Computational and Machine Learning-Assisted Discovery and Experimental Validation of Conjugated Sulfonamide Cathodes for Lithium-Ion Batteries
How Physics-based Modeling and Machine Learning Enable Accelerated Development of Battery Materials
OCT 1, 2024
How Physics-based Modeling and Machine Learning Enable Accelerated Development of Battery Materials
Speaker:
Garvit Agarwal, Senior Scientist II, Schrödinger
Abstract:
The rapid advancements in rechargeable Li-ion battery (LIB) technology over the last decade has revolutionized several key industries such as transportation and consumer electronics. However, new battery chemistries are needed to meet the rapidly growing demand and to improve the power density, safety, reliability, and lifetime of LIBs. Molecular modeling has become an integral part of the design cycle of new battery chemistries. Accurate physics-based modeling enables rapid evaluation and screening of large chemical and material design space thereby, helping industries reduce the time required to bring the new technology to the market. In this webinar, we will introduce the latest technological innovations in Schrödinger’s digital chemistry platform for battery materials design. In particular, the webinar will focus on examples to demonstrate the application of automated solutions for accurate prediction of thermodynamic stability and voltage profile of cathode materials, ion diffusion pathways and kinetics in electrode materials, transport properties of liquid electrolytes and modeling the nucleation and growth of solid electrolyte interphase (SEI) layers using Schrödinger’s SEI simulator module. We will also introduce an automated generalized framework for the development of customized machine learning force fields for complex materials such as liquid electrolytes, inorganic cathode coatings and solid polymer electrolytes, paving the way for efficient design of novel materials for next generation batteries.
Machine Learning for Materials Science
Advanced Machine Learning and Molecular Simulations for Formulation Design
Advanced Machine Learning and Molecular Simulations for Formulation Design
Overview
Complex chemical mixtures — or formulations — are used in a wide range of applications, such as gasoline blends in oil & gas, daily care products in consumer goods, and drug delivery in pharmaceutics. Given the vast number of potential formulations, evolving regulatory requirements, and increasing consumer demand for eco-friendly and sustainable products, we need innovative and cost-effective solutions for designing enhanced formulations. The latest advancements in atomic-scale modeling and machine learning (ML) have enabled computer-aided screening of large numbers of formulation candidates — thus, accelerating the identification of promising formulations and reducing costly experiments.
Schrödinger’s Formulation Machine Learning tool uses data-driven methods to correlate ingredient structure and composition to formulation properties. This tool uses advanced cheminformatics descriptors and automatic hyperparameter tuning to find the best ML model, and allows external features (e.g., temperature, pressure) from experiments or high-throughput molecular dynamics (MD) calculations to be used as additional input to the ML model. The Formulation ML tool enables R&D teams to quickly train and deploy ML models to rapidly explore the broad design space of formulations by varying the chemical ingredients, compositions, and external features.
