Multi-level learning: improving the prediction of protein, domain and residue interactions by allowing information flow between levels

Table 2 Data features at the domain level.

Feature	Feature of	Data type	Kernel
Phylogenetic tree correlations [68] of Pfam alignments	Domain family pairs	Real matrix	Empirical kernel map [70]
In all species, number of proteins containing an in stance of the domain family	Domain families	Integers	Polynomial (d = 3)
In all species, number of proteins containing domain instances only from the family	Domain families	Integers	Polynomial (d = 3)
Number of domain instances of parent protein	Domain instances	Integers	Polynomial (d = 3)
Fraction of non-yeast interacting protein pairs contain ing instances of the two domains respectively are mediated by the domain instances*	Domain family pairs	Real matrix	Constant shift embedding [71]
Fraction of protein pairs containing instances of the two domains respectively are known to be interacting in the PPI training set*	Domain family pairs	Real matrix	Constant shift embedding

*: These two features were used with the unidirectional and bidirectional flow architectures only since they involve information about the training set of the protein level.

ISSN: 1471-2105