Using machine learning to design extractants for improved - - PowerPoint PPT Presentation

Using machine learning to design extractants for improved separations

Marilu G. Perez In collaboration with: Federico Zahariev, Theresa L. Windus, Benjamin Hay

1

Purpose

• Critical Materials Institute
• keep the U.S. competitive in energy technology by securing “critical” materials
• Collaboration with industry is important
• Industry is practical – but can open up new avenues of research
• Social Science Example: Crisis Text Line (crisistextline.org)
• Correlation between high-risk users and words “Advil” and “ibuprofen” over

intuitive words such as “die” or “suicide”

• Result more important than reason for outcome

2

Collaboration

Experimental Chemistry Computer Science Data Science

Federico Zahariev Marilu Perez

Computational/Theoretical Chemistry Project

3

Goal

• Determine metal-ligand complex equilibrium constants, K
• Identify system descriptors – properties that may correlate to K values
• Use supervised learning
• Relatively limited amount of data
• Some knowledge of the system exists
• Ultimately – predict distribution coefficients/separation factors

4

Predicting Value of Interest

• Descriptors
• Considerations
• Challenges

Sample Descriptors Values 1 2 … n Known 1 D₁₁ D₂₂ … Dnn V₁ . . . … . . m D₁m D₂m … Dnm Vm Sample Descriptors Values 1 2 … n Known Predicted 1 D₁₁ D₂₂ … Dnn V₁ V₁ . . . … . . . m D₁

m

D₂m … Dnm Vm Vm

Training Set Test Set

5

Future Outlook

• Open access data
• Experimental conditions
• Experimental analysis
• Raw data
• Incorporating “failed” experiments
• More collaboration

6

Experimental Chemistry Computer Science Data Science Computational/Theoretical Chemistry