Advantages of Schrödinger Formulation Screening Technology
- Efficient ML model building and data generation: Leveraging deep learning technology to build accurate ML models to predict formulation properties, which can be coupled with MD simulations as a way to generate physically meaningful descriptors to improve ML model accuracy
- Scalable: ML can be trained and evaluated for mixtures with more than 100 components, extending the capabilities beyond simple mixtures to designing complex mixtures with enhanced properties
- Automated: Automatic hyperparameter tuning enables accurate ML model development using expert cheminformatic descriptors with minimal ML expertise required
- Rapid screening capabilities: ML can generate predictions in a fraction of a second, which can scale up to screening ~100K formulations in the order of minutes-hours
- Dedicated support: Dedicated support team consisting of scientific experts at Schrödinger are available to help users apply computational tools to their applications
- Multiple platform functionality: Can be used on laptops, desktops, and high performance clusters
Applications Across Industries
Consumer Products
Random copolymer systems are often found in packaging materials, and glass transition temperature (Tg) is an important parameter that dictates the stability of the polymer as a function of temperature. Formulation ML can accurately predict Tg for 365 examples with a test set coefficient of determination (R2) of 0.97.2
Energy Storage
Liquid electrolytes are often used in batteries to facilitate the movement of electrical charge between an anode and cathode, and viscosity is an important parameter that dictates how easy ions can move through an electrolyte solution. Formulation ML can accurately predict temperature-dependent viscosity given ~34K examples with a test set R2 of 0.96.3
Pharmaceutical Formulation
Solubility of drug molecules in pure and binary mixture solutions is crucial for drug delivery applications for pharmaceutical formulations. Formulation ML can accurately predict temperature-dependent drug solubility for either pure or binary mixture solutions given ~27K examples, which achieves a test set R2 of 0.93.4
Oil and Gas
Mixtures of hydrocarbons are critical in gasoline blends, facilitating efficient combustion for automotive engines, and motor octane number (MON) is an important parameter that measures the fuel behavior under external pressure. Formulation ML can accurately predict MON given ~700 examples with the number of components ranging from pure (single) component systems to 120 components, which achieves a test set R2 of 0.79.5
Learn more in the tutorial (Note: you will need a web account to access tutorials)
References
-
Leveraging high-throughput molecular simulations and machine learning for the design of chemical mixtures
Alex, C., et al. npj Comput Mater 11, 72, 2025, https://doi.org/10.1038/s41524-025-01552-2.
-
The glass transition temperature of random copolymers: 1. Experimental data and the Gordon-Taylor equation
Penzel, E., et al. Polymer, 38.2, 1997, 325-337.
-
Machine learning for predicting the viscosity of binary liquid mixtures
Bilodeau, C., et al. Chemical Engineering Journal, 464, 2023, 142454.
-
Towards the prediction of drug solubility in binary solvent mixtures at various temperatures using machine learning
Bao, Z., et al. J Cheminform 16, 117, 2024, https://doi.org/10.1186/s13321-024-00911-3.
-
Artificial intelligence-driven design of fuel mixtures
Kuzhagaliyeva, N., et al. Communications Chemistry, 5.1, 2022, 111.
Software and services to meet your organizational needs
Software Platform
Deploy digital materials discovery workflows with a comprehensive and user-friendly platform grounded in physics-based molecular modeling, machine learning, and team collaboration.
Modeling Services
Leverage Schrödinger’s expert computational scientists to assist at key stages in your materials discovery and development process.
Support & Training
Access expert support, educational materials, and training resources designed for both novice and experienced users.
Advancing material property prediction: using physics-informed machine learning models for viscosity
Leveraging atomistic simulation, machine learning, and cloud-based collaborative ideation for display materials discovery
AUG 7, 2024
Leveraging atomistic simulation, machine learning, and cloud-based collaborative ideation for display materials discovery
The rapid evolution of display technology requires the use of cutting-edge research methods to maintain progress. This webinar will explore the union of physics-based simulations, machine learning (ML), and cloud-native collaboration and informatics tools in revolutionizing R&D innovation for display materials.
We will delve into how physics-based simulations provide a robust foundation for understanding and predicting material behaviors, while ML modeling accelerates the discovery and optimization of new materials through data-driven insights. Furthermore, we will introduce Schrödinger’s LiveDesign, a cutting-edge web-based collaboration platform, designed to facilitate R&D in a modern, digital working environment. LiveDesign supports comprehensive functionalities, including modeling, data processing, data storage, and collaborative ideation, empowering teams to work seamlessly across diverse geographical locations.
Join us to gain a deeper understanding of:
- The principles and benefits of combining physics-based simulation and machine learning models
- Strategies for seamless integration of computational approaches in your R&D workflow
- Real-world examples illustrating the application and impact of integrated models in developing superior display materials
- How to leverage LiveDesign for collaborative ideation, advanced modeling, and project management
Our Speaker
Hadi Abroshan
Principal Scientist I, Schrödinger
Hadi Abroshan is the Product Manager for Organic Electronics at Schrödinger, Inc. He holds a Ph.D. from Carnegie Mellon University and has conducted research at Stanford University and Georgia Tech. Hadi specializes in multiscale simulations, leading projects to design cost-effective multifunctional materials for optoelectronics. His expertise lies in developing computational strategies that bridge atomistic structures to multilayered device scales, using a blend of physics-based methodologies and machine learning techniques. His work has led to the discovery of novel, environmentally friendly materials and processes with superior efficiencies.
