MRS Spring 2023, San Francisco, California

April 10, 2023 to April 14, 2023

Schrödinger is excited to be participating in the MRS Spring 2023 conference taking place on April 10th-14th in San Francisco, California. Join us for presentations by Schrödinger scientists.

Talk 1: Application of machine-learned potentials to accurately model polymer materials
 Mohammad Atif Faiz Afzal, Principal Scientist
Time: April 12 | 11:15am-11:30am PDT
Polymers are widely and increasingly used in a variety of industries, for example, in aerospace, electronics, and automotive. This popularity is due to polymer materials being easy to process and able to make the devices lighter and flexible. Furthermore, we can obtain desirable mechanical and thermophysical properties for specific applications by changing and controlling polymer chemistry. However, the scope of chemical space exploration and testing in an experimental setup is limited. Modeling and simulation of polymer systems provide an accelerated means of characterization and design of new polymer materials. Designing new polymers requires a careful understanding of the interactions between polymers and also with other components in a formulated material. Polymer systems are typically modeled using molecular dynamics (MD) techniques, but using classical force fields in MD has several shortcomings, especially for properties dependent on the polymer chain dynamics. We built scalable and generalizable machine-learned (ML) potentials that accurately capture the dynamics of polymer chains. Using the ML potentials, we can precisely calculate the polymer properties, including dynamical, thermophysical, and mechanical properties. In this presentation, I will present several case studies demonstrating the key benefits of using ML potentials in modeling polymeric materials.


Talk 2: Predicting Small-Molecule Viscosities Using Machine Learning and Physics-Informed Approaches 
 Alex Chew, Senior Scientist
Time: April 14 | 9:15am-9:30am PDT
Viscosity is a fundamental property that measures how “sticky” - or how resistant to flow - a fluid is, which dictates the performance of a wide range of materials, such as batteries, cosmetic formulations, and pharmaceuticals. Measuring viscosities using experiments or physics-based simulations is slow and tedious, which motivates the use of alternative, fast data-driven approaches to accelerate viscosity estimations. In this work, an extensive collection of over 4,000 experimental viscosity and temperature values for small organic molecules were curated from scientific literature and databases to develop quantitative structure-property relationship (QSPR) models. We compared both traditional-learning and deep-learning QSPR models and identified models that can accurately predict viscosities using molecular descriptors and temperature as inputs. To further improve model accuracy, we added physically relevant descriptors, extracted using classical molecular dynamics simulations, and encoded them into our QSPR models. We employ feature importance analysis tools to evaluate the influence of molecular-based and physics-informed descriptors on QSPR model performance. The data curation of a large viscosity dataset and the development of accurate, physically informed machine learning models enables the screening of viscosities for the design of new materials.



Back To Top