Fig. 1

Simplified diagram of the neural network of the Multi-resBind model. The input data with a length of 150 and width of d is first fed to a 1D convolutional layer. In this layer, the number of kernel filters, kernel size, and step size of the stride are set to 96, \(11 \times d\), and 1, respectively. The obtained feature maps are provided as input to a residual block (3 ×) with skip connection in the last block. Each block consists of three sequential layers: convolution 1D, batch normalization, and ReLU activation. After the residual block, the average pooling with 10 × 1 receptive fields (size = 10, stride = 10) and concatenate operations convert the feature maps to a 1D vector. The 1D vector then passes through a 3-layer fully connected network with 256 nodes in the hidden layer. The last fully connected layer consists of k nodes corresponding to the RBPs of interest independently. Finally, sigmoid is chosen as the activation function of each node in the last layer