SoluProtMut: Siamese Deep Learning for Predicting Solubility Effects of Protein Mutations with Experimental Validation

SoluProtMut, a Siamese deep geometric neural network trained on deep mutational scanning in bacterial and yeast expression assays, achieves top informedness on 44 proteins, with a 26.5% ternary informedness for multipoint mutations.

• Why it matters: Predicting how mutations change protein solubility is difficult because solubility datasets are heterogeneous and biased toward certain proteins, and limited data continues to slow progress.
• What they did: The final model was trained only on deep mutational scanning from recombinant protein expression assays, then compared against five established solubility predictors on an independent set of 44 proteins.
• The result: External validation using new solubility measurements for 277 haloalkane dehalogenase mutants and recent myoglobin deep-mutational-scanning data suggests protein-level generalization limits performance, while high-throughput assays plus deep learning can expand accuracy and scope.

AI summary of the abstract · may contain errors · not medical advice · read the paper.