BMC Systems Biology

Table 3 The hyper-parameters of the proposed deep architecture

From: Large-scale prediction of protein ubiquitination sites using a multimodal deep architecture

Subnet	Layer	Hyper-parameters
Subnet	Layer	Activation function	Size^c	Filters	Drop-out
One hot vector	1D Convolution	softsign	2	200	0.4
		softsign	3	150	0.4
		softsign	5	150	0.4
		softsign	7	100	0.4
	Dense^a	relu	256	–	0.3
		relu	128	–	0
		relu	128	–	–
Phsico- chemical properties	Dense	softplus	1024	–	0.2
		softplus	512	–	0.4
		softplus	256	–	0.5
		relu	128	–	–
PSSM profile	1D Convolution	relu	1	200	0.5
		relu	8	150	0.5
		relu	9	200	0.5
	1D Convolution^b	relu	1	200	0.5
		relu	3	150	0.5
		relu	7	200	0.5
	Dense	relu	128	–	0.3
	Dense	relu	128	–	0
Merged representations	Dense	softmax	2	–	0

^aDense layers represent for the fully connected layers in keras
^bThe layers were designed for trans-positioned PSSM profile
^cThe size of convolution layers means the kernel sizes, and the size of Dense layers denotes the number of hidden states

Back to article page

ISSN: 1752-0509

Contact us

General enquiries: ORSupport@springernature.com