Translational Biology Lab

Evidence should change the decision.

Adjust genetics, mechanism, cellular context, biomarkers and safety. The recommendation is deliberately revisable: the goal is to expose uncertainty and identify the next experiment with the highest decision value.

Evidence-to-decision exercise

Make the reasoning explicit.

This is a research-programme simulation, not a clinical recommendation. It teaches how heterogeneous evidence can support GO, PAUSE, REDIRECT or STOP while remaining open to revision.

Human genetics 65

Mechanistic validation 70

Cellular / tissue context 55

Biomarker readiness 40

Safety / redundancy confidence 50

Research-programme decision simulation

PAUSE

Evidence is promising but a decision-critical gap remains.

Highest-value next experimentStrengthen biomarker readiness

The weakest evidence layer currently limits confidence more than the stronger layers add it.

Worked example · published mechanismALC1 auto-inhibition and PARP-dependent activation

The 2017 single-first-author Molecular Cell study defined how poly-ADP-ribose releases ALC1 auto-inhibition, joining DNA-damage signalling, chromatin remodelling and oncology. That published mechanism forms part of the scientific lineage of later ALC1-focused therapeutic development — a real instance of the evidence-to-decision path this exercise models. Open the Molecular Cell study ↗

Principle: translation is the disciplined conversion of heterogeneous evidence into revisable decisions—not a single score pretending uncertainty has disappeared.

Translation atlas · 10 decision gates

A chain of evidence, not a confidence score.

Each gate asks what must be true for the programme to advance and what result should pause, redirect or stop it.

Decision rule: state the action for each plausible experimental outcome before running the next study. If every outcome leads to GO, the gate is not controlling the programme.

Define the decision contextWhich patient, disease state and intervention decision is the programme trying to change?

Evidence to inspect: Unmet need, current standard, population, stage, endpoint, feasible modality and acceptable risk.
What would change the decision: A more precise context reveals that the proposed mechanism cannot alter the intended decision.
Primary confound: A broad disease label combines biologically distinct populations and outcomes.
Highest-value next test: Write one target product and evidence statement with population, action, comparator and outcome.

Triangulate human relevanceDo genetics, tissue evidence and disease trajectory support the same directional biology?

Evidence to inspect: Allelic direction, penetrance, expression, spatial context, longitudinal association and independent replication.
What would change the decision: Human evidence points in the opposite direction or is explained by linkage, ascertainment or tissue composition.
Primary confound: Association is mistaken for target dependency or intervention direction.
Highest-value next test: Integrate directionally concordant genetics and tissue-state evidence, then predefine the perturbation prediction.

Resolve target mechanismWhich molecular step connects target engagement to the disease-relevant state?

Evidence to inspect: Direct engagement, proximal intermediate, temporal order, dose response, orthogonal perturbation and rescue.
What would change the decision: Adequate engagement fails to change the predicted proximal intermediate.
Primary confound: Pleiotropy or broad toxicity changes the endpoint without the proposed mechanism.
Highest-value next test: Use an acute, orthogonal perturbation–rescue experiment with proximal and distal readouts.

Choose a fit-for-decision modelWhich missing constraint would most threaten translation?

Evidence to inspect: Genotype, lineage, tissue architecture, immune context, exposure, phenotype fidelity and cross-model replication.
What would change the decision: The dependency disappears in every model that restores the decision-critical constraint.
Primary confound: Complexity is added without improving the mechanism or decision being tested.
Highest-value next test: Select a second model specifically for the largest unresolved translational risk.

Resolve patient heterogeneityWho carries the mechanism and who does not?

Evidence to inspect: Biological subgroup definition, prevalence, continuous state, interaction effect, ancestry, treatment history and uncertainty.
What would change the decision: The subgroup effect fails in independent data or follows a technical variable.
Primary confound: Post-hoc thresholds and multiple testing manufacture a responder group.
Highest-value next test: Validate a pre-specified mechanism-linked interaction in separated data.

Qualify the biomarkerCan the marker be measured reliably and change the intended decision?

Evidence to inspect: Analytical precision, biological validity, threshold, prevalence, calibration, reproducibility and context of use.
What would change the decision: Routine measurement error or realistic prevalence eliminates decision value.
Primary confound: Retrospective accuracy is mistaken for prospective utility.
Highest-value next test: Run blinded analytical validation and decision analysis near the intended threshold.

Match modality to mechanismCan the intervention reach and alter the required target state?

Evidence to inspect: Compartment, target form, occupancy or degradation, selectivity, reversibility, dosing logic and proximal response.
What would change the decision: The modality cannot access or control the mechanistically relevant target pool.
Primary confound: Biochemical potency is extrapolated across compartments and target states.
Highest-value next test: Measure compartment-specific engagement and proximal response for the proposed modality.

Establish exposure and safety windowIs target engagement achievable for long enough without unacceptable harm?

Evidence to inspect: Free exposure, tissue distribution, PK/PD, healthy-context dependency, selectivity, recovery and delayed toxicity.
What would change the decision: The engagement required for efficacy overlaps the exposure that causes mechanism-linked harm.
Primary confound: Short assays miss cumulative toxicity or active metabolites.
Highest-value next test: Measure exposure, engagement, response and toxicity in matched time-resolved samples.

Anticipate resistance and adaptationHow can the system bypass, compensate for or evolve around the intervention?

Evidence to inspect: Parallel pathways, target alteration, state switching, microenvironment, chronic exposure and combination interaction.
What would change the decision: The proposed combination does not prevent escape or destroys the safety window.
Primary confound: Acute response is treated as durable disease control.
Highest-value next test: Use longitudinal exposure with state profiling, washout and mechanistically justified combinations.

Make the programme decisionWhich action is justified now, and which result would reverse it?

Evidence to inspect: Causal coherence, model validity, subgroup, biomarker, modality, exposure, safety, resistance, feasibility and uncertainty.
What would change the decision: A veto criterion is met or the next experiment cannot plausibly change the decision.
Primary confound: A weighted average hides one fatal weakness or rewards redundant evidence.
Highest-value next test: Choose GO, PAUSE, REDIRECT or STOP; record decisive evidence, weakest link, owner, milestone and stopping rule.