XCPP: A Multi-model Explainable Deep Learning Framework for Accurate Identification of Cell-Penetrating Peptides from...

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Delivery Systems

Background:

• Cell Penetrating Peptides (CPPs) are short amino acid sequences enabling therapeutic molecule transport across cell membranes.
• CPPs offer a versatile platform for targeted drug delivery and molecular diagnostics.

Purpose of the Study:

• To develop and evaluate deep learning models for accurate in silico prediction of CPPs.
• To identify key sequence features contributing to CPP activity using explainable AI (XAI).

Main Methods:

• Analysis of 473 confirmed CPPs from the EnDM-CPP database.
• Computation of four sequence descriptors (PRIM, RPRIM, AAPIV, Reverse AAPIV).
• Training and testing of Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) models.
• Model evaluation using self-consistency, independent testing, and 10-fold cross-validation.
• Application of SHAP values for XAI to interpret model predictions.

Main Results:

• The CNN model achieved the highest accuracy (99.05%) during cross-validation, outperforming DNN and LSTM models.
• All models demonstrated reasonable prediction accuracy with structured input features.
• SHAP analysis successfully identified biologically relevant sequence descriptors, enhancing model transparency.

Conclusions:

• Deep learning, particularly CNNs, provides an effective framework for accurate CPP identification.
• The study highlights the potential of in silico CPP prediction for applications in drug delivery, diagnostics, and personalized medicine.
• SHAP-based XAI increases confidence in model predictions by linking sequence features to biological properties.