DeepConf: Leveraging ANI-ML Potentials for Exploring Local Minima with Application to Bioactive Conformations.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Drug Discovery

Background:

• Accurate prediction of molecular conformations is crucial for understanding biological activity.
• Existing methods for conformer generation can be computationally expensive and may not always achieve desired accuracy.

Purpose of the Study:

• To introduce DeepConf, a novel low-energy conformer generation algorithm.
• To evaluate the performance of ANI-ML potentials in reproducing bioactive conformations.
• To provide a guideline for evaluating bioactive conformation generation processes.

Main Methods:

• Development of the DeepConf algorithm for conformer generation.
• Utilizing ANI-ML potentials to achieve DFT accuracy.
• Benchmarking DeepConf against Auto3D using bioactive conformation datasets.
• Focusing on conformational changes due to single bond rotations.

Main Results:

• DeepConf efficiently generates high-quality conformers, especially for structures far from equilibrium or when ML potentials encounter non-smooth regions.
• ANI-ML potentials, when used with DeepConf, accurately reproduce bioactive conformations with a mean RMSD < 0.5 Å.
• The method outperforms conventional techniques in reproducing target conformations.

Conclusions:

• DeepConf offers an efficient and accurate approach for low-energy conformer generation.
• ANI-ML potentials show significant promise for drug discovery applications requiring precise conformational prediction.
• The developed algorithm and guidelines facilitate improved bioactive conformation evaluation.