data mining methods involving machine learning to discover SAR features classify active and inactive compounds and or induce knowledge rules d biologically based models such