Ph.D. Dissertation Defense: Adrian Ratwatte
Tuesday, May 26
10 AM
AVH 115
Zoom: https://unl.zoom.us/j/98424305210
"Exploring Gene Regulatory Neural Network Biocomputing of Bacteria"
Artificial Intelligence (AI) has evolved from brain-inspired algorithms into a multifaceted discipline that increasingly integrates with biological systems. Although silicon based computing platforms have achieved remarkable progress in machine learning, they remain limited in energy efficiency and applicability to environments that silicon cannot reach. AI is no longer limited to silicon chips. A new form of computing is emerging within living systems through biological processes. Biological computing offers an alternative by reducing energy consumption, enabling AI in non silicon environments, and supporting reliable computing that is efficient in resource utilization and reconfigurable across diverse tasks. This dissertation addresses these limitations by introducing a bacterial computing framework that models Gene Regulatory Networks (GRNs) as Gene Regulatory Neural Networks (GRNNs). The GRNN mirrors the structure and function of Artificial Neural Networks (ANNs) through gene-gene interactions across trans-omic layers, enabling natural, self-regulating information processing within living systems.
At the cellular scale, a dual-layered chemical reaction model was developed to describe transcription and translation dynamics, revealing the sigmoidal activation behavior of genes modeled as gene-perceptrons. This abstraction enabled the isolation of sub-GRNNs from the larger GRNN that function as non-linear classifiers, where Lyapunov-based stability analysis ensured reliable computation under concentration fluctuations. Building upon this theoretical foundation, a control-theoretic framework was introduced to determine optimal chemical input concentrations that guide the GRNN toward desired weight configurations using the Linear Quadratic Regulator (LQR) approach. This control model maintained a balance between stability and reconfigurability and demonstrated its effectiveness in mitigating Clostridioides difficile biofilm formation. Extending beyond control applications, recent work expands the GRNN framework to transform bacterial gene expression dynamics into a biocomputing library of mathematical solvers. Current biocomputing approaches mainly rely on engineered circuits with fixed logic, which limits stability and reliability under different conditions. In contrast, this work uses the GRNN framework to identify functional subnetworks directly from the native GRNs. A sub-GRNN search algorithm is developed as a general framework to identify subnetworks that match task-specific gene expression patterns under chemically encoded input codes. Mathematical calculation and classification tasks, including identifying Fibonacci numbers, prime numbers, multiplication, and Collatz step counts, are used as case studies to validate the proposed framework. The identified sub-GRNNs are evaluated using gene-wise perturbation, collective perturbation, and Lyapunov-based analysis to assess computing stability and reliability. The results show that native transcriptional dynamics can support diverse computing tasks while maintaining stable and reliable performance. Overall, this work establishes bacterial biocomputing as a viable paradigm for performing computing within living cells.
Committee:
Dr. Sasitharan Balasubramaniam, Advisor
Dr. Byrav Ramamurthy
Dr. Nirnimesh Ghose
Dr. Xu Li