Designing libraries with CNS activity

Area of Science:

• Computational Chemistry
• Drug Discovery
• Medicinal Chemistry

Background:

• Designing libraries of Central Nervous System (CNS)-active compounds is a challenging but crucial task in drug discovery.
• Existing databases like CMC and MDDR contain numerous compounds with documented CNS activity and inactivity.

Purpose of the Study:

• To develop and validate a computational model for predicting CNS activity to aid in library design.
• To create a filtering system for identifying potentially CNS-active molecules within large chemical libraries.

Main Methods:

• Selection of CNS-active and inactive compounds from CMC and MDDR databases.
• Description of molecules using 7 1D and 166 2D molecular descriptors.
• Training a neural network using Bayesian methods to predict CNS activity based on molecular descriptors.

Main Results:

• The model achieved prediction accuracies of 75% for actives and 65% for inactives using 1D descriptors.
• Incorporating 2D descriptors improved prediction accuracy to 83% for actives and 79% for inactives.
• On a literature-curated database, the model achieved 92% accuracy for actives and 71% for inactives.

Conclusions:

• The developed neural network models serve as an effective filter for evaluating potential CNS-active molecules in chemical libraries.
• The method was successfully applied to generate a library of potentially CNS-active molecules suitable for combinatorial chemistry.