Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (22 results)

  • Thesis
  • Peer Reviewed
UC Berkeley Electronic Theses and Dissertations (2023)

Adapting machine learning (ML) models to the science of solids holds great promise for designing functional materials with unprecedented scale and speed. However, the increasing popularity of materials-ML models raises concerns about how they should be leveraged. Validation is not standardized, making models difficult to compare and reproduce. Heterogeneous materials datasets generated with dissimilar methods are difficult to fuse into new discoveries and insights. Most importantly, the vast majority of materials data cannot be easily used to train models because it is dispersed in scientific text.

This thesis examines these concerns through ML approaches to semiconductor material design. This composite approach utilizes high-throughput density functional theory (HT-DFT) calculations, supervised ML models, natural language processing (NLP), and the classical and quantum theories of solids. Towards fusing disparate sources of materials data, the HT-DFT/ML screening presented identifies rare earth phosphide bulk thermoelectrics (REZnCuP2, zTexpt. ≈ 0.5) alongside classes of metallic (but near-band gap) thermoelectric Chevrel phases, La3Te4-type phases, clathrates, and others. By fitting simple linear ML models to HT-DFT data, we discern new chemical rules for engineering valence bands relevant to thermopower in Half-Heusler phases. Expanding the ML approach beyond thermopower, this thesis also describes the first widely-accepted benchmark for evaluating ML models, including state-of-the-art graph neural networks, on structure-property/composition-property prediction tasks. Finally, the NLP approaches detail a framework for materials data in text to be leveraged on a massive scale. Moving from word2vec to large language models such as GPT-3, this thesis presents NLP models of various complexities as tools to extract and analyze materials knowledge in the aggregate (e.g., using downstream supervised models). These tools are used to predict high-performance thermoelectric materials as well as gain a deeper understanding of dopant selection.

    Cover page: Machine Learning for Semiconducting Solids: Benchmarking, Thermoelectric Discovery, and Dopant Engineering
    • Article
    • Peer Reviewed
    LBL Publications (2020)
    We present a benchmark test suite and an automated machine learning procedure for evaluating supervised machine learning (ML) models for predicting properties of inorganic bulk materials. The test suite, Matbench, is a set of 13 ML tasks that range in size from 312 to 132k samples and contain data from 10 density functional theory-derived and experimental sources. Tasks include predicting optical, thermal, electronic, thermodynamic, tensile, and elastic properties given a material’s composition and/or crystal structure. The reference algorithm, Automatminer, is a highly-extensible, fully automated ML pipeline for predicting materials properties from materials primitives (such as composition and crystal structure) without user intervention or hyperparameter tuning. We test Automatminer on the Matbench test suite and compare its predictive power with state-of-the-art crystal graph neural networks and a traditional descriptor-based Random Forest model. We find Automatminer achieves the best performance on 8 of 13 tasks in the benchmark. We also show our test suite is capable of exposing predictive advantages of each algorithm—namely, that crystal graph methods appear to outperform traditional machine learning methods given ~104 or greater data points. We encourage evaluating materials ML algorithms on the Matbench benchmark and comparing them against the latest version of Automatminer.
      Cover page: Benchmarking materials property prediction methods: the Matbench test set and Automatminer reference algorithm
      • Article
      • Peer Reviewed
      UC Santa Barbara Previously Published Works (2016)
      Integrins mediate cell adhesion to the extracellular matrix and enable the construction of complex, multicellular organisms, yet fundamental aspects of integrin-based adhesion remain poorly understood. Notably, the magnitude of the mechanical load experienced by individual integrins within living cells is unclear, due principally to limitations inherent to existing techniques. Here we use Förster resonance energy transfer-based molecular tension sensors to directly measure the distribution of loads experienced by individual integrins in living cells. We find that a large fraction of integrins bear modest loads of 1-3 pN, while subpopulations bearing higher loads are enriched within adhesions. Further, our data indicate that integrin engagement with the fibronectin synergy site, a secondary binding site specifically for α5β1 integrin, leads to increased levels of α5β1 integrin recruitment to adhesions but not to an increase in overall cellular traction generation. The presence of the synergy site does, however, increase cells resistance to detachment by externally applied loads. We suggest that a substantial population of integrins experiencing loads well below their peak capacities can provide cells and tissues with mechanical integrity in the presence of widely varying mechanical loads.
        Cover page: Single Molecule Force Measurements in Living Cells Reveal a Minimally Tensioned Integrin State.
        • Article
        • Peer Reviewed
        LBL Publications (2020)
        The typical strategy to design high performance thermoelectric materials is to dope a semiconducting material until optimal properties are obtained. However, some known thermoelectric materials such as La3Te4, Mo3Sb7, Yb14MnSb11, and NbCoSb are actually gapped metals,i.e., their band structure displays a gap slightly above or below the band crossed by the Fermi level. This key feature makes these metals comparable to degenerate semiconductors and thus suitable for thermoelectric applications. In this work, we perform a computational high-throughput search for such gapped metals exhibiting attractive thermoelectric properties. Several thousands of metals are found to present this key feature, and about one thousand of them show decent thermoelectric properties as evaluated by a computedzT. We present the different chemistry of gapped metals we discovered such as clathrates, Chevrel phases, or transition metal dichalcogenides and discuss their previous studies as thermoelectric and their potential as new thermoelectric materials.
          Cover page: Gapped metals as thermoelectric materials revealed by high-throughput screening
          • Article
          • Peer Reviewed
          Energy Storage & Distributed Resources (2020)
          Half-Heusler materials are strong candidates for thermoelectric applications due to their high weighted mobilities and power factors, which is known to be correlated to valley degeneracy in the electronic band structure. However, there are over 50 known semiconducting half-Heusler phases, and it is not clear how the chemical composition affects the electronic structure. While all the n-type electronic structures have their conduction band minimum at either the Γ- or X-point, there is more diversity in the p-type electronic structures, and the valence band maximum can be at either the Γ-, L-, or W-point. Here, we use high throughput computation and machine learning to compare the valence bands of known half-Heusler compounds and discover new chemical guidelines for promoting the highly degenerate W-point to the valence band maximum. We do this by constructing an "orbital phase diagram" to cluster the variety of electronic structures expressed by these phases into groups, based on the atomic orbitals that contribute most to their valence bands. Then, with the aid of machine learning, we develop new chemical rules that predict the location of the valence band maximum in each of the phases. These rules can be used to engineer band structures with band convergence and high valley degeneracy.
            Cover page: Machine Learning Chemical Guidelines for Engineering Electronic Structures in Half-Heusler Thermoelectric Materials
            • Article
            • Peer Reviewed
            UC Santa Barbara Previously Published Works (2021)