Electrodes, electrolytes & interfaces: Harnessing molecular simulation and machine learning for rapid advancements in battery materials development
JUN 26, 2024
Electrodes, electrolytes & interfaces: Harnessing molecular simulation and machine learning for rapid advancements in battery materials development
Schrödinger is excited to be presenting in a webinar hosted by the Battery Technology Platform, taking place on June 26th. Join us for a presentation by Garvit Agarwal, Ph.D., Scientific Lead at Schrödinger, titled “Electrodes, electrolytes & interfaces: Harnessing molecular simulation and machine learning for rapid advancements in battery materials development.”
Abstract:
Li-ion battery (LIB) technology has revolutionized industries like transportation and consumer electronics. However, new battery chemistries are needed to address rapidly growing demand and to improve the power density, safety, reliability, and lifetime of LIBs.
In this webinar we will explore the key materials challenges for improving battery performance and demonstrate how atomistic simulation and machine learning (ML) enable swift evaluation and screening of vast design spaces, accelerating the introduction of innovative technology to market.
We will introduce the latest technological innovations in Schrödinger’s digital chemistry platform for battery materials design. In particular, the webinar will focus on examples to demonstrate the application of automated solutions for accurate prediction of thermodynamic stability and voltage profile of cathode, ion diffusion pathways and kinetics in electrode materials, transport properties of liquid electrolytes and modeling the nucleation and growth of solid electrolyte interphase (SEI) layers using Schrödinger’s SEI simulator module. We will also introduce an automated generalized framework for the development of customized machine learning force fields for complex materials such as liquid electrolytes, inorganic cathode coatings and solid polymer electrolytes, paving the way for efficient design of novel materials for next generation batteries.
Key Learning Objectives:
- Gain insight into how a digital chemistry approach reduces development cycle time for new battery materials
- Learn how Schrödinger’s automated high throughput simulation workflows and ML models enable rapid screening of battery materials candidates
- Learn how to leverage automated solutions for accurate prediction of key battery properties: thermodynamic stability and voltage profile of electrodes, ion diffusion pathways and kinetics in electrodes, transport properties of liquid electrolytes, and the nucleation and growth of solid electrolyte interphase (SEI) layers
- Hear applications of advanced machine learning force fields for accurate modeling of electrolyte materials, cathode coatings and interfaces
Garvit Agarwal, Ph.D.
Scientific Lead, Energy Storage Materials Science Group, Schrödinger
Garvit Agarwal is Senior Scientist and Scientific Lead for Energy Storage at Schrödinger, working to extend and apply molecular modeling tools for the accelerated discovery of next-generation clean energy technologies. Garvit obtained his Ph.D. in Materials Science and Engineering from the University of Connecticut. He worked as a post-doctoral researcher in the Materials Science Division at Argonne National Laboratory prior to joining the Materials Science team at Schrödinger.
Hole transport materials for QLEDs: a combined approach of machine learning and atomistic simulation
Expediting FEP+ model optimization for challenging systems with a fully automated, machine learning-driven workflow
JUN 25, 2024
Expediting FEP+ model optimization for challenging systems with a fully automated, machine learning-driven workflow
FEP+ is a powerful predictive technology in drug discovery – with applications from hit discovery through lead optimization. A critical first step in deploying FEP+ is to validate and optimize the model for the protein-ligand system of interest. While some systems perform well with out-of-the-box FEP+ settings, others require manual protocol refinement.
In this webinar, we will introduce Schrödinger’s FEP+ Protocol Builder, an automated machine learning workflow for FEP+ model optimization. This workflow is designed for systems with insufficient predictive accuracy using default settings or after initial manual protocol optimization attempts. FEP+ Protocol Builder saves researcher time and increases the chances of successfully enabling FEP+ by efficiently identifying an optimized predictive model for your system of interest.
