Chemical, material and pharmaceutical industries, focus on producing new and better products to satisfy the growing consumer demands. Molecular level investigation using old trial and error methodologies to produce new products, with specific desired properties, is capital intensive, highly time consuming and difficult to automate. Use computer-aided molecular design (CAMD), an emerging technology, can partially replace the same as one can generate and evaluate numerous candidates in a very short time by determining the molecular structures that match a specific set of desired (target) properties representing the product. Extending CAMD techniques for predicting macromolecular structure (like polymers), for a set of target properties, however, requires new property models. Developing new property models for predicting polymer properties based on the analysis of polymer structures is the main objective of this paper.

This work involves inductive learning. That is, experimental data are collected, a property model with predictive capability is derived from the data set and finally the properties are predicted using the developed model. Group contribution (GC) as well as atom-connectivity index(CI) based polymer property prediction models are developed in this work. The properties considered are those usually needed in polymer design, such as, amorphous volume, crystalline volume, glass transition temperature, solubility parameter, refractive index, permeability, thermal conductivity, dielectric constant and heat capacity of solid and liquid polymers.

Polymer property modeling requires the estimation of linear and non-linear parameters, and, solving those needs high computational resources and time. When modeling several properties, time taken for parameter estimation and computational power demands in general, increases proportionately with the number of properties modeled. Grid technology greatly facilitates this work as different property models (based on either group contributions, connectivity indices or other methods) can be generated in parallel. OFFICE Grid software offered by Mesh Technologies is used for this purpose, where one processor is used as a ‘master' and five others connected to it as ‘clients'. The job is submitted to the master from the command line, and the master looks for freely available clients (using the free computer resources) to perform the submitted job. The completed job is reported as the output by the master.

In this work, we present a suite of new polymer property models based on groups and atomic contributions using the Grid technology. These models are more suitable for computer-aided polymer design algorithms than their predecessors and have been developed simultaneously. That is, around 350-700 parameters for 20 property models (10 of GC type and 10 of CI type) have been developed in parallel. A comparison of the time required for generating group contribution and connectivity indices based models in series and in parallel, using grid technology will be presented together with details on the application of the Grid Technology.