With a 20-letter alphabet, conceivable protein sequence-space is enormous; sparks of structure and function are vanishingly rare. Despite massive advances in AI-guided protein design, we remain largely ignorant of the sequences and structures that populate the depths of protein space more than a handful of mutations away from what nature has tried. In this work, we leverage the potential of one specific class of AI protein model \u2014 the protein language model, or PLM \u2014 to internalize the essential features of the protein sequence-structure map while retaining the capacity to explore its extremes. Guided by a \"novelty first, fitness next\" mentality, we harness this balance towards systematic discovery of new-to-nature sequences and structures throughout deep protein space.
\r\n\r\nIn the first section, we dissect the ability of PLMs to explore natural and novel regimes of sequence and structure during free generation. We find that while these models readily emit novel sequences encoding artificial proteins that appear biophysically feasible in silico, they fail to completely or representatively capture the known distribution of natural protein structures. We expose a fundamental tradeoff between the ability of a PLM to generate with sequence novelty or structural coverage but not both simultaneously; prioritizing sampling of far-from-natural sequences triggers a collapse to a handful of simple structural motifs and disordered regions.
\r\n\r\nTurning this sequence novelty vs. structural breadth tradeoff to our advantage, the second section is devoted to the development of \"foldtuning\" \u2014 a structure-preserving, sequence-remodeling engine for navigating the far corners of sequence-space with PLM-based probes. We successfully scale and deploy foldtuning for >700 targets, pushing artificial sequences past the point of detectable homology to any real protein documented in nature, discovering novel sequence-level semantics and grammar for mimicking known protein folds, and accessing potential reservoirs of downstream structural and functional innovation. Experimental validation of select targets reveals that foldtuning produces realizable and functional binders in contexts including a toxin/antitoxin system and peptide hormone signaling.
\r\n\r\nShifting to focus on structural novelty, the final section introduces two PLM-driven methods for the discovery of new-to-nature structures. We show that with appropriate steering functions, PLMs readily yield well-structured domains (featuring diverse secondary and supersecondary elements) outside the several thousand such families cataloged from among known proteins. Overall, this work makes substantial inroads towards the challenge of locating viable far-from-natural regions of protein density across the global sequence-structure map, and revises our notions of the physical constraints on sequence and structure in valid proteins. Moreover, it sets the stage for future assembly of synthetic biological systems composed fully of new-to-nature parts and ultimately for modeling efforts that close the design loop from sequence all the way to complex phenotype.
", "doi": "10.7907/p4st-m614", "publication_date": "2026", "thesis_type": "phd", "thesis_year": "2026" }, { "id": "thesis:17566", "collection": "thesis", "collection_id": "17566", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:07282025-205651638", "primary_object_url": { "basename": "white_elephants_and_cash_cows.pdf", "content": "final", "filesize": 12002392, "license": "other", "mime_type": "application/pdf", "url": "/17566/1/white_elephants_and_cash_cows.pdf", "version": "v4.0.0" }, "type": "thesis", "title": "White Elephants and Cash Cows: Economically Wrangling the Zoo of AI Models", "author": [ { "family_name": "Zellinger", "given_name": "Michael J.", "orcid": "0009-0001-7499-148X", "clpid": "Zellinger-Michael-J" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" }, { "family_name": "B\u00fchlmann", "given_name": "Peter", "orcid": "0000-0002-1782-6015", "clpid": "B\u00fchlmann-Peter" } ], "thesis_committee": [ { "family_name": "Perona", "given_name": "Pietro", "orcid": "0000-0002-7583-5809", "clpid": "Perona-P" }, { "family_name": "Wierman", "given_name": "Adam C.", "orcid": "0000-0002-5923-0199", "clpid": "Wierman-A-C" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" }, { "family_name": "B\u00fchlmann", "given_name": "Peter", "orcid": "0000-0002-1782-6015", "clpid": "B\u00fchlmann-Peter" } ], "local_group": [ { "literal": "div_eng" } ], "abstract": "The capabilities of artificial intelligence are rapidly expanding, but deploying AI systems in practice still poses significant challenges. Specifically, practitioners find limited guidance on selecting the most suitable AI model for a concrete use case, balancing the economics of an AI deployment, and managing the risk of AI errors. These challenges call for a unified framework addressing pain points in a conceptually clear and statistically sound manner. In this thesis, we present several components of such a framework: 1) uncertainty-aware system optimization, 2) economic evaluation, 3) error reduction with human-in-the-loop, and 4) a proof-of-concept system for synthetic data generation. Our work presents novel technical and conceptual approaches for orchestrating natural language-based systems, advancing the economical and reliable deployment of artificial intelligence.", "doi": "10.7907/xj31-xm14", "publication_date": "2026", "thesis_type": "phd", "thesis_year": "2026" }, { "id": "thesis:17550", "collection": "thesis", "collection_id": "17550", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:07232025-195327620", "primary_object_url": { "basename": "Thesis_Enrique (4).pdf", "content": "final", "filesize": 30948826, "license": "cc_by_nc_sa", "mime_type": "application/pdf", "url": "/17550/1/Thesis_Enrique (4).pdf", "version": "v4.0.0" }, "type": "thesis", "title": "Designing Intelligent Agents for Real-Time Experimental Control and Multi-Task Generalization", "author": [ { "family_name": "Amaya Perez", "given_name": "Enrique", "orcid": "0000-0003-3166-8583", "clpid": "Amaya-Perez-Enrique" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Sternberg", "given_name": "Paul W.", "orcid": "0000-0002-7699-0173", "clpid": "Sternberg-P-W" }, { "family_name": "Rutishauser", "given_name": "Ueli", "orcid": "0000-0002-9207-7069", "clpid": "Rutishauser-U" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" }, { "family_name": "Van Valen", "given_name": "David A.", "orcid": "0000-0001-7534-7621", "clpid": "Van-Valen-D" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "Scientific discovery has traditionally relied on human-led iterative loops of observation, modeling, and intervention. This thesis explores the possibility of automating components of this loop using artificial intelligence (AI), particularly in systems characterized by non-equilibrium dynamics, high dimensionality, and emergent behaviors. Two foundational challenges are addressed: automating physical modeling and enabling adaptive interaction with dynamic experimental systems, and generalizing agent behavior across tasks and contexts without retraining.
