Peptide Hemolytic Activity Analysis using Visual Data Mining of Similarity-based Complex Networks

Abstract

Peptides are promising drug development frameworks thanks to their high target selectivity, tolerability and relatively low production cost. However, despite the fact that several thousand potentially therapeutic peptides reported, only sixty have arrived at the market. This concerning low proportion is partially explained by undesired properties such as peptide-induced hemolytic activity. Hence, we aim to get a better insight into the chemical space of hemolytic peptides using a novel approach based on network science and interactive data mining as an alternative to design more effective peptide drugs with low hemolytic activity. Metadata networks (METNs) were used to characterize and find general patterns associated to hemolytic peptides, whereas Half-Space Proximal Networks (HSPNs), created using five different two-way dissimilarity measures, represented the hemolytic peptide space. Then, using the best candidate HSPNs, we extracted various scaffolds that capture information of almost all the chemical space but avoiding peptide overrepresentation. Such scaffolds can have many applications, such as training accurate ML-based prediction models, constructing one-class multi-query similarity searching models and characterizing the diversity of hemolytic peptides using a manageable set of peptides. Finally, by means of an alignment-free approach, we reported 47 putative hemolytic motifs, which might provide hints about the mechanisms of hemolysis and can also be used as toxic signatures when developing novel peptide-based drugs.

Multiquery Similarity Searching Models: An Alternative Approach for Predicting Hemolytic Activity from Peptide Sequence

“…It is a graph-based database containing 45120 peptides with annotated activities retrieved from 40 bioactive databases embedded in the StarPep toolbox software . A subset consisting of 2004 hemolytic peptides was used in ref for generating scaffolds. Here, we constructed our MQSS models based on the 144 scaffolds derived from HSPNs.…”

“…These networks were built using three similarity metrics: angular separation (AS), Chebyshev (Ch) distance, and Euclidean (Eu) distance. See SM4.2 in ref for more details. This database was used to characterize the prevalence of hemolytic toxicity in antibacterial, anticancer, antifungal, antiparasitic, and antiviral peptides.…”

“…In this data set, peptides containing ‘X’ several times in a sequence and Nphe or Nleu in their sequences were discarded. The resulting Big-Hemo data set contains 2196 highly hemolytic peptides …”

“…In a previous report, we explored the chemical space of the StarPep DB hemolytic peptides and generated a variety of subsets (scaffolds) that represent the entire space but retain a relatively low number of peptides, avoiding overrepresentation. Such scaffolds were derived from half-space proximal networks (HSPNs) constructed by using five different similarity measures.…”

“…Our method is based on complex network science and multiquery similarity searching (MQSS) models. This procedure has demonstrated not only to accurately predict tumor-homing and antiparasitic peptide activities but also to get a deeper insight into the peptide chemical space, , hence improving peptide-based drug design. Furthermore, a software specifically designed for this purpose, named StarPep toolbox v0.8.5 has been developed to facilitate the workflow …”