An explicit-solvent hybrid QM and MM approach for predicting pKa of small molecules in SAMPL6 challenge

Abstract

In this work we have developed a hybrid QM and MM approach to predict pKa of small drug-like molecules in explicit solvent. The gas phase free energy of deprotonation is calculated using the M06–2X density functional theory level with Pople basis sets. The solvation free energy difference of the acid and its conjugate base is calculated at MD level using thermodynamic integration. We applied this method to the 24 drug-like molecules in the SAMPL6 blind pKa prediction challenge. We achieved an overall RMSE of 2.4 pKa units in our prediction. Our results show that further optimization of the protocol needs to be done before this method can be used as an alternative approach to the well established approaches of a full quantum level or empirical pKa prediction methods.

Holistic Prediction of the pKa in Diverse Solvents Based on a Machine‐Learning Approach

“…In the year of 2017, a blind p K a prediction challenge named SAMPL6 has been established by the Drug Design Data Resource Community, which consists of predicting microscopic and macroscopic p K a of 24 drug‐like small molecules (17 drug‐fragment‐like and 7 drug‐like). The submitted prediction strategies included quantum‐chemistry based calculation, EC‐RISM theory, QM/MM approach, ab initio quantum mechanical prediction as well as machine learning . The machine learning methods were built with the general Gaussian process with unfortunately only moderate accuracy.…”

Evaluation of log P, pKa, and log D predictions from the SAMPL7 blind challenge

“…Previous SAMPL challenges have looked at the prediction of solvation free energies [8][9][10][11][12], guest-host [13][14][15][16][17][18][19] and protein-ligand binding affinities [20][21][22][23][24][25][26], pK a [27][28][29][30][31][32][33], distribution coefficients [34][35][36][37], and partition coefficients [38][39][40][41]. These challenges have helped uncover sources of error, pinpoint the reasons various methods performed poorly or well and their strengths and weaknesses, and facilitate (1) log P = log 10 K ow = log 10 [unionized solute] octanol [unionized solute] water dissemination of lessons learned after each challenge ends, ultimately leading to improved methods and algorithms.…”

“…A slightly different approach was used by one participant (0wfzo) where QM calculations of the free energy of deprotonation and thermodynamic integration, an MM method, were combined to calculate the difference of the solvation free energies between the acid and its conjugate base [33]. This approach yielded an average level of performance, with an RMSE of 2.89 for the macroscopic acidity constants calculated from the submitted microscopic acidity constants, excluding two compounds (SM14 and SM18) from the analysis as they exhibited multiple pK a values too close to each other.…”

Holistic Prediction of the pKa in Diverse Solvents Based on a Machine‐Learning Approach

“…In the year of 2017, ab lind pK a prediction challenge named SAMPL6 has been established by the Drug Design Data Resource Community, [18] which consists of predicting microscopic and macroscopic pK a of 24 drug-like small molecules (17 drug-fragment-like and 7drug-like). The submitted prediction strategies included quantum-chemistry based calculation, [19] EC-RISM theory, [20] QM/MM approach, [21] ab initio quantum mechanical prediction [22] as well as machine learning. [23] Them achine learning methods were built with the general Gaussian process with unfortunately only moderate accuracy.B ased on the HM-XGBoost model, we obtained aprediction with MAE = 0.80 and RMSE = 1.07 (Figure 9and Figure S17 A).…”