\r\n\r\nTo address the first challenge, we introduce a hierarchical AI framework for controlling active biomolecular matter, exemplified by microtubule\u2013kinesin networks driven by light-activated motors. At the foundation are predictive models that learn the system\u2019s response to static light patterns, enabling inverse design by selecting inputs that yield desired structural outcomes. Building on this, dynamic models construct low-dimensional representations of the system\u2019s evolving state under time-varying stimuli, supporting forward simulation and real-time tracking. At the highest level, reinforcement learning agents\u2014trained in simulation\u2014discover and execute closed-loop control policies that achieve fine-grained manipulation objectives. These agents are deployed across ~100 parallel experimental setups, demonstrating autonomous operation with robustness, scalability, and reliable transfer.
\r\n\r\nTo address the second challenge, we investigate how generalist reinforcement learning agents can be constructed by leveraging the geometry of policy parameter space. We show that agents trained on distinct tasks self-organize into functionally segregated regions of weight space that encode both task identity and strategic variability. This insight enables the design of a hypernetwork\u2014a network that generates the weights of other networks\u2014that can interpolate smoothly between tasks and strategies via a single scalar input. Combined with a meta-controller, this architecture enables real-time modulation of agent behavior\u2014ranging from conservative to risk-seeking\u2014without retraining.
\r\n\r\nTogether, these contributions demonstrate that intelligent systems can both design and control physical experiments in real time, and adapt cognitive strategies across tasks through principled representations in policy space. This work establishes a foundation for closed-loop scientific autonomy, programmable biomaterials, and generalist AI agents, converging at the intersection of machine learning, biophysics, and automation.
", "doi": "10.7907/nmvs-7b59", "publication_date": "2026", "thesis_type": "phd", "thesis_year": "2026" }, { "id": "thesis:17383", "collection": "thesis", "collection_id": "17383", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:06022025-234901151", "type": "thesis", "title": "The Topology of Cellular Ontogeny", "author": [ { "family_name": "Flores-Bautista", "given_name": "Emanuel", "orcid": "0000-0002-2810-1757", "clpid": "Flores-Bautista-Emanuel" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Prober", "given_name": "David A.", "orcid": "0000-0002-7371-4675", "clpid": "Prober-D-A" }, { "family_name": "Marcolli", "given_name": "Matilde", "orcid": "0000-0002-2045-2907", "clpid": "Marcolli-M" }, { "family_name": "Pachter", "given_name": "Lior S.", "orcid": "0000-0002-9164-6231", "clpid": "Pachter-L" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "A fundamental goal of modern biology is to build global, predictive models of gene regulation that encompass diverse physiological contexts. Single-cell transcriptomics has enabled the creation of developmental cell atlases--detailed catalogs of gene expression patterns and differentiation trajectories at an organismal scale. The widespread availability of cell atlases across metazoan model organisms presents an opportunity to construct global theories of cell-state control. In this thesis, we introduce a framework that uses persistent homology to decompose cell atlases into topological structures that provide signatures of gene regulation at the scale of an organism. Using this framework, we found that the topological structure of a broad set of developmental atlases contains only a discrete set of topological structures\u2014such as clusters, trees, and loops\u2014-revealing the recurrent use of global gene regulatory strategies. Our analysis revealed that the tree topology, while predominant, is not universal. Indeed, we identified non-trivial topologies containing loops in the development of human immune cells, seam-hypodermal cells in \\textit{C. elegans}, and the cnidocytes of multiple cnidarians. Analysis of cell-state manifolds with non-trivial topology demonstrated an important role of convergent structures in increasing cellular diversity along paths to a common cell fate, and of cyclic structures in self-renewal of progenitor-like states. Together, this work provides a global perspective on principles of cell-state regulation, and suggests that loops are important organizing structures for controlling cell differentiation.", "doi": "10.7907/t8hc-yq15", "publication_date": "2025", "thesis_type": "phd", "thesis_year": "2025" }, { "id": "thesis:16525", "collection": "thesis", "collection_id": "16525", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:06152024-132652470", "primary_object_url": { "basename": "Wang_Zitong_2025.pdf", "content": "final", "filesize": 14747054, "license": "other", "mime_type": "application/pdf", "url": "/16525/2/Wang_Zitong_2025.pdf", "version": "v3.0.0" }, "type": "thesis", "title": "Theoretical and Computational Analysis of Cell Migration in Complex Tissue Environments", "author": [ { "family_name": "Wang", "given_name": "Zitong (Jerry)", "orcid": "0000-0001-8008-7318", "clpid": "Wang-Zitong-Jerry" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Cai", "given_name": "Long", "orcid": "0000-0002-7154-5361", "clpid": "Cai-Long" }, { "family_name": "Eberhardt", "given_name": "Frederick", "clpid": "Eberhardt-Frederick" }, { "family_name": "Merchant", "given_name": "Akil Abid", "orcid": "0000-0001-7472-822X", "clpid": "Merchant-Akil-Abid" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "Cells sense and respond in spatially structured environments, including soils and tissue. My Ph.D. projects centered on developing new theoretical models and computational methods to understand how cells migrate in complex environments.
