Computational analysis of crystallization trials
Computational analysis of crystallization trials
复制标题
DOI：
10.1107/s0907444902016840
复制
发表时间：
2002-11-01
期刊：
影响因子：
--
通讯作者：
Priestle, JP
中科院分区：
文献类型：
--
作者：
Spraggon, G;Lesley, SA;Priestle, JP
提出了一种对机器人筛选产生的结晶实验结果进行自动分类的系统。以预设的时间间隔拍摄来自机器人生成的结晶屏幕的图像，并通过计算机程序晶体实验评估程序 (CEEP) 进行分析。该程序试图将各个晶体实验自动分类为许多简单的类别，从透明水滴到可安装晶体。该算法首先通过边缘检测和纹理分析从图像中选择特征。分类是通过自组织神经网络实现的，该神经网络是由用作训练集的一组手动分类图像生成的。然后根据该神经网络对新图像进行分类。事实证明，时间序列信息的结合可以提高分类的准确性。海洋栖热袍菌蛋白质组筛选的初步结果显示了该系统的实用性。
A system for the automatic categorization of the results of crystallization experiments generated by robotic screening is presented. Images from robotically generated crystallization screens are taken at preset time intervals and analyzed by the computer program Crystal Experiment Evaluation Program (CEEP). This program attempts to automatically categorize the individual crystal experiments into a number of simple classes ranging from clear drop to mountable crystal. The algorithm first selects features from the images via edge detection and texture analysis. Classification is achieved via a self-organizing neural net generated from a set of hand-classified images used as a training set. New images are then classified according to this neural net. It is demonstrated that incorporation of time-series information may enhance the accuracy of classification. Preliminary results from the screening of the proteome of Thermotoga maritima are presented showing the utility of the system.