            Introduction

            Cell structure and migration is impacted by the mechanical properties and geometry of the cell adhesive environment. Most studies to date investigating the effects of 3D environments on cells have not controlled geometry at the single-cell level, making it difficult to understand the influence of 3D environmental cues on single cells. Here, we developed microwell platforms to investigate the effects of 2D vs. 3D geometries on single-cell F-actin and nuclear organization.

            Methods

            We used microfabrication techniques to fabricate three polyacrylamide platforms: 3D microwells with a 3D adhesive environment (3D/3D), 3D microwells with 2D adhesive areas at the bottom only (3D/2D), and flat 2D gels with 2D patterned adhesive areas (2D/2D). We measured geometric swelling and Young's modulus of the platforms. We then cultured C2C12 myoblasts on each platform and evaluated the effects of the engineered microenvironments on F-actin structure and nuclear shape.

            Results

            We tuned the mechanical characteristics of the microfabricated platforms by manipulating the gel formulation. Crosslinker ratio strongly influenced geometric swelling whereas total polymer content primarily affected Young's modulus. When comparing cells in these platforms, we found significant effects on F-actin and nuclear structures. Our analysis showed that a 3D/3D environment was necessary to increase actin and nuclear height. A 3D/2D environment was sufficient to increase actin alignment and nuclear aspect ratio compared to a 2D/2D environment.

            Conclusions

            Using our novel polyacrylamide platforms, we were able to decouple the effects of 3D confinement and adhesive environment, finding that both influenced actin and nuclear structure.
              Cover page: 3D Microwell Platforms for Control of Single Cell 3D Geometry and Intracellular Organization
              • Article
              • Peer Reviewed
              UC Santa Barbara Previously Published Works (2020)

              Introduction

              Cell structure and migration is impacted by the mechanical properties and geometry of the cell adhesive environment. Most studies to date investigating the effects of 3D environments on cells have not controlled geometry at the single-cell level, making it difficult to understand the influence of 3D environmental cues on single cells. Here, we developed microwell platforms to investigate the effects of 2D vs 3D geometries on single-cell F-actin and nuclear organization.

