Where I want the research to go.

Finding the biological state worth changing.

I am interested in target identification that moves beyond a static protein name or an expression measurement. A disease-associated state may be defined by conformation, interacting partners, location within a cell, trafficking behaviour or its timing during disease progression. This leads naturally to chemical biology, allosteric modulators, induced-proximity pharmacology, PROTACs and molecular-glue degraders, and phenotypic discovery. I want to learn how click chemistry and bioorthogonal ligation can be used to construct or interrogate degraders, probes and ligand conjugates, while separately testing binding, target engagement, degradation selectivity and functional rescue. The important experiment is the one that connects target engagement to a change in disease-relevant function.

Possible tools for doctoral training: medicinal and click chemistry, structure–activity relationships, chemical probes, target-engagement assays, proteomic selectivity measurements, genetic perturbation, live-cell measurements and mechanistic modelling.

Designing the intervention and its route together.

I want to investigate molecular recognition alongside the chemistry that determines tissue access and intracellular delivery. In nucleic-acid formulations that means measuring uptake and endosomal release as well as functional activity; in antibody–drug conjugates it means examining antigen recognition together with linker and payload behaviour. Aptamer conjugates, lipid and polymeric nanocarriers, responsive biomaterials and engineered cell therapies raise related questions, although their experimental endpoints differ substantially. I am interested in aptamer selection and stabilization, site-selective conjugation, RNA and small-molecule cargo, controlled release, and microfluidic preparation of delivery systems when it enables reproducible manufacturing or precise formulation control.

Therapeutic engineering can also mean modifying a cell or its immediate environment. Cell-surface-bound drug depots, engineered macrophages and other immune cells, and controlled cytokine release offer ways to steer inflammatory or tissue-repair responses. For example, cellular backpacks for macrophage immunotherapy and backpack-assisted myeloid cells in a preclinical multiple-sclerosis model illustrate how particle engineering and cell-state control can be studied together. I am interested in the underlying methods of cell manipulation, signalling measurement and local therapeutic action, not in requiring a particular carrier design.

Possible tools for doctoral training: bioorthogonal bioconjugation, nucleic-acid chemistry, microfluidic formulation, nanocarrier characterization, uptake and endosomal-release assays, receptor trafficking, cellular pharmacology, quantitative PK/PD and biocompatibility tests.

Building models of disease before its late manifestations.

I am particularly interested in how a healthy tissue becomes susceptible to disease and how that transition might be studied before an irreversible endpoint is reached. Patient-derived organoids, co-cultures, tissue engineering and microfluidic organs-on-chips can introduce interactions absent from conventional two-dimensional assays. Engineering methods such as controlled perfusion, shear stress, barrier transport and tissue-interface measurements can help explain why a molecularly promising therapy fails in a more realistic biological environment. In pulmonary or renal models, flow, mechanical loading, barrier transport and inflammatory signalling can be measured under controlled conditions. In oncology, tumour organoids and engineered immune microenvironments could help investigate heterogeneous response and therapeutic resistance.

I want such models to reduce animal use where they can answer the question reliably, and eventually replace particular animal experiments when fit-for-purpose evidence justifies that substitution. An organoid is not automatically more predictive because it is three-dimensional; maturity, vascular or immune components, reproducibility and clinically relevant exposure must be demonstrated.

Possible tools: microphysiological systems, organoids, 3D bioprinting, longitudinal imaging, tissue function assays and systematic model qualification.

Making experiments and computational models correct one another.

I want to develop computational tools that follow more of a therapeutic candidate's life: from molecular properties and formulation through administration, transport, metabolism, target exposure and biological response. Molecular simulation, in-silico screening, physiologically based pharmacokinetics, quantitative systems pharmacology and machine learning can all contribute, but they answer different questions and carry different assumptions. The objective would be a testable model linked to measurements, with its scope and uncertainty explicit.

A valuable result would be a workflow in which a model identifies the next informative experiment, the experimental result updates the model, and uncertainty is reported alongside predictions. I am interested in this logic of design–make–test–analyse, including automation where the underlying measurements justify it.

Possible tools: scientific Python, molecular descriptors, PBPK/QSP, mechanistic simulation, imaging analytics, uncertainty analysis and human-in-the-loop closed-loop experimentation.