Join us as we share key features of FEP+ Protocol Builder and highlight case studies that have shown success utilizing the technology to accelerate projects.
Highlights:
- Overview of FEP+ Protocol Builder technology, which uses an Active Learning workflow to iteratively search the protocol parameter space to develop accurate FEP+ protocols
- Case studies on effective prospective use of protocols generated by FEP+ Protocol Builder in drug discovery programs
- Comparison of time efficiency between manual and automated FEP+ optimization
- Details on requirements and services options to leverage the technology in your own program
- Overview of the different options to enable FEP+ at almost all levels of structural information for your protein-ligand system of interest
Our Speakers
Jeremie Vendome
Senior Director, Applications Science, Schrödinger
Dr. Jeremie Vendome, senior director of applications science, joined Schrödinger in 2015. He received his PhD from the Ecole Normale Superieure in France and completed his training at Columbia University in the lab of Prof. Barry Honig, where he worked on various problems related to protein-protein interaction energetics and specificity by combining computational approaches and experiments. Prior to joining Schrödinger, Jeremie acquired 10+ years of drug discovery experience both as a computational chemist in the industry and as the head of a CADD collaborative platform at Columbia Medical Center. At Schrödinger, he has held several roles of increasing responsibility and has continuously been at the interface between the company’s latest technological developments and their applications in active drug discovery projects. Most recently, Jeremie has been spearheading Schrödinger’s research enablement initiative, meant to advance drug discovery programs though key stages by providing access to our latest technologies and workflows at scale as a collaborative service.
Jordan Epstein
Product Manager, Schrödinger
Jordan Epstein joined Schrödinger in 2017 after studying Chemistry at New York University. Upon joining Schrödinger as a software developer, he worked on products such as FEP+, Desmond, and LiveDesign. More recently, he transitioned into his role as product manager where he has helped to see FEP+ Protocol Builder to its full release.
Sathesh Bhat
Executive Director, Therapeutics Group, Schrödinger
Sathesh Bhat, Ph.D., executive director in the therapeutics group, joined Schrödinger in 2011. He is responsible for overseeing computational chemistry efforts on internal and partnered drug discovery programs at Schrödinger. Previously, Sathesh worked at both Merck and Eli Lilly leading computational efforts in several drug discovery programs. He obtained his Ph.D. from McGill University, which involved developing structure-based methods to predict binding free energies. Sathesh has co-authored multiple patents and publications and continues to publish on a wide variety of topics in computational chemistry.
Machine learning force field ranking of candidate solid electrolyte interphase structures in Li-ion batteries
Trends in modern hit discovery: How your ultra-large screens can benefit from machine learning
FEB 2, 2022
Trends in modern hit discovery: How your ultra-large screens can benefit from machine learning
Speaker:
Matt Repasky
Senior Vice President
Abstract:
While traditional structure-based virtual screening has been successful in finding diverse hits to advance projects there is significant room for improvement of hit rates, diversity of hit chemotypes, available IP space explored, and the potency of unoptimized hits. Ultra-large, on-demand synthesizable libraries from vendors have enabled ~100x expansion of purchasable compound space, now billions of compounds, while DNA encoded libraries (DEL) can be even larger. In order to screen these much larger chemical spaces in the billions of compounds, results of two machine learning enabled approaches are described that make it easy and cost effective to find novel hits through virtual and DEL screens of billion compound plus libraries. DNA encoded libraries (DEL) enable screening billions of synthesized compounds but are limited due to high rates of experimental false negatives and positives. Employing machine learning trained to experimental DEL results we demonstrate significantly reduced false negative rates while identifying byproducts in a more favorable property space. To enable efficient, extrapolative chemical space exploration with an accurate docking scoring function, we have developed an active learning-based method employing AutoQSAR/DC machine learning and Glide SP docking as the learner. Results from Active Learning Glide screening of 100 million to billion compound screens show increased chemical diversity and GlideScore of hits relative to brute force screening of subsets of the libraries. Results and costs from these two new methods suggest billion compound library screens could replace smaller, traditional screens commonly employed today.