\r\n \r\nThe first project is more theoretical in nature, leveraging information theory to study how the spatial organization of cell signaling pathways are adapted to the cell's natural environment. In tissue and soil, cells must localize to their targets by navigating distributions of extracellular ligands that are spatially discontinuous, consisting of local concentration peaks, due to binding a non-uniform network of ECM fibers. It is unclear how cells navigate patchy environments while not getting trapped in local concentration peaks. To answer this question, we framed navigation as a problem of maximizing mutual information in space and developed a computational algorithm for computing signaling pathway architectures that maximize mutual information in simulated natural environments. We found that for cells in tissues and soils, dynamic localization of membrane receptors dramatically boosts sensing precision and enables cells to navigate to chemical sources 30 times faster, but this receptor localization strategy is relatively inconsequential for cells in purely diffusive environments. Further, we found that anisotropic receptor dynamics previously observed in immune cells and growth cones are nearly optimal as predicted by our model.
\r\n\r\nThe second project is more computational in nature, leveraging multiplexed tissue imaging to understand T-cell migration in tumor microenvironments. Immunotherapies can halt or slow down cancer progression by activating either endogenous or engineered T-cells to detect and kill cancer cells. T-cells must infiltrate the tumor core for immunotherapies to be effective. However, many solid tumors resist T-cell infiltration, challenging the efficacy of current therapies. In collaboration with clinician scientists at Cedars-Sinai Medical Center, we developed an integrated deep learning framework, Morpheus, that takes large-scale spatial omics profiles of patient tumors, and combines a formulation of T-cell infiltration prediction as a self-supervised machine learning problem with a counterfactual optimization strategy to generate minimal tumor perturbations predicted to boost T-cell infiltration. We applied Morpheus to 368 metastatic melanoma and colorectal cancer samples assayed using 40-plex imaging mass cytometry, discovering cohort-dependent, combinatorial perturbations, involving CXCL9, CXCL10, CCL22 and CCL18 for melanoma and CXCR4, PD-1, PD-L1 and CYR61 for colorectal cancer, predicted to support T-cell infiltration across large patient cohorts. Using only raw image data, Morpheus also identified distinct therapeutic strategies for different patient strata such as cancer stage or fatty liver presence. Our work presents a paradigm for counterfactual-based prediction and design of cancer therapeutics using spatial omics data.
", "doi": "10.7907/mj08-b258", "publication_date": "2025", "thesis_type": "phd", "thesis_year": "2025" }, { "id": "thesis:16533", "collection": "thesis", "collection_id": "16533", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:07052024-170119371", "primary_object_url": { "basename": "Thesis.pdf", "content": "final", "filesize": 43525786, "license": "other", "mime_type": "application/pdf", "url": "/16533/1/Thesis.pdf", "version": "v4.0.0" }, "type": "thesis", "title": "Active Acquisition Methods for Single Cell Genomics", "author": [ { "family_name": "Chen", "given_name": "Xiaoqiao", "orcid": "0000-0003-4685-3466", "clpid": "Chen-Xiaoqiao" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Cai", "given_name": "Long", "orcid": "0000-0002-7154-5361", "clpid": "Cai-Long" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" }, { "family_name": "Yue", "given_name": "Yisong", "orcid": "0000-0001-9127-1989", "clpid": "Yue-Yisong" }, { "family_name": "Bouman", "given_name": "Katherine L.", "orcid": "0000-0003-0077-4367", "clpid": "Bouman-K-L" } ], "local_group": [ { "literal": "div_eng" } ], "abstract": "We introduce two novel computational methodologies, ActiveSVM and Active Cell Inference, aimed at reducing the costs and enhancing the efficiency of single-cell mRNA sequencing and spatial transcriptomics, respectively. ActiveSVM employs an active learning approach to identify minimal yet highly informative gene sets for cell-type classification, physiological state identification, and genetic perturbation responses in single-cell datasets. By focusing on misclassified cells through an iterative process, ActiveSVM efficiently scales to analyze over a million cells, demonstrating around 90% accuracy across various datasets, including cell atlas and disease characterization studies.
