神经网络模型对造血干细胞体外扩增效果评价及预测

采用神经网络技术对造血干细胞HSCs体外扩增能力建立了评价及预测模型.总结了前人的实验结果,共获得341组数据.选取细胞接种密度、细胞因子组合、细胞来源、血清、基质细胞、反应器类型和培养时问等7个影响因子作为网络输入特征参数,分别对有核细胞(nUClear cells,NCs)、CD34 细胞和成集落细胞(colony-forming units,CFU-Cs)进行体外扩增能力的拟合评价及预测.选取124、90及86组实验数据分别用于NCs、CFU-Cs和CD34 细胞为评价指标的神经网络训练;而17、10及14组实验数据分别用于NCs、CFU-Cs和CD34 细胞为评价指标的神经网络预测.结果表明.对NCs、CFU-Cs和CD34 细胞的区间训练准确率分别为85.5%、86.7%和86.1%;区间预测准确率分别为82.4%、70.0%和71.4%.由此可知.人工神经网络的非线性模拟使定量描述HSCs体外扩增效果与培养条件问的关系及预测HSCs的最佳体外扩增条件成为可能.

Evaluation and prediction of ex-vivo expansion of HSCs with neural network model

An evaluating and predictive model for the ex-vivo expansion of hematopoietic stem cells (HSCs) was built up with artificial neural network (ANN) technology. 341 groups of data were summarized from literatures, in which 124, 90 and 86 data were employed to train the network and 17, 10 and 14 data were applied to predict respectively. Expansion folds of nuclear cells (NCs), colony-forming units (CFU-Cs) and CD34 + cells were chosen as evaluation objectives and inoculated density, cytokines, cell resources, serum, stromal cells, bioreactor types and culture time were chosen as network inputs. The calculated results show that for the training of network, the interval accuracy of the expansion folds for the different cells is 85.5%, 86.7% and 86.1% respectively. While for the prediction of network, the interval accuracy can be up to 82.4%, 70.0% and 71.4% respectively. Therefore, this nonlinear modeling makes it possible to quantitatively describe the effects of the culture conditions on the HSCs expansion and to predict the optimal culture conditions for higher ex-vivo expansion of HSCs.