Live broadcast of materials science ｜ When molecular design and experimental design have AI blessing： MaXFlow artificial intelligence application is upgraded again.

Lecture topic:When molecular design and experimental design have AI blessing, the new version of MaXFlow3.6.1 material will be released.

Lecture time:Thursday, March 16th, 2023, 15:00.

Speaker:Chuangteng Technology Lecturer

It takes only 2 seconds to obtain a stable 3D molecular structure?

The generation of three-dimensional conformation of molecules plays a key role in the design of small molecules in materials science, life science and other fields. Traditional methods usually need to predict the distance between atoms, the gradient of distance or the local structure such as twist angle, and then reconstruct its 3D conformation.

In MaXFlow3.6.1, the DMCG component will be launched soon. DMCG is a method to directly predict the atomic scale, that is, to directly generate the three-dimensional coordinates of molecules without considering the intermediate value. This method can adaptively combine bond and atomic information, and refine the generated conformational coordinates through continuous iteration. This method has achieved good performance on both GEOM-QM9 and GEOM-Drugs data sets.

MaXFlow3.6.1 is about to launch the online molecular 3D coordinate generation module DMCG. DMCG can directly predict the three-dimensional coordinates of each atom without providing intermediate values such as bond length and bond angle, and visualize the optimized three-dimensional structure. This method automatically extracts bond and atom information from molecular SMILES, and obtains the lowest energy conformation through iterative optimization. DMCG has achieved excellent performance on both GEOM-QM9 and GEOM-Drugs data sets, and its intrinsic properties such as band gap of the generated conformation are consistent with the real values. This method will provide an effective basis for subsequent property calculation and molecular simulation.

With it, the optimization of reaction conditions will no longer make you "bald"

In the process of preparing new materials, many reaction parameters are usually involved. In order to find the best preparation scheme and obtain the best yield, a lot of experimental verification is often needed, which brings a lot of workload to researchers and is very time-consuming and labor-intensive. How to optimize the experimental design scheme by means of AI, so that researchers can find the best scheme only through a limited number of experiments?

MaXFlow, a molecular simulation and artificial intelligence platform, integrates machine learning algorithm with experimental design (DOE), which helps researchers quickly lock the optimal interval and greatly reduces the number of verification experiments.

In MaXFlow3.6.1, EDBO component will be launched soon, which integrates Bayesian optimization algorithm. Bayesian optimization is a global optimization algorithm based on iterative response surface, which shows excellent performance in the tuning of machine learning model. In recent years, Bayesian optimization has also been applied to the chemical field. Using Bayesian optimization method in experimental design can effectively help optimize process parameters, thus realizing data-driven decision-making.

At this conference, you will also learn about MaXFlow’s rich model building functions.