reverse engineering – recent advances and applications - OpenLibra

224
Reverse Engineering – Recent Advances and Applications
the activation level of a neuron into an output signal (regulation result). The induced local
field and the output of the neuron, respectively, are given by:
N
i j � ik k j�1 � i j
k�1
v ( T ) � w x ( T ) b ( T )
(1)
x ( T ) � �(
v ( T ))
(2)
i j i j
where the synaptic weights wi1, wi2,…, wiN define the strength of connections between the ith
neuron (e.g. ith gene) and its inputs (e.g. expression level of genes). Such synaptic weights
exist between all pairs of neurons in the network. bi( Tj) denotes the bias for the ith neuron
at time T j . We denote w � as a weight vector that consists of all the synaptic weights and
biases in the network. w � is adapted during the learning process to yield the desired network
outputs. The activation function �() introduces nonlinearity to the model. When information
about the complexity of the underlying system is available, a suitable activation function can
be chosen (e.g. linear, logistic, sigmoid, threshold, hyperbolic tangent sigmoid or Gaussian
function.) If no prior information is available, our algorithm uses the hyperbolic tangent
sigmoid function.
Fig. 3. Four NMs discovered in human: (A) multi-input module; (B) single input module; (C)
feed-forward loop - 1; and (D) feed-forward loop - 2.
As a cost function, we use the RMSE between the expected output and the network output
across time and neurons in the network. The cost function is written as: