Quantum-level machine learning calculations of Levodopa.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Drug molecules possess functional groups leading to torsional barriers crucial for molecular simulations.
• Accurate potential energy surface (PES) calculations are vital in medicinal chemistry and drug design.
• Machine learning (ML), particularly deep learning (DL), is a rapidly advancing tool in computer-aided drug discovery.

Purpose of the Study:

• To utilize the ANI-1x neural network potential for predicting the PES of Levodopa, an antiparkinsonian drug.
• To compare ML predictions with density functional theory (DFT) calculations for accuracy and efficiency.

Main Methods:

• Employed the ANI-1x neural network potential, a quantum-level ML model.
• Performed DFT calculations using the wB97X method with various Pople's basis sets.
• Investigated vibrational frequencies to correlate DFT and ML data.

Main Results:

• The wB97X functional with the 6-31G(d) basis set showed results comparable to the ANI-1x model.
• A linear correlation was observed between DFT and ML data for vibrational frequencies.
• ANI-1x calculations were completed rapidly, demonstrating high computational efficiency.

Conclusions:

• The ANI-1x model provides an efficient and effective method for predicting PES in drug molecules.
• The findings suggest the ANI-1x dataset is valuable for computational structure-based drug design.
• This ML approach accelerates molecular simulations and aids in the discovery of new therapeutics.