A Deep Learning Approach for Peptide-based Treatment Design

The great Human Genome Project revealed the heterogeneity of human society. Geneticvariants drive the healthcare process and treatment design toward individualization and per-persontreatment methods. Finding the causative agent and rational drug design, to increase or inhibit theactivity of the target protein is the first stage of treatment design. Side effects and drug resistance haveincreased the interest in exploiting natural compounds.The use of therapeutic peptides has been known as a therapeutic strategy for years due to their lowertoxicity and immunogenicity, type of physical and chemical structure, and specificity. However, theexpensive production process and the limitations of human drug testing, as well as the existence of bigdata, have caused computational methods based on machine learning and deep learning to be used toidentify the effective features of peptide therapy or to introduce new therapeutic peptides for synthesiswith generative algorithms.In the current research, while introducing a new data library consisting of ۱۶۲ pairs of therapeuticpeptides and target proteins of the diseases, which were collected from bioinformatics servers andarticles, feature extraction was done using the iFeature python toolkit [۱], and data classification wasconducted by Random Forest algorithm, which ۸۰% precision and ۷۸% accuracy were achieved. Next,according to the linguistic nature of peptide and protein sequences, another classification was modeledfrom the deep learning method based on the transformer self-attention mechanism and the embeddingvector resulting from the encoder layer in the ProtBERT [۲] architecture, which in comparison to themachine learning model, respectively It resulted in a precision and accuracy of ۷۱% and ۷۸%. Thisresearch shows that despite the limitations of the peptide data library, the classification of therapeuticpeptides with the transformer model can compete with the results of the machine learning model.

کلیدواژه ها:

Peptide ، Treatment Design ، Machine Learning ، Transformer