\r\n\r\nActive Cell Inference complements this by utilizing ordered gene sets, developed through ActiveSVM, to streamline spatial genomics measurements. This end-to-end pipeline significantly reduces measurement time and costs by up to 100-fold in scientific and clinical settings. It optimizes the gene probing process by identifying well-classified cells early, allowing for targeted gene application based on cell classification certainty. This method's efficacy is further enhanced by a temporal scaling calibration scheme, improving calibration accuracy throughout its iterative process.
\r\n\r\nBoth methodologies were rigorously tested on the expansive Human Cell Atlas dataset, using the advanced computational tool, CellxGene-Census, involving over 60 million cells. This integration facilitated the creation of precise gene sets for various human tissues, dramatically improving the efficiency and reliability of these cutting-edge genomic techniques. Together, ActiveSVM and Active Cell Inference represent significant advancements in the application of genomics to clinical diagnostics, therapeutic discovery, and genetic screens, promising substantial reductions in the operational complexities and costs associated with next-generation sequencing technologies.
", "doi": "10.7907/nsn8-nd79", "publication_date": "2025", "thesis_type": "phd", "thesis_year": "2025" }, { "id": "thesis:17326", "collection": "thesis", "collection_id": "17326", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:05312025-073148050", "primary_object_url": { "basename": "Caltech_Thesis___Alec_Lourenc\u0327o-1.pdf", "content": "final", "filesize": 9832833, "license": "other", "mime_type": "application/pdf", "url": "/17326/2/Caltech_Thesis___Alec_Lourenc\u0327o-1.pdf", "version": "v5.0.0" }, "type": "thesis", "title": "Building Closed-Loop Frameworks for AI-Guided Protein Design", "author": [ { "family_name": "Louren\u00e7o", "given_name": "Alexandre Luiz", "orcid": "0009-0005-0758-2968", "clpid": "Louren\u00e7o-Alexandre-Luiz" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Mayo", "given_name": "Stephen L.", "orcid": "0000-0002-9785-5018", "clpid": "Mayo-S-L" }, { "family_name": "Zinn", "given_name": "Kai George", "orcid": "0000-0002-6706-5605", "clpid": "Zinn-K-G" }, { "family_name": "Bjorkman", "given_name": "Pamela J.", "orcid": "0000-0002-2277-3990", "clpid": "Bjorkman-P-J" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "The design of proteins with tailored properties remains a central challenge in protein engineering, with profound implications for therapeutics, sustainable manufacturing, and environmental remediation. Recent advances in artificial intelligence have dramatically improved our ability to design novel proteins, yet the precision required for many applications remains elusive. This thesis details the development and implementation of closed-loop frameworks that integrate AI-guided protein design with quantitative experimental data to iteratively improve design outcomes.
\r\n\r\nFirst, I present Protein CREATE (Computational Redesign via an Experiment-Augmented Training Engine), a high-throughput platform that combines phage display with molecular counting techniques to generate quantitative binding data at scale. This platform enables rapid evaluation of thousands (and is in the process of being scaled to millions) of designed protein variants against multiple targets simultaneously.
\r\n\r\nIn subsequent chapters, I explore two separate strands of protein design as they reach for each other to close the loop. One thread focuses on collecting data on binders I engineered to the interleukin 7 receptor alpha (IL7RA) and Insulin receptor while the other investigates the value data, even when limited, adds to improve the design process of enzymes to solve a pressing environmental remediation problem: cleaning up per and polyfluoroalkyl substances (PFAS).
\r\n\r\nWhile all of the targets discussed so far have benefited from developments in artificial intelligence, I explore one target where the benefits are limited, the human sweet taste receptor. Here, I leverage alternative computational methods coupled to experimental testing to chart a course for design.
\r\n\r\nFinally, I discuss the technologies we are integrating within the Protein CREATE framework to enable rapid in vitro and in vivo testing.
\r\n\r\nThroughout my PhD, I have been bringing the two threads of computational design and experimental characterization closer together for not only theoretically interesting, but also practically relevant, engineering cases. The methodologies developed here represent a significant advancement in our ability to design proteins with precisely tailored properties for diverse applications.
