Defense Notices

Accurate protein characterization, precision therapeutics design, and privacy-preserving clinical modeling are important challenges in biomedical artificial intelligence (AI). Addressing these challenges requires computational frameworks that are not only accurate and scalable but also capable of supporting reliable decision-making. This dissertation develops efficient, scalable, and reliable computational frameworks across three areas: Computational characterization of proteins, precision therapeutics design, and privacy-aware clinical representation learning. The first part of this dissertation focuses on efficient protein characterization at both the sequence and residue levels. GPCR-SLM, a knowledge-distilled small protein language model, classifies G-protein-coupled receptors (GPCRs), a family targeted by approximately 35% of FDA-approved drugs. Across 86 families, GPCR-SLM achieves 99% accuracy, outperforming BLAST (86.4%) and HMMER (91%), with a 33.5-fold speedup over large protein language models. Extending characterization to the residue level, a protein-metal-ion binding site prediction framework evaluates five protein language models and incorporates positional encoding to capture long-range dependencies. Under 10-fold cross-validation, it achieves a Matthews correlation coefficient of 0.89, with precision, recall, and F1 scores exceeding 95% across six metal ions.

Building on these protein characterization methods, the second part advances precision therapeutics design through uncertainty quantification and generative learning for therapeutic antibodies. NbBayesLM combines protein language model embeddings and physicochemical descriptors in a Bayesian neural network to predict nanobody thermostability. Evaluated on 10,630 VHH sequences, it achieves a mean absolute error of 1.89C and an R2 of 0.67 while estimating epistemic and aleatoric uncertainty to support candidate prioritization. CIgFlow employs classifier-free guidance to generate de novo, antigen-conditioned antibody sequences without fixed scaffolds or paired structural data. Generated sequences achieve a median OASis humanness score of 0.90 and mean predicted local distance difference test (pLDDT) score of 85.0, demonstrating promising novelty and diversity. The third part addresses privacy-aware clinical modeling using electrocardiograms (ECGs), which encode diagnostic information and sensitive demographic attributes. A variational autoencoder with dual discriminators suppresses demographic information while preserving clinically relevant representations.

Collectively, this dissertation advances efficient protein representation learning, computational antibody design, and privacy-preserving clinical AI, contributing methodological foundations for trustworthy biomedical applications.

Mismatched filtering (MMF) is a widely used pulse compression technique that reduces cross-correlation sidelobe levels at the cost of some loss in signal-to-noise ratio (SNR). Applying MMF to stochastically generated radar waveforms makes the filter's performance inherently random, which can be problematic if the variability is excessive. As well as the computational cost being relatively high, as per pulse, a new MMF optimization is required. In this thesis, the feasibility of a fixed MMF strategy for stochastic radar waveforms to reduce computational cost is investigated. Two MMF formulations were evaluated, the optimal MMF (Diagonal Loaded Filter) and a projection-based MMF. Both filters were designed to minimize sidelobe energy within specified regions of the Cross-Ambiguity Function (CAF) for Stochastic Waveform Generation (StoWGe) waveforms, and their performance was assessed across multiple time-bandwidth products and sidelobe-regions for two center frequencies, Fc = 30 Ghz and Fc  = 3.5 Ghz. The MMF performance was characterized using SNR loss and sidelobe depth over hundreds of trails. It is observed that the variability in SNR and sidelobe depth decreases as the waveform time-bandwdith product and the sidelobe region targeted for minimization increase, although the Fc = 3.5 GHz cases exhibited a plateau in achievable sidelobe depth. Additional validation using 1000-trail histograms and loop-back testing confirmed the robustness for the fixed MMF approach for higher time-bandwidth product stochastic waveforms. Where, overall, the results demonstrate that a fixed MMF strategy is feasible for stochastic waveforms when the time-bandwdith product and sidelobe region are sufficiently large. Furthermore, the Projection-based MMF provides a suboptimal alternative to the optimal MMF as it achieves comparable MMF performance when constrained to similar SNR-loss and may be desirable when the size of the domination eigenvalues is small relative to the filter length and can be efficiency obtained.