Free Energy Calculation (John Chodera) and AI Assisted Committor - - PowerPoint PPT Presentation

HKUST Deutschland ZIB of Freie U

Selected talk at the Molecular Kinetics conference held in Berlin:

Free Energy Calculation (John Chodera) and AI Assisted Committor Discovery (Gerhard Hummer)

Siqin Cao

Department of Chemistry The Hong Kong University of Science and Technology

Molecular Kinetics and Free Energy Calculation

John Chodera

Drug discovery usually ends in failure

• S. Paul, D. Mytelka, ..., A. Schacht, Nature Reviews Drug Discovery 9, 203 (2010)

4.5 years and $219M

R&D model yielding costs to successfully discover and develop a single new molecular entity:

p(TS): probability to next stage WIP: work in progress NME: new molecule entity Unit of money: million USD

AMBER/GAFF for Relative Free Energy Calculations

Lin Frank Songa, Tai-Sung Leeb, Chun-Zhua, Darrin M. Yorkb, and Kenneth M. Merz Jr, ChemRxiv 7653434 (2019)

OPLS2.1+REST2: MUE = 0.9 kcal/mol, RMSE = 1.14 kcal/mol AMBER99SB/GAFF+MD for 330 mutations: MUE = 1.15 kcal/mol, RMSE = 1.5 kcal/mol

SMIRNOFF: SMIRks Native Open Force Field

FF frustrations: Rely on expert intuition; Poor choice for data structure; Massive reduction in parameters.

SMARTS: direct chemical perception Atom types: indirect chemical perception

FF optimization: Auomated through Bayesian approach.

Machine learning will have a large impact on physical modeling

We can learn:

• more accurate potential functions
• optimal alchemical protocols for each

transformation

• “difficulty” of transformations
• estimates of ΔG and ΔΔG directly
• estimates of conformational reorganization

energies

AI-Assisted Discovery of Molecular Mechanisms from Simulations

Peter G. Bolhuis, David Chandler, Christoph Dellago, and Phillip L. Geissler, Annu. Rev. Phys. Chem. 53, 291 (2002)

Gerhard Hummer Challenges in MD of biological systems:

• 1. Sampling problem
• 2. Interpretation problem

Transition Path Sampling

Transition path “shooting from the top”

Hendrik Jung , Kei-ichi Okazaki, and Gerhard Hummer, JCP 147, 152716 (2017)

Machine learning of committor

• P. G. Bolhuis, D. Chandler, C. Dellago, & P. L. Geissler, Annu. Rev. Phys. Chem. 53, 291 (2002); H. Jung , K. Okazaki, &
• G. Hummer, JCP 147, 152716 (2017); H. Jung, R. Covino, & G. Hummer, arxiv 1901.04595 (2019)

AI assisted discovery of reaction coordinates

Hendrik Jung, Roberto Covino, and Gerhard Hummer, arxiv 1901.04595 (2019)

Committor (p) vs reaction coordinate (q): Transition state ensumble: Committor modeled with self-normalizing ANN

Metropolis-Hastings criterion of path acceptance:

Total likelihood: Loss function to minimize:

AI assisted discovery of reaction coordinates

Hendrik Jung, Roberto Covino, and Gerhard Hummer, arxiv 1901.04595 (2019)