", "doi": "10.7907/8can-jz97", "publication_date": "2025", "thesis_type": "phd", "thesis_year": "2025" }, { "id": "thesis:16459", "collection": "thesis", "collection_id": "16459", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:06012024-054725051", "primary_object_url": { "basename": "240531_PB_thesis_final.pdf", "content": "final", "filesize": 44817586, "license": "other", "mime_type": "application/pdf", "url": "/16459/1/240531_PB_thesis_final.pdf", "version": "v4.0.0" }, "type": "thesis", "title": "Modeling and Design of Synthetic Biochemical Circuits for Biological Phenotypes", "author": [ { "family_name": "Bhamidipati", "given_name": "Pranav Subramanyam", "orcid": "0000-0002-6199-6505", "clpid": "Bhamidipati-Pranav-Subramanyam" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Elowitz", "given_name": "Michael B.", "orcid": "0000-0002-1221-0967", "clpid": "Elowitz-M-B" }, { "family_name": "Bois", "given_name": "Justin S.", "orcid": "0000-0001-7137-8746", "clpid": "Bois-J-S" }, { "family_name": "Barr", "given_name": "Alan H.", "clpid": "Barr-A-H" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "Biological behaviors arise from the dynamical interactions of biochemical networks. For example, the various immune responses to damage are manifestations of signaling networks between immune cell types. A central goal in systems and synthetic biology is to elucidate the design principles of these networks, or circuits, both in the sense of dissecting how function arises from structure in the natural context and in the sense of understanding the guidelines for optimal engineering of synthetic biological systems. The study of design principles in both senses is aided by mathematical modeling and simulation, which provide a self-consistent framework for evaluating the theoretical implications of biological hypotheses as well as a testbed for the development of novel circuits for desired biological phenotypes. This thesis pertains to two related challenges in this field, namely the scaling of computational design to larger circuits and the engineering of global phenotypes that emerge nonlinearly from local interactions.
\r\n \r\nThe first section of this thesis presents a novel design platform for biological circuits, called CircuiTree, that uses a game-playing paradigm to overcome the combinatorial complexity of \\textit{de novo} circuit design. This platform treats circuit design as a game of circuit assembly and traverses the tree of possible assemblies using Monte Carlo tree search (MCTS). Borrowed from artificial intelligence (AI) agents that have mastered complex games, MCTS is a reinforcement learning (RL)-based search algorithm that efficiently searches for the most effective design strategies and naturally discovers design principles in the form of network motifs, which appear as clusters of solutions in the search tree. Finally, when tasked with designing fault-tolerant oscillators with five components, CircuiTree finds a novel design strategy, which we call motif multiplexing, in which multiple sub-oscillators are interleaved so as to render the circuit highly resistant to deletions and knockdowns. This design principle, which may be responsible for the multiple oscillatory loops observed in eukaryotic circadian clocks, opens the possibility of engineering synthetic circuits at a larger scale and suggests that larger biological circuits contain yet-unknown design features that are not simply extensions of smaller circuits.
\r\n\r\nThe second section describes a novel mechanosensitive property of the SynNotch synthetic chimeric receptor and uses a multicellular modeling framework to show how it can be used to control spatiotemporal patterning \\textit{in vitro}. Modified from the endogenous juxtacrine receptor Notch, SynNotch binds to an arbitrary extracellular ligand and, in response, releases an arbitrary transcription factor, thus acting as a user-defined signal transducer. We show that, in mouse fibroblasts, a simple sender-receiver SynNotch circuit ceases to transduce a membrane-bound GFP signal at high cell densities in 2D culture. Because of this feature, a lawn of cells expressing a signal-relay circuit, which we call the transceiver circuit, can undergo spatially limited activation, where the signal propagates in a wave outward from a GFP-expressing sender cell until, due to cell division, the cell density crosses a threshold value and the signaling system shuts down. Using a multicellular lattice-based model combined with experiments, we demonstrate that perturbations of growth parameters can be used to control the size of activated spots. Finally, we achieve spatiotemporal patterns of activation by seeding the growth dish nonuniformly, creating a wave of activation at the millimeter scale that recapitulates the kinematic wave patterning phenomenon observed during vertebrate somitogenesis.
\r\n\r\nTogether, this body of work represents an advance in the use of computational methods and mathematical modeling to guide the design and control of complex biological phenotypes. Advances in these methods promise to catalyze the development of more advanced cell-based therapies and engineered tissues.
