Deep Autoencoder Learning Features for Classifying Peptides Data as Recombinant Bacterial Material Using PTVPSO-Deep ELM Algorithm

Abstract

Codon codes directly in the form of vector or 3D molecular to support feature learning often impede machine learning reading for the epitope classification process in extracting feature values that are more easily understood by computers.  The compute time required for molecular 3D is also quite significant, which is directed more to a good visualization art approach only because the DNA visualization method for determining epitope and non-epitope classes is more suitable for observation by human experts, not computers.  In addition, the length of each codon code in the peptides varies greatly.  Therefore, computers for such dataset processing require standardized visualization transformation results before the classification process is carried out. In this research, a 2-stage approach was carried out. First, the deep autoencoder approach was used as a pre-process to transform the learning features from multi-dimensional variations to 2D or 3D which is more powerful than PCA and kernel trick. Then, the classification stage was carried out using the PTVPSO-Deep ELM algorithm.  PSO is used to optimize the transformation and classification results. The test results showed that the proposed method is able to produce better and more stable metric values..