Acta Chimica Sinica ›› 2008, Vol. 66 ›› Issue (19): 2093-2098.

Original Articles

### 用拓扑指数和神经网络研究有机污染物的生物富集因子

1. (a徐州工程学院化学化工学院 徐州 221008)
(b徐州师范大学化学化工学院 徐州 221116)

• 投稿日期:2008-01-20 修回日期:2008-05-14 发布日期:2008-10-14
• 通讯作者: 冯长君

### Research on the Bioconcentration Factors of Organic Pollutants with Topological Indices and Artificial Neural Network

FENG, Chang-Jun *,a MU, Lai-Long b YANG, Wei-Hua b CAI, Ke-Ying a

1. (a School of Chemistry & Chemical Engineering, Xuzhou Institute of Technology, Xuzhou 221008)
(b School of Chemistry & Chemical Engineering, Xuzhou Normal University, Xuzhou 221116)
• Received:2008-01-20 Revised:2008-05-14 Published:2008-10-14
• Contact: FENG, Chang-Jun

On the basis of the revision of Randic’s molecular connectivity index and conjugation matrix, a novel molecular connectivity index (mF) was defined and calculated for 239 organic pollutant molecules in this paper. The QSAR model of bioconcentration factor (lgBCF) for 239 organic pollutants was constructed from 1F, the traditional correlation coefficient (R2) and the cross-validation correlation coefficient (Q2) of leave-one-out (LOO) were 0.747 and 0.742, respectively. The five-parameter QSAR model was constructed from 1F and the four electronegativity distance vectors (Mk), the traditional correlation coefficient (R2) and the cross-validation correlation coefficient (Q2) were 0.829 and 0.819, respectively. The result demonstrates that the model is highly reliable and has good predictive ability from the point of view of statistics. From the five parameters of the model, it is known that the dominant influence factors of bioconcentration factor are the molecular structure fragments: —C—, ＞C—, —O—, —S—, —X and the space factors: the flexibility and the puckered degree of molecules for organic pollutant. The five structural parameters were used as the input neurons of artificial neural network, and a 5∶26∶1 network architecture was employed. A satisfying model could be constructed with the back-propagation algorithm, with the correlation coefficient R2 and the standard error s being 0.987 and 0.157, respectively, showing that the relationship between lgBCF and five structural parameters has a good nonlinear correlation. The results show that the new parameters 1F and Mk have good rationality and efficiency for the bioconcentration factors of organic compounds. It can be expected that the 1F and Mk will be used widely in quantitative structure-property/activity relationship research.