SMMPPI

SMMPPI predicts small-molecule modulators of protein-protein interactions using machine learning to identify candidates relevant to anticancer, antiviral, and antimicrobial therapies.

Key Features:

• Two-Stage Prediction Process: Uses a two-stage approach with a random forest (RF) classifier to predict general PPI modulators and family-specific predictors to identify modulators for 11 PPI families.
• Training Data: Trained on a large dataset of experimentally validated PPI modulators to develop machine learning classifiers.
• High Predictive Accuracy: Demonstrates ROC-AUC > 0.9 and prediction accuracies exceeding 90% in most cases, with protocols to mitigate biases in training/test data division.
• Clinically Relevant PPI Families: Targets RBD:hACE2, Bromodomain_Histone, BCL2-Like_BAX/BAK, LEDGF_IN, LFA_ICAM, MDM2-Like_P53, RAS_SOS1, XIAP_Smac, WDR5_MLL1, KEAP1_NRF2, and CD4_gp120.

Scientific Applications:

• Novel Chemical Scaffold Identification: Identification of novel chemical scaffolds that inhibit the RBD_hACE PPI implicated in SARS-CoV-2 host cell entry.
• Docking Studies: Facilitates docking studies showing some predicted compounds bind with high affinity at key receptor-binding domain (RBD) interaction hotspots to inhibit the RBD_hACE2 interaction.

Methodology:

Applies machine learning techniques including a random forest classifier for general modulator prediction, family-specific predictors for 11 PPI families, and data-division protocols to reduce bias between training and test sets.