TPdsm: a method based on TabPFN for prediction of deleterious synonymous mutations.

Motivation: Assessing the deleteriousness of synonymous mutations is of considerable importance for understanding human health, and the development of corresponding prediction methods offers a rapid and efficient approach. However, the scarcity of training data remains a major bottleneck in building high-performance predictors for synonymous mutation deleteriousness. Herein, we present TPdsm, a novel method for predicting deleterious synonymous mutations. Results: TPdsm leverages TabPFN, a tabular foundation model specifically designed for small-sample prediction. The features are retrieved from CDsyn, a comprehensive database dedicated to deleterious synonymous mutation prediction. Our evaluation demonstrates that TPdsm delivers better predictive performance than a set of 14 current state-of-the-art predictors, as evidenced by its results across multiple independent testing datasets and real-world cases. Availability and implementation: TPdsm is available on GitHub at https://github.com/Project4bz2023/TPdsm. The pre-computed score of TPdsm can be accessed at https://doi.org/10.5281/zenodo.18265619. The result can be queried at https://bingtseng-tpdsm.share.connect.posit.cloud/.