Combining Deep Neural Networks for Protein Secondary Structure Prediction

Abstract

By combining convolutional neural networks (CNN) and long short term memory networks (LSTM) into the learning structure, this paper presents a supervised learning method called combining deep neural networks (CDNN) for protein secondary structure prediction. First, we use multiple convolutional neural networks with different number of layers and different size of filters to extract the protein secondary structure features. Second, we use bidirectional LSTM to extract features continually based on the raw features and the features which extracted by CNNs. Third, a fully connected dense layer is used to map the extracted features by LSTM to the different protein secondary structure classes. CDNN architecture is trained by RMSProp optimizer based on the cross entropy error between protein secondary structure labels and dense layer's outputs. CDNN not only inherits the abstraction ability of CNN and sequence data process ability of LSTM, but also demonstrates the attractive classification ability for handling protein secondary structure data. The empirical validation on two protein secondary structure prediction datasets demonstrates the effectiveness of CDNN method.

Expression, purification, and biological characterization of recombinant human interleukin‐31 protein

“…The parameters used for the CD scan were set up as: wavelength: 190.0–260.0 nm, step resolution: 1.0 bandwidth, 1.0 nm, temperature: 25.0°C, and sensitivity: 0.5 s time‐per‐point. The secondary structural elements of hIL‐31 were analyzed by submitting the raw data to software cdnn using the CDNN algorithm with a reference set 20 …”

Geometric Deep Learning on Molecular Representations

“…Generative molecular design [19,52,53], synthesis planning [54], protein structure prediction [55,56] and prediction of properties in drug discovery [57,58] Transformers String notation (encoded as a graph)…”

“…a few studies, one-letter amino acid sequences were employed for peptide design [52,[143][144][145][146]. RNNs have also been applied to predict ligand-protein interactions and the pharmacokinetic properties of drugs [57,58], protein secondary structure [55,56], and the temporal evolution of molecular trajectories [147]. RNNs have been applied for molecular feature extraction [148,149], showing that the learned features outperformed both traditional molecular descriptors and graph-convolution methods for virtual screening and property prediction [148].…”

Advances in Optimisation Algorithms and Techniques for Deep Learning

“…An advantage provided by RMSProp is that it requires less tuning compared to the SGD [36]. Besides, the RMSProp has been used in different DL architectures including to optimise the parameters of a combined deep CNN and LSTM for classifying protein structures [42] as well as the design of a new deep convolutional spiking neural network for time series classification [43].…”

“…where A and C are the coefficient vector, t represents the t − th iteration, X is the wolf vector position and X P is the prey vector position. The vectors A and C is obtained by (42) where r 1 and r 2 are random vectors located in the slope of [0, 1] and the value of a lies between 0and2. The GWO has been recently applied in optimising flight models, especially to identify the flight state using CNNs [58] as well as modifying the hidden parameters of the SAE architecture [56].…”