Ideas in progress

Research Notes

Papers that stay with me, questions I am working through, and connections to my research in AI for precision oncology.

Four parallel sources of temporal information: differences across people, molecular and spatial signals, sparse measurements at known times, and learned patterns from real longitudinal records.

· Research perspective

Virtual cohorts and the missing disease timeline

How tissue snapshots, sparse experiments, and clinical histories support different kinds of disease trajectories, and how to choose a useful starting point.

Virtual cohorts · Disease progression · Longitudinal modeling

Molecular data, imaging, and clinical history inform an estimated state. A dynamics model uses intervention and elapsed time to predict possible trajectories; a selected experiment supplies new evidence to update the state.

· Research perspective

World models in biomedicine

A practical map of patient simulators, imaging dynamics, and cellular response models: how they work, what their data support, and how predictions can guide the next experiment.

World models · Disease dynamics · Perturbation prediction

Four stages from tissue inspection to patient representations, modality comparisons, and survival evaluation, keeping patients within one data split.

· Research perspective

From pathology slides to survival predictions

A practical map of the pathology AI stack: what each tool does, how to choose foundation models, and how the pieces become a patient-level study.

Computational pathology · Open-source tools · Foundation models

Six stages in a research cycle: measure cell states, observe perturbations, learn response models, prioritize hypotheses, test experimentally, and evaluate and refine.

· Research perspective

Closing the loop in precision oncology

A practical route from perturbation data to treatment hypotheses, informative experiments, and patient evidence.

Virtual cells · Experimental design · Digital twins