GAN-GA: A Deep-Evolutionary Approach for Synthesizing Functionally Similar Heat Shock Proteins.

Area of Science:

• Protein engineering
• Computational biology
• Synthetic biology

Background:

• De novo protein design enables functional protein generation without natural templates.
• Generative deep learning and evolutionary computation advance synthetic protein engineering.
• Traditional methods often lack functional and physicochemical validation.

Purpose of the Study:

• To propose a hybrid framework (GAN-GA) for de novo protein design.
• To address limitations in sequence-based generation and improve biological relevance.
• To design synthetic Heat Shock Proteins (HSPs) with enhanced properties.

Main Methods:

• Utilized Wasserstein Generative Adversarial Networks (GANs) for novel sequence generation.
• Incorporated Non-dominated Sorting Genetic Algorithm II (NSGA-II) for multi-objective physicochemical optimization.
• Developed dual-objective (GAN-GA2) and six-objective (GAN-GA6) configurations.

Main Results:

• Generated novel protein sequences with improved functional and physicochemical validation.
• Optimized sequences demonstrated structural soundness and biological plausibility.
• GAN-GA outperformed other state-of-the-art algorithms like GAN, ProtBert, ESM, VAE, and RL.

Conclusions:

• GAN-GA framework offers a promising strategy for de novo protein engineering.
• The hybrid approach enhances the biological relevance of designed proteins.
• Applicable for therapeutic and synthetic biology applications.