Annealed fractional Lévy-Itō diffusion models for protein generation.

Area of Science:

• Computational biology and bioinformatics
• Artificial intelligence in drug discovery
• Protein structure modeling

Background:

• Protein generation is crucial for therapeutic antibodies and drug development.
• Current generative models struggle with the high dimensionality and complexity of protein conformational space.
• Efficiently sampling protein structures remains a significant challenge.

Purpose of the Study:

• To develop novel generative diffusion models for improved protein structure sampling.
• To address limitations in current models for protein generation tasks.
• To enhance the exploration of high-dimensional conformational spaces in proteins.

Main Methods:

• Introduction of Markovian and non-Markovian generative diffusion models.
• Utilizing fractional stochastic differential equations and the Lévy distribution.
• Application to a dataset of proteins and evaluation using Fréchet distance, fidelity, and diversity.

Main Results:

• The novel diffusion models demonstrate more effective exploration of protein conformational space.
• Significant improvements observed in Fréchet distance (25.4%), fidelity (35.8%), and diversity (11.8%) compared to state-of-the-art methods.
• The proposed approach outperforms existing techniques in protein generation benchmarks.

Conclusions:

• The developed Markovian and non-Markovian diffusion models offer a superior approach to protein generation.
• This advancement has significant implications for accelerating the design of therapeutic antibodies and novel drugs.
• The models provide a more effective means to navigate and sample complex protein conformational landscapes.