The ability to predict the properties of materials before they are fabricated, or to explore their behavior under conditions which are impractical to replicate under laboratory conditions (eg extreme temperatures or pressures) is critical for next generation materials design. Improving the predictive power of materials modelling guides planning and design, saves money, and can accelerate discovery.
Conventional ball and stick model of the sucrose molecule [Source]
Unlike the conventional “ball and stick” picture, the electrons within molecules are mobile; they fluctuate, redistribute themselves and are distorted by their complex environments. This behavior leads to a wide range of interactions, such as electron fluctuations (polarization) and charge distortion (dispersion) which influence how matter behaves. A good way of handling these effects, (incorporating quantum mechanical effects), without prohibitive computational expense, hasn’t yet been developed. As a result, these effects are normally neglected in conventional simulations of materials, and predictive power is limited as a result.
As an added problem, simulating molecules is even more difficult when dealing with outside influences, such as when the molecule is in a solution, encapsulated (such as when used to aid drug delivery), tagged with other molecules, or in contact with a cell surface. Coping with such a wide range of conditions presents fundamental challenges for current methodologies which tend to be designed to cope with a single, average, environment rather than adapt to changing circumstances.
IBM researchers, in conjunction with the Hartree Centre have developed a fundamentally new strategy for capturing these electronic effects in molecules, which can predict the weird properties of water – from the freezing point to the critical point (when water’s liquid and vapor states can coexist) and beyond. The approach – called “electronic coarse graining” – represents the electron cloud bound to a molecule as a single charged object tethered to the molecular frame by a spring. The simplicity of the “electronic coarse graining” approach captures the full spectrum of intermolecular forces, including those that arise from dispersion and polarization, which are typically not included in conventional models.
An example of the model created when using the new, “electronic corse graining” approach
A particularly challenging and important area for this type of research is in the life sciences and biotechnology space, where the development pipeline for new drugs and compounds is often long and costly – and molecular-level insight is priceless. Here, it is essential to understand the structure and simulations of elements at the molecular scale in a variety of complex and changing environments, as the molecular structure of drug design is critical to their effectiveness.
This work lays the foundation for much more predictive, and reliable materials modelling involving fewer assumptions, and taking into account a much more complete set of molecular forces. The method will be extended first in life sciences and simple biomolecules and eventually on to short proteins and self-assembling structures. All of these circumstances are relevant to the drug discovery cycle from early design concepts, understanding mode of action at the molecular scale, through to the development of a new medicine.
At the 18th European Conference on Computational Biology and the 27th Conference on Intelligent Systems for Molecular Biology, IBM will present significant, novel research that led to the implementation of three machine learning solutions aimed at accelerating and guiding cancer research.
Our team of IBM researchers published research in Radiology around a new AI model that can predict the development of malignant breast cancer in patients within the year, at rates comparable to human radiologists.