Protein pK_a Prediction by Tree-Based Machine Learning

Abstract

Protonation states of ionizable protein residues modulate many essential biological processes. For correct modeling and understanding of these processes, it is crucial to accurately determine their pK a values. Here, we present four tree-based machine learning models for protein pK a prediction. The four models, Random Forest, Extra Trees, eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), were trained on three experimental PDB and pK a datasets, two of which included a notable portion of internal residues. We observed similar performance among the four machine learning algorithms. The best model trained on the largest dataset performs 37% better than the widely used empirical pK a prediction tool PROPKA and 15% better than the published result from the pK a prediction method DelPhiPKa. The overall root-mean-square error (RMSE) for this model is 0.69, with surface and buried RMSE values being 0.56 and 0.78, respectively, considering six residue types (Asp, Glu, His, Lys, Cys, and Tyr), and 0.63 when considering Asp, Glu, His, and Lys only. We provide pK a predictions for proteins in human proteome from the AlphaFold Protein Structure Database and observed that 1% of Asp/Glu/Lys residues have highly shifted pK a values close to the physiological pH.

AlphaFold, Artificial Intelligence (AI), and Allostery

“…It also points to research projects that should be reconsidered. The richness of high quality data that are being compiled in databases (e.g., refs and − ) is already strengthening studies that require protein structures, such as mapping binding sites and interactions in signaling pathways, and identification of hot spots, including latent and rare cancer driver mutations. − The most profound impact will likely be in accelerating and improving production of new medications (e.g., ref ), and in generating data that can be used toward this vital aim (e.g., refs , , , and − ). AI developments and applications may further help foretell whether the signal propagating downstream will be strong enough to reach its genomic target to activate (suppress) gene expression, and predict pathways. − Altogether, these powerful approaches and the databases that they create revamp and transform traditional and ongoing research involving the use of structures.…”

A Fast and Interpretable Deep Learning Approach for Accurate Electrostatics-Driven pKa Predictions in Proteins

“…Both these issues contribute to the risk of model overfitting and poor generalizability. Chen et al trained tree-based machine learning models, such as XGBoost or LightGBM, on experimental data, and their best model exhibited an RMSE of 0.69 . To compare pKAI with these models and illustrate the data leakage problem at hand, we have refined our pKAI model by training it on same data split reported in ref .…”

“…As a comparison, in PROPKA3, only 85 experimental values of aspartate and glutamate residues were used to fit 6 parameters . Recently, traditional ML models have been trained on ∼1500 experimental p K a values. , However, testing the real-world performances of such methods is difficult, as there is a high degree of similarity among available experimental data. Our larger data set translates into more diversity in terms of protein and residue types and, more importantly, a wider variety of residue environments.…”

“…experimental pK a values 36,37 . However, testing the real world performance of such methods is difficult as there is a high degree of similarity among available experimental data.…”

AlphaFold 2 and NMR Spectroscopy: Partners to Understand Protein Structure, Dynamics and Function

“…Fortunately, a rather successful empirical method for estimating pKas is available called PropKa ( Li et al, 2005 ); it uses a protein 3D structure and takes into account factors like burial, which tends to favor the neutral state and the proximity of other charged groups to calculate approximate pKa values. Thanks to AlphaFold 2, the availability of accurate 3D structures has now enabled the complete calculation of all titratable residues in the whole human proteome ( Chen et al, 2022 ).…”