", "doi": "10.7907/gpc6-hb40", "publication_date": "2024-06-14", "thesis_type": "phd", "thesis_year": "2024" }, { "id": "thesis:16486", "collection": "thesis", "collection_id": "16486", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:06032024-182223499", "primary_object_url": { "basename": "Thesis_Draft_final_final.pdf", "content": "final", "filesize": 21944874, "license": "other", "mime_type": "application/pdf", "url": "/16486/1/Thesis_Draft_final_final.pdf", "version": "v4.0.0" }, "type": "thesis", "title": "Revealing Regulatory Network Organization Through Single-Cell Perturbation Profiling and Maximum Entropy Models", "author": [ { "family_name": "Jiang", "given_name": "Jialong", "orcid": "0000-0001-8560-8397", "clpid": "Jiang-Jialong" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Elowitz", "given_name": "Michael B.", "orcid": "0000-0002-1221-0967", "clpid": "Elowitz-M-B" }, { "family_name": "Phillips", "given_name": "Robert B.", "orcid": "0000-0003-3082-2809", "clpid": "Phillips-R" }, { "family_name": "Pachter", "given_name": "Lior S.", "orcid": "0000-0002-9164-6231", "clpid": "Pachter-L" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "Gene regulatory networks within cells modulate the expression of the genome in response to signals and changing environmental conditions. Reconstructions of gene regulatory networks can reveal the information processing and control principles used by cells to maintain homeostasis and execute cell-state transitions. In this thesis, we introduce a computational framework, D-SPIN, that generates quantitative models of gene regulatory networks from single-cell mRNA-seq datasets collected across thousands of distinct perturbation conditions. D-SPIN models the cell as a collection of interacting gene-expression programs, and constructs a probabilistic model to infer regulatory interactions between gene-expression programs and external perturbations. Using large Perturb-seq and drug-response datasets, we demonstrate that D-SPIN models reveal the organization of cellular pathways, sub-functions of macromolecular complexes, and the logic of cellular regulation of transcription, translation, metabolism, and protein degradation in response to gene knockdown perturbations. D-SPIN can also be applied to dissect drug response mechanisms in heterogeneous cell populations, elucidating how combinations of immunomodulatory drugs can induce novel cell states through additive recruitment of gene expression programs. D-SPIN provides a computational framework for constructing interpretable models of gene-regulatory networks to reveal principles of cellular information processing and physiological control.", "doi": "10.7907/5zta-9818", "publication_date": "2024", "thesis_type": "phd", "thesis_year": "2024" }, { "id": "thesis:15123", "collection": "thesis", "collection_id": "15123", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:03172023-050019811", "primary_object_url": { "basename": "Guru_PhD_thesis_v2.pdf", "content": "final", "filesize": 16253785, "license": "other", "mime_type": "application/pdf", "url": "/15123/1/Guru_PhD_thesis_v2.pdf", "version": "v5.0.0" }, "type": "thesis", "title": "Engineering Artificial Systems with Natural Intelligence", "author": [ { "family_name": "Raghavan", "given_name": "Guruprasad", "orcid": "0000-0002-1970-9963", "clpid": "Raghavan-Guruprasad" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Winfree", "given_name": "Erik", "orcid": "0000-0002-5899-7523", "clpid": "Winfree-E" }, { "family_name": "Rutishauser", "given_name": "Ueli", "orcid": "0000-0002-9207-7069", "clpid": "Rutishauser-U" }, { "family_name": "Lois", "given_name": "Carlos", "orcid": "0000-0002-7305-2317", "clpid": "Lois-Carlos" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "Although Deep neural networks achieve human-like performance on a variety of perceptual and decision-making tasks, they perform poorly when confronted with changing tasks or goals, and broadly fail to match the flexibility and robustness of human intelligence. Additionally, artificial neural networks rely heavily on human-designed, hand-programmed architectures for their remarkable performance. In this thesis, I work towards achieving two goals: (i) development of a set of mathematical frameworks inspired by facets of natural intelligence, to endow artificial networks with flexibility and robustness, two key traits of natural intelligence; and (ii) inspired by the development of the biological vision system, I propose an algorithm that can \u2018grow\u2019 a functional, layered neural network from a single initial cell, with the aim of enabling autonomous development of artificial networks akin to living neural networks.
\r\n\r\nFor the first goal of endowing networks with flexibility and robustness, I propose a mathematical framework to enable continuous training of neural networks on a range of objectives by constructing path connected sets of networks, resulting in the discovery of a series of networks with equivalent functional performance on a given machine learning task. In this framework, I view the weight space of a neural network as a curved Riemannian manifold and move a network along a functionally invariant path in weight space while searching for networks that satisfy secondary objectives. A path-sampling algorithm trains computer vision and natural language processing networks with millions of weight parameters to learn a series of classification tasks without performance loss while accommodating secondary objectives including network sparsification, incremental task learning, and increased adversarial robustness. Broadly, for achieving this goal, I conceptualize a neural network as a mathematical object that can be iteratively transformed into distinct configurations by the path- sampling algorithm to define a sub-manifold of networks that can be harnessed to achieve user goals.
\r\n\r\nFor the second goal of \u2018growing\u2019 artificial neural networks in a manner similar to living neural networks, I develop an approach inspired by the mechanisms employed by the early visual system to wire the retina to the lateral geniculate nucleus (LGN), days before animals open their eyes. I find that the key ingredients for robust self- organization are (a) an emergent spontaneous spatiotemporal activity wave in the first layer and (b) a local learning rule in the second layer that \u2018learns\u2019 the underlying activity pattern in the first layer. As the bio-inspired developmental rule is adapt- able to a wide-range of input-layer geometries and robust to malfunctioning units in the first layer, it can be used to successfully grow and self-organize pooling architectures of different pool-sizes and shapes. The algorithm provides a primitive procedure for constructing layered neural networks through growth and self-organization. Finally, I also demonstrate that networks grown from a single unit perform as well as hand-crafted networks on a wide variety of static (MNIST recognition) and dynamic (gesture-recognition) tasks. Broadly, the work in the second section of this thesis shows that biologically inspired developmental algorithms can be applied to autonomously grow functional \u2018brains\u2019 in-silico.
