Abarbanel, Omri David
(2024)
Combining Quantum Mechanical Calculations with Machine Learning and Genetic Algorithms for the Design of Better Materials.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
In the past, the discovery process of new materials was done mainly through trial and error, which was time-consuming and expensive. However, computational simulations and models can quickly filter through a large number of potential candidates and narrow down the search space efficiently and cost-effectively. For example, this process can help identify new electronic materials that use less energy or have novel properties that open the door for new applications and find life-saving drugs for hard-to-cure or rare diseases.
In this work, we present how the combination of several of these computational techniques, namely quantum mechanical (QM) calculations with machine learning (ML) and Genetic Algorithms (GA), can help accelerate the discovery of new materials. We have used GFN2-xTB throughout this work because it has a good balance of accuracy and speed and shows how it can be used as a surrogate for the more costly density functional theory (DFT) and how it can be used to generate molecular features for ML applications.
Three different molecular properties were selected to show how the combination of QM with ML and GAs is greater than the sum of its parts. First, we used GFN2-xTB to calculate geometrical features for a random forest ML algorithm to identify new thiophene-based pi-conjugated polymers with low reorganization energies, achieving a RMSE of 0.036 eV and a speed-up of ~13x over DFT. Second, we used GFN2-xTB calculations in a GA to help identify novel pi-conjugated polymers with stable triplet ground states, finding more than 1,400 potential candidates. Finally, we present QupKake, a graph-neural-networks based ML model that used GFN2-xTB calculated features to predict the micro-pka of drug-like molecules, achieving a $30\%$ improvement over existing models.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
27 August 2024 |
| Date Type: |
Publication |
| Defense Date: |
6 August 2024 |
| Approval Date: |
27 August 2024 |
| Submission Date: |
24 July 2024 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
203 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Dietrich School of Arts and Sciences > Chemistry |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
computational chemistry, material discovery, pka, quantum mechanics, machine learning, genetic algorithms, reorganization energy, ground state triplet |
| Date Deposited: |
27 Aug 2024 13:28 |
| Last Modified: |
27 Aug 2024 13:28 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/46744 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}