Gaining insights into the physicochemical properties and sequence space of blood–brain barrier penetrating peptides

Abstract

The blood–brain barrier (BBB) poses a significant obstacle to the administration of drugs to the brain for the development of therapies of central nervous system (CNS) disorders. Blood-brain barrier penetrating peptides (BBBPps) are a group of peptides that can traverse the BBB by different processes without causing harm to the BBB. These peptides show promise as potential drugs for CNS ailments. Nevertheless, the process of identifying BBBPps using experimental approaches is both time-consuming and labour-intensive. To discover additional BBBPps as potential treatments for CNS diseases, it is critical to develop insilico methods that can distinguish BBBPps from non- BBBPps rapidly and precisely. In the current work, machine learning aided models are developed for accurate prediction of BBBPps using physicochemical and evolutionary information. The best model achieved an accuracy of 84.8 % by using 5-fold cross-validation and 79.8 % based on the holdout testing set on a reduced set of features. Further an extensive analysis is carried out for the black box models using model agnostic interpretation approaches to infer the physicochemical and sequence space of BBBPps. Basic amino acids and conservation of Arginine and Lysine are found to be more favoured in BBBPps. Further, basic amino acid property group and its interactions with other features are found to be prominent important interactions.

Keywords

• Blood brain barrier penetrating peptides
• Sequence properties
• Physicochemical space
• Gradient boosting machines
• XGboost
• Model agnostic interpretation methods