Computational Prediction of Thermostable Fusion Proteins 9 views

Protein Engineering, the act of designing novel and unnatural proteins, has been used in many facets of sciences. Chimeric proteins that are the combination of two of more different proteins have been used in medicine to create more efficacious vaccines. We wanted to use computational methods in order to engineer chimeric proteins for vaccination in a simple and high-throughput fashion. We used Alphafold2’s (AF2) confidence score to test our hypothesis that this AF2 metric could identify homologous protein sequence that could be recombined with each other to create thermostable chimeras. These chimeras would be graded by comparing the change in confidence score between their native and chimeric states. We termed this change relative stability (RS) and found it to be a moderate predictor of stability with an accuracy of 5/7 SARS-CoV-2 spike chimeras that were taken through experimental validation. Once two recombination partners are selected, there are near infinite number of chimeric splices to make between them. We hypothesized that the language context in embeddings created by protein Language Models (pLMs) could help indicate which of these splice sites would lead to stable chimeras. We compared these contextualized numerical representations of different chimeric sequences and compared them to the context of their parents with dot product. We labeled this comparison Embedding Similarity (ES) and sequences with higher ES values are predicted to be more stable. We found our ES metric had a correlation of 0.79 when compared expression values from a SCHEMA-RASPP chimeric library and have now we have moved to experimental validation using expression and circular dichroism.

PHD (Doctor of Philosophy)

2025-11-10