              Methods

              We used microfabrication techniques to fabricate three polyacrylamide platforms: 3D microwells with a 3D adhesive environment ( 3D/3D ), 3D microwells with 2D adhesive areas at the bottom only ( 3D/2D ), and flat 2D gels with 2D patterned adhesive areas ( 2D/2D ). We measured geometric swelling and Young’s modulus of the platforms. We then cultured C2C12 myoblasts on each platform and evaluated the effects of the engineered microenvironments on F-actin structure and nuclear shape.

              Results

              We tuned the mechanical characteristics of the microfabricated platforms by manipulating the gel formulation. Crosslinker ratio strongly influenced geometric swelling whereas total polymer content primarily affected Young’s modulus. When comparing cells in these platforms, we found significant effects on F-actin and nuclear structures. Our analysis showed that a 3D/3D environment was necessary to increase actin and nuclear height. A 3D/2D environment was sufficient to increase actin alignment and nuclear aspect ratio compared to a 2D/2D environment.

              Conclusions

              Using our novel polyacrylamide platforms, we were able to decouple the effects of 3D confinement and adhesive environment, finding that both influenced actin and nuclear structure.
                Cover page: 3D Microwell Platforms for Control of Single Cell 3D Geometry and Intracellular Organization
                • Article
                • Peer Reviewed
                UC Santa Barbara Previously Published Works (2014)
                External forces play complex roles in cell organization, fate, and homeostasis. Changes in these forces, or how cells respond to them, can result in abnormal embryonic development and diseases in adults. How cells sense and respond to these mechanical stimuli requires an understanding of the biophysical principles that underlie changes in protein conformation and result in alterations in the organization and function of cells and tissues. Here, we discuss mechano-transduction as it applies to protein conformation, cellular organization, and multi-cell (tissue) function.
                  Cover page: Mechano-Transduction: From Molecules to Tissues
                  • Article
                  • Peer Reviewed
                  UC Berkeley Previously Published Works (2020)
                  Machine learning has emerged as a novel tool for the efficient prediction of material properties, and claims have been made that machine-learned models for the formation energy of compounds can approach the accuracy of Density Functional Theory (DFT). The models tested in this work include five recently published compositional models, a baseline model using stoichiometry alone, and a structural model. By testing seven machine learning models for formation energy on stability predictions using the Materials Project database of DFT calculations for 85,014 unique chemical compositions, we show that while formation energies can indeed be predicted well, all compositional models perform poorly on predicting the stability of compounds, making them considerably less useful than DFT for the discovery and design of new solids. Most critically, in sparse chemical spaces where few stoichiometries have stable compounds, only the structural model is capable of efficiently detecting which materials are stable. The nonincremental improvement of structural models compared with compositional models is noteworthy and encourages the use of structural models for materials discovery, with the constraint that for any new composition, the ground-state structure is not known a priori. This work demonstrates that accurate predictions of formation energy do not imply accurate predictions of stability, emphasizing the importance of assessing model performance on stability predictions, for which we provide a set of publicly available tests.
                    Cover page: A critical examination of compound stability predictions from machine-learned formation energies
                    • Article
                    • Peer Reviewed
                    UC Berkeley Previously Published Works (2019)
                    The overwhelming majority of scientific knowledge is published as text, which is difficult to analyse by either traditional statistical analysis or modern machine learning methods. By contrast, the main source of machine-interpretable data for the materials research community has come from structured property databases1,2, which encompass only a small fraction of the knowledge present in the research literature. Beyond property values, publications contain valuable knowledge regarding the connections and relationships between data items as interpreted by the authors. To improve the identification and use of this knowledge, several studies have focused on the retrieval of information from scientific literature using supervised natural language processing3-10, which requires large hand-labelled datasets for training. Here we show that materials science knowledge present in the published literature can be efficiently encoded as information-dense word embeddings11-13 (vector representations of words) without human labelling or supervision. Without any explicit insertion of chemical knowledge, these embeddings capture complex materials science concepts such as the underlying structure of the periodic table and structure-property relationships in materials. Furthermore, we demonstrate that an unsupervised method can recommend materials for functional applications several years before their discovery. This suggests that latent knowledge regarding future discoveries is to a large extent embedded in past publications. Our findings highlight the possibility of extracting knowledge and relationships from the massive body of scientific literature in a collective manner, and point towards a generalized approach to the mining of scientific literature.
                      Cover page: Unsupervised word embeddings capture latent knowledge from materials science literature
                      Top