Application of Improved Three-Dimensional Kernel Approach to Prediction of Protein Structural Class

Table 1

LOOCV success rates by component-coupled, neural network, SVMs, AdaBoost, and improved 3D kernel approach.

Dataset	Algorithm	All-𝛼	All-𝛽	𝛼/𝛽	𝛼+𝛽	Overall
Dataset A (277 domains)	Component-coupled	84.3%	82.0%	81.5%	67.7%	79.1%
	Neural networks	68.6%	85.2%	86.4%	56.9%	74.7%
	SVMs	74.3%	82.0%	87.7%	72.3%	79.4%
	AdaBoost	87.1%	95.1%	98.7%	81.5%	90.9%
	3D kernel	88.6%	85.3%	93.8%	77.0%	86.6%
Dataset B (498 domains)	Component-coupled	93.5%	88.9%	90.4%	84.5%	89.2%
	Neural networks	86.0%	96.0%	88.2%	86.0%	89.2%
	SVMs	88.8%	95.2%	96.3%	91.5%	93.2%
	AdaBoost	96.2%	92.1%	98.5%	89.9%	94.2%
	3D kernel	91.6%	95.3%	99.3%	92.3%	95.0%