pKALM: Accurate and Rapid Protein pK_a Prediction: Protein Language Models Reveal the Sequence-pK_a Relationship

pKALM is a web server for predicting protein pK_a values from amino acid sequences, based on protein language model and transfer learning. It achives high accuracy and speed compared to existing methods. It supports the pK_a prediction of eight ionizable groups: N-terminal, C-terminal, and six ionizable side chains of Asp, Glu, His, Lys, Cys, and Tyr, with 0.8321 RMSE and a speed of 4,961 pK_a / sec. [ GitHub | Paper | License ]

There is a variant of pKALM, called pKALMs, which can predict largely shifted pK_a values. pKALMs was trained on the simulated data rather than the experimental data. Users can choose either pKALM or pKALMs for the prediction.

Predict by pKALM Example Sequence Clear

Prediction Results

The prediction is quite fast and the results is a CSV-formatted table with 5 columns:

1. The first column is the sequence ID.
2. The second column is the index of the ionizable group.
3. The third column is the name of the ionizable group.
4. The fourth column is the predicted pK_a shift values.
5. The fifth column is the predicted pK_a values.

Significantly shifted pK_a values are highlighted in yellow. The predicted pI are also attached at the end of the table.

References

Users are kindly requested to utilize the following citation when referencing this method:

• Shijie Xu and Akira Onoda, Accurate and Rapid Prediction of Protein pK_a: Protein Language Models Reveal the Sequence-pK_a Relationship." bioRxiv (2024): 2024-09. DOI: 10.1101/2024.09.16.613101

Please contact shijie.xu@ees.hokudai.ac.jp for any questions.

Changelogs

• 2024/10/22: First release.