Data-driven discovery of cardiolipin-selective small molecules by computational active learning.

Area of Science:

• Biophysics
• Computational Chemistry
• Materials Science

Background:

• Mitochondrial membrane lipid composition critically influences cellular function.
• Cardiolipin, a unique phospholipid in the inner mitochondrial membrane, serves as a biomarker for various pathologies.
• Targeted visualization and quantification of cardiolipin are essential for understanding and diagnosing diseases.

Purpose of the Study:

• To discover novel small organic compounds with high selectivity for cardiolipin-containing membranes.
• To develop a data-driven computational workflow for efficient molecular design.
• To establish design principles for cardiolipin-selective molecules.

Main Methods:

• A hybrid approach combining deep learning-enabled active learning with coarse-grained molecular dynamics (CG MD) simulations.
• Alchemical free energy calculations to assess binding affinities.
• Utilizing transferable CG models to explore a vast chemical space of small molecules (<500 Da).

Main Results:

• Identification of novel small molecules exhibiting significantly enhanced cardiolipin selectivity compared to the established probe 10-N-nonyl acridine orange.
• Efficient exploration of the molecular design space, simulating only 0.42% of the CG search space.
• Derivation of interpretable structure-selectivity relationships from simulation data.

Conclusions:

• The study demonstrates the power of multiscale modeling and coarse-grained simulations for accelerated materials discovery.
• The identified compounds and design rules offer potential for developing advanced diagnostic tools and therapeutics targeting mitochondrial dysfunction.
• This work highlights the utility of computational approaches in designing functional molecules for specific biological membranes.