Kinome-wide interaction modelling using alignment-based and alignment-independent approaches for kinase description and linear and non-linear data analysis techniques

Table 1 Results of proteochemometric modelling of kinase-inhibitor interactions using different types of kinase descriptions and different data analysis methods

Data analysis method:	DT		1-NN		k-NN		SVM		PLS		PLS (w/o cross-terms)
Kinase description:	P ²	P ² _kin	P ²	P ² _kin	P ²	P ² _kin	P ²	P ² _kin	P ²	P ² _kin	P ²	P ² _kin
Composition, transition, and distribution (CTD) of amino acid properties	0.45	0.38	0.48	0.43	0.58	0.53	0.66	0.60	0.48	0.45	0.32	0.30
Sequence order and pseudo-amino acid (SO-PAA) descriptors	0.44	0.33	0.52	0.49	0.60	0.55	0.68	0.63	0.49	0.44	0.32	0.29
Amino acid and dipeptide composition (AAC-DC)	0.43	0.33	0.50	0.46	0.62	0.57	0.68	0.64	0.58	0.53	0.34	0.31
Maximums of auto- and cross-covariances (MACCs) of z-scales	0.46	0.30	0.55	0.55	0.63	0.63	0.70	0.67	0.66	0.63	0.35	0.32
Auto- and cross-covariances (ACCs) of z-scale descriptors	0.48	0.42	0.53	0.49	0.64	0.53	0.72	0.69	0.66	0.64	0.35	0.32
Z-scales of aligned sequences	0.49	0.43	0.55	0.58	0.65	0.64	0.73	0.70	0.67	0.65	0.34	0.32

Shown are the performances of proteochemometric models based on decision trees (DT), one nearest neighbour (1-NN) and k-nearest neighbour (k-NN) approaches, support vector machines (SVM), and partial least-square projections to latent structures, with (PLS) and without cross-terms (PLS w/o cross-terms). P² and P²_kin indicate the squared correlation coefficient from outer loop of cross-validation for, respectively, new kinase-inhibitor combinations and new kinases.

ISSN: 1471-2105