Which targets and candidate molecules are worth pursuing?

Integrate cellular phenotypes, genetic perturbations, pathway evidence and, where appropriate, transcriptomic or proteomic data to prioritize targets. Identify known ligands and candidate natural-product or synthetic scaffolds. Docking and molecular dynamics can inform hypotheses about recognition; biochemical target engagement, cellular assays and appropriate counter-screens are needed to test them.

Which chemical or biological route is feasible?

Trace candidate molecules to purchasable intermediates, natural precursors or metabolites. Investigate the enzymes and host organisms that can generate relevant intermediates, including pathway engineering and enzyme promiscuity where supported by evidence. Compare computer-assisted retrosynthetic routes with biocatalytic, chemoenzymatic and continuous-flow methods; measure actual yield, selectivity and process robustness.

Can screening guide the next synthesis rather than merely report a hit?

Use well-controlled automated assays to quantify dose response, target specificity and cellular phenotype. Feed those data, alongside reaction performance and analytical quality, into experimental design or active-learning models to choose the next set of compounds and conditions. A useful success criterion is prospective improvement confirmed in held-out experiments, not a higher predicted docking score.

Potential methods for future training: target deconvolution and chemical proteomics; structure-based screening; enzyme and metabolic-pathway mining; retrosynthesis planning; reaction and flow engineering; high-throughput assay development; design of experiments and Bayesian optimization.

Can the vulnerable state be identified before choosing the drug?

Take a signalling interaction involved in pathological repair or tumour invasion. Instead of ranking targets only by abundance, characterize when the relevant protein complex forms, where it resides and what changes after injury or treatment. Compare orthogonal binding measurements with genetic perturbation and functional rescue. A compound that binds convincingly but fails to alter a disease-associated state is not an encouraging hit merely because its affinity is strong. Small-molecule discovery targeting the SLIT2–ROBO axis illustrates how to connect a molecular screening programme to questions of biological function.

First possible test: time-resolved cell states, target engagement, genetic perturbation and rescue. Decisive question: does the target causally control the selected transition?

Where, exactly, does a promising therapeutic disappear?

In pulmonary RNA therapy a formulation must survive aerosol generation, deposit in the relevant region, interact with lung lining fluid, enter the intended cell and release functional cargo. I would compare delivery systems across those separate barriers in healthy and injury-conditioned models, rather than rank them by size or fluorescent uptake alone. Functional knockdown, intracellular release, barrier recovery and inflammation should be measured alongside delivered dose. Experimental work showing that endosomal escape can provoke inflammatory damage makes the design problem sharper: greater entry need not produce a safer or more effective medicine.

First possible test: matched RNA cargo and administered dose across formulations, measuring aerosol mass balance, cellular delivery, knockdown and tissue function. Decisive question: which barrier limits productive and tolerable exposure?

Why does recognition fail to become treatment?

Antibody–drug conjugates, aptamer conjugates and engineered T cells make recognition central to therapeutic design, but recognition is only the beginning. In an ADC, antigen heterogeneity, internalization, linker cleavage, payload release and bystander toxicity matter. In aptamer-directed carriers the questions include nuclease stability, receptor-dependent uptake and intracellular escape; aptamer-guided oligonucleotide delivery to muscle stem cells is a useful research example. CAR-T and other engineered immune cells confront trafficking, persistence and suppressive microenvironments; tumour-on-chip research on CAR-T action illustrates a possible experimental route. These modalities share questions about access and spatial response, not interchangeable measures of success.

First possible test: one modality and one co-culture or tissue chip, with mapped recognition sites and time-resolved functional response. Decisive question: does failure arise from recognition, access, release or tissue resistance?

Can a model predict the next experiment before seeing it?

I would first model a tractable sequence: formulation characteristics, delivered dose, tissue transport, intracellular disposition and measured pharmacological response. Mechanistic transport, PBPK, QSP or statistical models could each be used where their parameters are identifiable. I would specify predictions before collecting a held-out dataset and assess calibration and systematic error afterward. The model should be able to recommend an informative experiment, and it should also be allowed to fail.

First possible test: a modular inhalation-to-cell-response model linked to aerosol and epithelial-barrier experiments. Decisive question: does it predict unseen results better than a simpler baseline, and where does its uncertainty become unacceptable?