Direct Force Fieldセミナー

高精度分子シミュレーション支援ソフトウエア「Direct Force Field」は、「力場パラメータが足りなくて計算できない」、「化学的に誤った構造を結果として与える」というような力場パラメータの問題を解決するために開発され、精度の高い力場を提供することで、高精度な分子シミュレーションを支援してきました。

弊社ではDirect Force Fieldと背景にある技術を紹介するために、Direct Force Fieldの開発元のAeon Technology社の開発責任者であるDr. Huai Sunを招聘し、オンラインセミナーを開催いたします。セミナーでは、力場パラメータの自動作成を視野に現在開発中の「Automation of Force Field Development」機能を中心にDirect Force Fieldの概要と応用事例について講演する予定です。



* Direct Force Fieldに関する詳細は、こちらをご覧ください。


日付 時間 オンラインシステム
10:30-11:40 Zoom




時間 講演内容 講演者
10:30~10:40 はじめに(開発者ならびにDFFについての概要紹介) 株式会社モルシス 佐藤史一
10:40~11:30 Automation of Force Field Development Aeon Technology Inc
Shanghai Jiao Tong University
Dr. Huai Sun
11:30~11:40 質疑応答



①セミナー名「Direct Force Fieldセミナー」



Force field can expand the time and length scales of computation significantly, is one of the most important tools in molecular simulation. Meanwhile, force field method can be daunting. This is because force field is a nonlinear regression model that predicts potential energies  as functions of molecular structures. All regressions inevitably depend on the data used for the regression. In addition, the empirical functions used in force field cannot fully describe all scientific phenomena. Therefore, the reliability and transferability of force field is limited.
A force field works only for what it is designed for, beyond the scope making a new force field is often required. Recent advances in machine learning illustrates that complicated regression models can be made successfully by using big data. As robust computational methods based on quantum mechanics and statistical mechanics are generally accessible, the possibility of making big data for force field development from first principles becomes reality. The problem that needs to be solved is how to generate appropriate big data and obtain force field parameters from it. The automated workflow method implemented in DFF8 is dedicated to solving this problem.

The main functionalities of the workflow are:
A) It finds representative fragments suitable for parameterization for target molecules.
B) It carries out QM calculations to sample the potential-energy hypersurfaces and generate data for regression.
C) It fits the data to a nonlinear regression model using least squares technique.
D) It optimizes the model by using MD simulations to fit condense-phase properties to include multibody and polarization effects.

The workflow can be executed in batch automatically, it can be used for making either a specific force field for solving a research problem or a general force field for group or enterprise usages.
In this presentation we will go over the functionality with real examples in life and material sciences and discuss its advantage as well as limitations.