ContactGAN

ContactGAN refines predicted protein residue-residue contact maps using a generative adversarial network (GAN) to improve accuracy of tertiary structure prediction.

Key Features:

• Generative Adversarial Network integration: Uses a GAN framework to process input contact maps and distinguish true from false residue-residue contacts.
• Improved contact precision: Reports increased precision of contact predictions on benchmark datasets, including CASP13 and CASP14 targets.
• Enhanced tertiary model accuracy: Refined contact maps translate into more accurate protein tertiary structure models compared with the inputs.
• Comparison to trRosetta: Demonstrates a measurable, albeit relatively small, improvement over trRosetta in the reported comparisons.

Scientific Applications:

• Protein structure prediction: Provides refined contact maps to improve downstream tertiary structure modeling accuracy.
• Benchmarking and validation: Applicable for performance evaluation using established datasets such as CASP13 and CASP14.

Methodology:

ContactGAN takes a predicted protein contact map as input and processes it through a GAN framework to output an enhanced contact map that better captures true residue-residue contacts, focusing on refinement of existing predictions rather than de novo generation.