", "doi": "10.7907/374f-1202", "publication_date": "2023", "thesis_type": "phd", "thesis_year": "2023" }, { "id": "thesis:14496", "collection": "thesis", "collection_id": "14496", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:02132022-064810187", "primary_object_url": { "basename": "David_Brown_Thesis_V4.pdf", "content": "final", "filesize": 10109719, "license": "other", "mime_type": "application/pdf", "url": "/14496/1/David_Brown_Thesis_V4.pdf", "version": "v5.0.0" }, "type": "thesis", "title": "Principles of Massively Parallel Sequencing for Engineering and Characterizing Gene Delivery", "author": [ { "family_name": "Brown", "given_name": "David", "orcid": "0000-0002-9757-1744", "clpid": "Brown-David" } ], "thesis_advisor": [ { "family_name": "Gradinaru", "given_name": "Viviana", "orcid": "0000-0001-5868-348X", "clpid": "Gradinaru-V" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Yue", "given_name": "Yisong", "orcid": "0000-0001-9127-1989", "clpid": "Yue-Yisong" }, { "family_name": "Arnold", "given_name": "Frances Hamilton", "orcid": "0000-0002-4027-364X", "clpid": "Arnold-F-H" }, { "family_name": "Gradinaru", "given_name": "Viviana", "orcid": "0000-0001-5868-348X", "clpid": "Gradinaru-V" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "The advent of massively parallel sequencing and synthesis technologies have ushered in a new paradigm of biology, where high throughput screening of billions of nucleid acid molecules and production of libraries of millions of genetic mutants are now routine in labs and clinics. During my Ph.D., I worked to develop data analysis and experimental methods that take advantage of the scale of this data, while making the minimal assumptions necessary for deriving value from their application. My Ph.D. work began with the development of software and principles for analyzing deep mutational scanning data of libraries of engineered AAV capsids. By looking at not only the top variant in a round of directed evolution, but instead a broad distribution of the variants and their phenotypes, we were able to identify AAV variants with enhanced ability to transduce specific cells in the brain after intravenous injection. I then shifted to better understand the phenotypic profile of these engineered variants. To that end, I turned to single-cell RNA sequencing to seek to identify, with high resolution, the delivery profile of these variants in all cell types present in the cortex of a mouse brain. I began by developing infrastructure and tools for dealing with the data analysis demands of these experiments. Then, by delivering an engineered variant to the animal, I was able to use the single-cell RNA sequencing profile, coupled with a sequencing readout of the delivered genetic cargo present in each cell type, to define the variant\u2019s tropism across the full spectrum of cell types in a single step. To increase the throughput of this experimental paradigm, I then worked to develop a multiplexing strategy for delivering up to 7 engineered variants in a single animal, and obtain the same high resolution readout for each variant in a single experiment. Finally, to take a step towards translation to human diagnostics, I leveraged the tools I built for scaling single-cell RNA sequencing studies and worked to develop a protocol for obtaining single-cell immune profiles of low volumes of self-collected blood. This study enabled repeat sampling in a short period of time, and revealed an incredible richness in individual variability and time-of-day dependence of human immune gene expression. Together, my Ph.D. work provides strategies for employing massively parallel sequencing and synthesis for new biological applications, and builds towards a future paradigm where personalized, high-resolution sequencing might be coupled with modular, customized gene therapy delivery.
", "doi": "10.7907/yqjm-6609", "publication_date": "2022", "thesis_type": "phd", "thesis_year": "2022" }, { "id": "thesis:14399", "collection": "thesis", "collection_id": "14399", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:10172021-215439860", "primary_object_url": { "basename": "Dobreva_Tatyana_2021_v7.pdf", "content": "final", "filesize": 10892340, "license": "other", "mime_type": "application/pdf", "url": "/14399/1/Dobreva_Tatyana_2021_v7.pdf", "version": "v4.0.0" }, "type": "thesis", "title": "Engineering Tools to Probe and Manipulate the Immune System at Single-Cell Resolution", "author": [ { "family_name": "Dobreva", "given_name": "Tatyana", "orcid": "0000-0002-2625-8873", "clpid": "Dobreva-Tatyana" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" }, { "family_name": "Gradinaru", "given_name": "Viviana", "orcid": "0000-0001-5868-348X", "clpid": "Gradinaru-V" } ], "thesis_committee": [ { "family_name": "Gao", "given_name": "Wei", "orcid": "0000-0002-8503-4562", "clpid": "Gao-Wei" }, { "family_name": "Gradinaru", "given_name": "Viviana", "orcid": "0000-0001-5868-348X", "clpid": "Gradinaru-V" }, { "family_name": "Elowitz", "given_name": "Michael B.", "orcid": "0000-0002-1221-0967", "clpid": "Elowitz-M-B" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_eng" } ], "abstract": "My thesis focuses on developing experimental and computational tools to probe and manipulate cellular transcriptomes in the context of human health and disease. Chapter 1 and 2 focus on published work where we leverage single-cell RNA sequencing (scRNA-seq) to understand human immune variability, characterize cell-type specific biases of multiple viral variants within an animal, and assess temporal immune response in the brain to delivery of genetic cargo via an adeno-associated virus (AAV). Chapter 3 and 4 present progress I have made on tools for exporting RNA extracellularly and engineering of a transcription factor for modulating macrophage state.
