A simple, automated quasi-4D-QSAR, quasi-multi way PLS approach to develop highly predictive QSAR models for highly flexible CXCR4 inhibitor cyclic pentapeptide ligands using scripted common molecular modeling tools

A methodology for developing highly predictive (r(2) > 0.9) 3D-QSAR models (q(2) > 0.7) based on sixteen flexible CXCR4 cyclic pentapeptide inhibitors is reported. The effective automated use of common molecular modeling tools such as Macromodel and Sybyl is demonstrated. The recently developed multi-way Partial Least Square (PLS) approach for discovering the bioactive conformers and alignment was used in a quasi-multi-way PLS approach. Twenty-five conformers for each compound were generated by Monte Carlo conformational searches and alignments (seventy five in total) were based on the templates from the three most active compound conformers. These were aligned in Sybyl Molecular Databases and Sybyl Molecular Spreadsheets. All repetitive tasks were automated by use of simple Unix shell, python and Sybyl Programming Language (SPL) scripts. This efficient protocol furnished three 3D-QSAR models with q(2) values of 0.714, 0.734 and 0.657 and predictive r(2) values of 0.951, 0.990, and 0.956 respectively. The best 3D-QSAR model predicted the biological activities of nine test compounds from all activity ranges within 0.5 log units.