Machine learning-driven QSAR modeling combined with molecular dynamics suggests high-affinity CD33-targeting peptides for leukemia therapy

Abstract

Despite advances in antibody-based therapies for leukemia, significant limitations persist, including immunogenicity, toxicity, and resistance development. To address these challenges, we pursued an innovative peptide-based targeting strategy against CD33, a well-validated surface marker in leukemia. This study establishes a comprehensive pipeline integrating computational design with experimental validation to develop novel CD33-targeting peptides with optimal therapeutic properties.

Our methodology combined machine learning-based QSAR modeling (achieving R² = 0.93) with molecular docking and 100-ns molecular dynamics simulations to predict and validate peptide-CD33 interactions. From this pipeline, we identified two lead peptides A3K2L2 (AKAKLAL-NH₂) and K4I3 (KKKKIII-NH₂), demonstrating exceptional binding affinities (−146.11 and − 108.08 kcal/mol, respectively) and complex stability (RMSD 0.25–0.35 nm).

Experimental characterization revealed these peptides form stable reverse β-sheet structures and self-assembling nanostructures while maintaining remarkable selectivity, showing potent cytotoxicity against K-562 cells (IC₅₀ 60–90 μM) with minimal effects on normal PBMCs. Additional safety profiling confirmed low hemolytic activity (<5 %) and favorable induction of cancer cell death through combined apoptotic and necrotic pathways.

These findings represent a significant advancement in targeted leukemia therapy, offering peptide-based alternatives that overcome key limitations of current antibody treatments. Our integrated design platform provides a generalizable framework for developing receptor-specific anticancer peptides, with particular promise for hematological malignancies where target specificity is paramount.