Abstract:Despite breakthroughs achieved in protein sequence-to-structure and function-to-sequence predictions, the affinity-to-mutation prediction problem remains unsolved. Such a problem is of exponential complexity deemed to find a mutated protein or protein complex having a guaranteed binding-affinity change. Here we introduce an adversarial learning-based mutation method that creates optimal amino acid substitutions and changes the mutant's affinity change significantly in a preset direction. The key aspect in our method is the adversarial training process that dynamically labels the real side of the protein data and generates fake pseudo-data accordingly to construct a deep learning architecture for guiding the mutation. The method is sufficiently flexible to generate both single- and multipointed mutations at the adversarial learning step to mimic the natural circumstances of protein evolution. Compared with random mutants, our mutated sequences have in silico exhibited more than one order of change in magnitude of binding free energy change towards stronger complexes in the case study of Novavax–angiotensin-converting enzyme-related carboxypeptidase vaccine construct optimization. We also applied the method iteratively each time, using the output as the input sequence of the next iteration, to generate paths and a landscape of mutants with affinity-increasing monotonicity to understand SARS-CoV-2 Omicron's spike evolution. With these steps taken for effective generation of protein mutants of monotone affinity, our method will provide potential benefits to many other applications including protein bioengineering, drug design, antibody reformulation and therapeutic protein medication.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the problem of predicting binding affinity after protein mutation. Specifically, the paper proposes an adversarial learning-based method to generate protein mutants with binding affinity changes in a specific direction (enhanced or weakened). While there have been breakthroughs in predicting protein sequence to structure and function to sequence, the prediction of binding affinity changes due to mutations remains unresolved. The complexity of this problem grows exponentially as it requires finding a mutated protein or protein complex that ensures the expected change in binding affinity. ### Main Contributions 1. **Adversarial Learning Framework**: The paper introduces a mutation method based on adversarial learning that can create optimal amino acid substitutions and significantly alter the binding affinity of the mutants. 2. **Flexibility**: This method can generate single-point or multi-point mutations, simulating the process of natural protein evolution. 3. **Iterative Application**: By iteratively applying this method, it is possible to generate mutation paths with monotonically increasing binding affinity, thereby understanding the evolutionary path of viral spike proteins (e.g., SARS-CoV-2 Omicron). 4. **Practical Application**: This method performs excellently in optimizing the binding affinity between the Novavax vaccine and the ACE2 receptor, with the mutated sequences showing an order of magnitude higher change in binding free energy compared to random mutation sequences. ### Method Overview - **DeepDirect Framework**: Includes a mutation generator, two discriminators, and a binding affinity change predictor. The mutation generator selects mutation sites and amino acid substitutions based on the structural information of the input protein, generating mutants with specific directional changes. - **Adversarial Training**: Constructs a deep learning architecture to guide mutation generation by dynamically labeling real data and generating pseudo-data. - **Iterative Generation**: By iteratively applying the mutation generator multiple times, it generates mutation paths with monotonically increasing binding affinity. ### Experimental Results - **Novavax Vaccine Optimization**: Mutants generated by DeepDirect show an order of magnitude higher change in binding free energy compared to random mutants, significantly enhancing the binding affinity between the vaccine and the ACE2 receptor. - **SARS-CoV-2 Omicron Spike Protein Evolution Path**: By iteratively applying DeepDirect, mutation paths with monotonically increasing binding affinity were generated, revealing the potential evolutionary direction of the virus. ### Conclusion DeepDirect provides an efficient method for generating protein mutants with specific directional changes in binding affinity. This method has broad application potential in protein bioengineering, drug design, antibody recombination, and therapeutic protein drugs.

Generating mutants of monotone affinity towards stronger protein complexes through adversarial learning

AF2-Mutation: Adversarial Sequence Mutations against AlphaFold2 on Protein Tertiary Structure Prediction

DDAffinity: predicting the changes in binding affinity of multiple point mutations using protein 3D structure

Retrieval-Enhanced Mutation Mastery: Augmenting Zero-Shot Prediction of Protein Language Model

Predicting the immune escape of SARS-CoV-2 neutralizing antibodies upon mutation

Zero-shot prediction of mutation effects with multimodal deep representation learning guides protein engineering

MuToN Quantifies Binding Affinity Changes upon Protein Mutations by Geometric Deep Learning

Exploring Pathogenic Mutation in Allosteric Proteins: the Prediction and Beyond

MutaGAN: A Seq2seq GAN Framework to Predict Mutations of Evolving Protein Populations

Multi-level Interaction Modeling for Protein Mutational Effect Prediction

Predicting a Protein's Stability under a Million Mutations

Multi-level Protein Representation Learning for Blind Mutational Effect Prediction

Accelerating protein engineering with fitness landscape modeling and reinforcement learning

Semantical and Geometrical Protein Encoding Toward Enhanced Bioactivity and Thermostability

Efficiently Predicting Mutational Effect on Homologous Proteins by Evolution Encoding

Protein Engineering with Lightweight Graph Denoising Neural Networks

PANDA: Predicting the change in proteins binding affinity upon mutations using sequence information

Predicting protein thermal stability changes upon single and multi-point mutations via restricted attention subgraph neural network

A machine learning approach for reliable prediction of amino acid interactions and its application in the directed evolution of enantioselective enzymes

AttABseq: an Attention-Based Deep Learning Prediction Method for Antigen-Antibody Binding Affinity Changes Based on Protein Sequences.

wwLearning the language of proteins and predicting the impact of mutations