07 · Glossary

The vocabulary of the field.

Compact definitions for the core terms — biology, pharmacology, causal inference, and modeling — used across the course.

ADMET: Absorption, Distribution, Metabolism, Excretion, Toxicity — the disposition properties that determine whether a molecule can be a drug.
Attractor: A region of state space toward which the dynamics tend. Stable cell types and disease states often correspond to attractors.
Active learning: Iterative experimental design where the model selects the next experiment to maximize information about the unknown.
Allostery: Regulation of protein activity at a site distinct from the active site, often via conformational change.

Bistability: A dynamical regime with two stable states; small perturbations can switch the system between them.

Cell state: A configuration of a cell described by its molecular contents (transcripts, proteins, metabolites) and context. Cells live in a high-dimensional, partially observed state space and move between states over time.
CRISPR knockout: Editing the genome to disrupt a gene's coding sequence so the functional protein is no longer produced. Strong, persistent loss-of-function.
Causal graph: A directed graph encoding causal relationships, with semantics under intervention (do-operator).
Counterfactual: A prediction of what would have happened under an alternative intervention, holding everything else equal.
Confounder: A variable that influences both the treatment and the outcome, biasing observational estimates of causal effect.

Disease signature: A characteristic pattern (e.g. gene-expression vector) that distinguishes a disease state from a healthy reference.
Drug target: A molecular entity (often a protein) whose modulation is intended to produce a therapeutic effect.

Expected information gain: The expected reduction in posterior uncertainty from running a candidate experiment.

Feedback loop: A pathway motif where downstream output influences upstream signaling, producing amplification, adaptation, or oscillation.
Foundation model: A large model pretrained on broad data (e.g. cells, sequences, structures) and adapted to specific tasks.

Hybrid mechanistic-neural model: A model that combines known mechanistic structure (graphs, ODEs) with learned neural components for the unknown residuals.

Knockdown: Reducing the abundance of a gene's transcript or protein (e.g. via RNAi or degraders) without removing the gene.

Metabolome: The set of small-molecule metabolites in a cell or sample.
Mechanism of action: The biological cascade by which an intervention produces its phenotypic effect, ideally specified at multiple scales.
Molecular docking: Computational placement of a small molecule in a protein binding pocket, scored by an approximate binding energy.

Overexpression: Increasing the level of a gene product above baseline, typically by introducing extra copies or stronger promoters.

Pathway: A set of interacting molecular components (genes, proteins, metabolites) and the directed signed relationships between them. Pathways are priors, not executable truth.
Perturbation: An intentional intervention applied to a biological system: drug, CRISPR edit, knockdown, overexpression, ligand, environmental change.
Proteome: The full set of proteins present in a cell or sample, including modifications and complexes.
Phenotype: An observable trait or behavior, from molecular readouts to cell morphology to organismal outcomes.
Perturb-seq: Combination of CRISPR perturbation with single-cell RNA-seq readout, giving per-cell perturbation-response data.
Potency: Concentration of a compound required to produce a given biological effect (e.g. IC50, EC50). Distinct from efficacy.
PK/PD: Pharmacokinetics (what the body does to the drug) and pharmacodynamics (what the drug does to the body).

Single-cell RNA-seq: Measurement of RNA expression at the resolution of individual cells, producing high-dimensional, sparse, noisy data.
Selectivity: Degree to which a compound acts on its intended target relative to off-targets.

Transcriptome: The full set of RNA transcripts in a cell or sample at a given time.
Toxicity: Harm to the organism from a compound, on-target or off-target, acute or chronic.
Therapeutic index: Ratio between the dose producing harm and the dose producing benefit.

Virtual cell: A computational model of a cell that predicts state and behavior under interventions, integrating multiple data types.