\r\n\r\nFor probing cellular transcriptome states, we have developed a platform using multiplexed single-cell sequencing and out-of-clinic capillary blood extraction to understand temporal and inter-individual variability of gene expression within immune cell types. Our platform enables simplified, cost-effective profiling of the human immune system across subjects and time at single-cell resolution. To demonstrate the power of our platform, we performed a three day time-of-day study of four healthy individuals, generating gene expression data for 24,087 cells across 22 samples. We detected genes with cell type-specific time-of-day expression and identified robust genes and pathways particular to each individual, all of which could have been missed if analyzed with bulk RNA-sequencing. Also, using scRNA-seq, we have developed a method to screen and characterize cellular tropism of multiple AAV variants. Additionally, I have looked at AAV-mediated transcriptomic changes in animals injected with AAV-PHP.eB three days and twenty-five days post-injection. I have found that there is an upregulation of genes involved in p53 signaling in endothelial cells three days post-injection.
\r\n\r\nIn the context of manipulating cellular transcriptomic states, I demonstrate that a fusion between RNA targeting enzyme, dCas13, and capsid-forming neuronal protein, Arc, is able to form a capsid-like structure capable of encapsulating RNA. I also present methods and preliminary data for tuning macrophage states through mutations in transcription factor EB (TFEB) using scRNA-seq as a readout.
", "doi": "10.7907/n3rs-ft69", "publication_date": "2022", "thesis_type": "phd", "thesis_year": "2022" }, { "id": "thesis:13838", "collection": "thesis", "collection_id": "13838", "cite_using_url": "https://resolver.caltech.edu/CaltechTHESIS:07082020-113341068", "type": "thesis", "title": "Guiding Self-Organization in Active Matter with Spatiotemporal Boundary Conditions", "author": [ { "family_name": "Ross", "given_name": "Tyler David", "orcid": "0000-0002-7872-3992", "clpid": "Ross-Tyler-David" } ], "thesis_advisor": [ { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "thesis_committee": [ { "family_name": "Winfree", "given_name": "Erik", "orcid": "0000-0002-5899-7523", "clpid": "Winfree-E" }, { "family_name": "Rothemund", "given_name": "Paul W. K.", "orcid": "0000-0002-1653-3202", "clpid": "Rothemund-P-W-K" }, { "family_name": "Qian", "given_name": "Lulu", "orcid": "0000-0003-4115-2409", "clpid": "Qian-Lulu" }, { "family_name": "Phillips", "given_name": "Robert B.", "orcid": "0000-0003-3082-2809", "clpid": "Phillips-R" }, { "family_name": "Brady", "given_name": "John F.", "orcid": "0000-0001-5817-9128", "clpid": "Brady-J-F" }, { "family_name": "Shapiro", "given_name": "Mikhail G.", "orcid": "0000-0002-0291-4215", "clpid": "Shapiro-M-G" }, { "family_name": "Thomson", "given_name": "Matthew", "orcid": "0000-0003-1021-1234", "clpid": "Thomson-M-W" } ], "local_group": [ { "literal": "div_bbe" } ], "abstract": "In this thesis, I demonstrate that self-organized structures and forces can be guided by modulating the interactions between force-generating molecules in space and time. The physics of self-organizing systems is an open frontier. We do not have a complete set of principles that can describe how a dynamic structure forms based on the non-equilibrium dynamics of its constituent components. Yet, living systems appear to depend on some set of rules of self-organization in order to reliably carry out their mechanical functions. Force-generating, active, molecules in the form of motor proteins and filamentous polymers are responsible for performing fundamental tasks in living matter, such as locomotion and division. While it is known that the regulation of motor-filament interactions is necessary to achieve the dynamic structures that drive movement and propagation, the role of spatial and temporal patterning in self-organizing systems has not been explored. I design a artificial system of purified molecules where the interactions between motors and filaments are toggled with light. By patterning molecular interactions in space and time, I show that it is possible to localize the formation of spherically symmetric asters, which can be moved, merged, and used to generate advective fluid flows. The ability to pattern molecular interactions in space and time offers a new perspective in the search for principles of active self-organization. Spatial and temporal control makes it possible to start distilling how the interactions between active molecules determine the mesoscopic behaviors of self-organized structures. These rules ultimately govern the physics of living matter and may eventually be harnessed to build new materials and cell-like machines.
", "doi": "10.7907/q85h-j730", "publication_date": "2021", "thesis_type": "phd", "thesis_year": "2021" } ]