ComputChem Products

iTitrate

Accurate knowledge of protonation states and reactive (nucleophilic) amino acid locations can significantly enhance the power of computer-aided drug design. Nucleophilicity is related to the pKavalue, but traditional approaches to estimate protein pKa's are often plagued by low accuracy and sensitivity to the input protein structure. iTitrate accurately predicts the pKa's of titratable residues by simulating an acid-base titration experiment, i.e., letting a protein adjust its protonation and conformational states to solution pH. The state-of-the-art approach employs the physics-based Amber force field (energy functions) and implicit solvent model, and has been rigorously validated for hundreds of proteins. iTitrate can be seamlessly integrated in a computer-aided drug design workflow or used as a standalone tool.

Key Features

Accuracy

Root mean square error of pKa < 1 pH unit for Asp, Glu, His, Cys, and Lys; accuracy of predicting reactive Cys > 80%

Web Interface

Seamless integration with a web GUI that lets you manage jobs, inputs, and data

Local or Cloud

Create and launch jobs locally or on AWS with the press of a button, no more waiting on slow machines

Data & Network Security

Keep your data private and secure with your local environment or with the highest standard by AWS

iKa (beta)

Absorption, distribution, metabolism, excretion, and toxicity of drug molecules are significantly affected by the ionization states. iKa accurately predicts small molecule pKa’s using machine learning models trained on large experimental data sets. Models have been validated for kinase inhibitor-like small molecules and GPCR ligands.

Key Features

Accuracy

Root mean square error of pKa < 1.5 pH units for drug-like small molecules; recognizes acid or base

SMILES Only

Only SMILES strings for molecules are needed for training, validating, and predicting

Local Operation

Local data storage and computation; security guaranteed by your own environment

Self-Trainable

Can either predict using trained machine learning models or can train new models given experimental data