How does a system change its behavior?

The Question

An enzyme can look almost identical in sequence and behave very differently after a catalytic-site mutation. How can a model reflect that difference without implying more numerical precision than the evidence supports?

My Contribution

I developed CatRange with collaborators, connecting biochemical data curation, enzyme–substrate representations, model development, evaluation, and usable inference.

• Protein representations
• Substrate chemistry
• Gradient-boosted classification
• Mutation-aware evaluation
• Research inference

What the record shows

The PNAS Nexus publication and public repository provide the research and software record. The contribution is a range-based, mutation-sensitive modeling framework—not an experimental measurement of every predicted variant.

What this does—and does not—establish

A predicted regime is not a measured kinetic constant. Neighboring-bin recovery is an evaluation criterion, not automatically a calibrated confidence interval. Assay context, sequence coverage, and the deployed bin definitions still matter.

The Question

Rhodopseudomonas palustris reorganizes its proteome across lignin-derived substrates and oxygen conditions. Which patterns remain informative about growth across those different environments?

My Contribution

I developed CorePredX to connect quantitative proteomics with growth prediction and dependence-aware interpretation. My work spans the computational question, methodology, software, analysis, validation, and scientific communication.

• Quantitative proteomics
• Neural prediction
• SHAP attribution
• Redundancy analysis
• Cross-condition interpretation

What the record shows

The mSystems study identifies a compact hierarchy of candidate growth-associated proteins beneath broader proteome remodeling, motivating more focused biological follow-up.

What this does—and does not—establish

Predictive importance does not establish a causal regulator or a directed regulatory edge. Correlated features, the sampled environments, and independent experimental validation constrain the interpretation.

The Question

How do you quantify transient metabolic fluxes when the experimental data is sparse and noisy?

My Contribution

r-DMFA framework, 15N isotope-labeling, targeted metabolomics, dynamic flux sampling. Identified SBH, LCBK control nodes.

• Isotope labeling integration
• Dynamic flux analysis
• Uncertainty quantification
• Control point identification
• Arabidopsis metabolism

What the record shows

iScience publication demonstrates flux identification under data variability.

What this does—and does not—establish

Point estimates of fluxes carry uncertainty; validated against biological priors, not independent kinetic measurements.

The Question

Can early cytokine and gene-expression patterns predict vaccine-induced antibody responses at later timepoints?

My Contribution

Built multi-horizon ensemble integrating cytokine, transcriptomic, and antibody data.

What the record shows

Demonstrates predictive utility of early-stage multi-omics integration in vaccine response forecasting.

What this does—and does not—establish

Identifies predictive biomarkers but does not directly prove mechanistic causality without further biological validation.

The Question

What metabolic pathways do pancreatic cancer cells activate when simultaneously stressed by low oxygen, acidic pH, and nutrient deprivation?

My Contribution

Building temporally resolved metabolic models, analyzing PDAC RNA-seq.

What the record shows

Reveals shifting dependencies across multiple simultaneous stress axes, informing context-specific vulnerabilities.

What this does—and does not—establish

Models suggest potential therapeutic targets based on transcriptomic shifts, which require subsequent empirical targeting to confirm essentiality.

The Question

How do you design a controller that handles nonlinear behavior, model uncertainty, and real-time constraints simultaneously?

My Contribution

Designed/fabricated cascaded three-tank system, implemented NMPC, MHE, EKF, economic MPC. Evaluated ANN-based predictive control.

What the record shows

Demonstrated real-time constraint satisfaction and setpoint tracking in physical hardware under disturbances.

What this does—and does not—establish

Validates controller architectures on specific pilot-scale hardware, not universal bounds on stability under arbitrary noise.

The Question

If a greenhouse has 56 sensor locations, how do you find the minimum set that captures meaningful variation without redundant monitoring?

My Contribution

Online K-Means++ clustering, psychrometric feature construction, multi-season validation.

What the record shows

Demonstrates reliable microclimate estimation from a reduced sensor footprint, lowering deployment costs.

What this does—and does not—establish

Optimizes sensor placement for specific geometries and seasons; re-calibration may be necessary for structural changes.

The Question

How can an AI assistant help a researcher find, evaluate, and act on biochemical evidence about enzyme mutations?

My Contribution

Training/fine-tuning LLMs with biochemical evidence. Building research prototype.

What the record shows

Early integration of biochemical heuristics and literature context inside an accessible chat agent.

What this does—and does not—establish

Provides augmented literature exploration, not a substitute for human scientific judgment or rigorous simulation.