Ranking Targets by Mechanistic Reversibility and Tissue Reach

Why reversibility and tissue reach belong in target ranking

A target can produce a large predicted improvement in one simulated tissue and still be a poor development choice. The effect may persist after the perturbation is removed, suggesting limited control over the altered system, or the required modulation may not be feasible in the tissues that matter for disease. Conversely, a moderate predicted benefit may be more useful if it occurs across relevant tissues, can be achieved there, and subsides when the perturbation is withdrawn.

A disciplined comparison therefore separates three questions: how much disease improvement does the model predict, how does the system respond when target modulation ends, and can a proposed therapeutic approach plausibly reach the required modulation in each tissue? These are related but distinct properties. A high score on one should not be used to imply a high score on the others.

A comparison is interpretable only if candidates are simulated under a shared protocol: the same disease context, virtual population, tissue definitions, initial conditions, intervention duration, recovery period, and disease-burden outcome. Use units and outcome definitions that permit tissue-by-tissue interpretation. If raw outcomes have different scales across tissues, normalize them against a meaningful tissue-specific baseline rather than summing incompatible measurements.

Define benefit and recovery before comparing candidates

Let \(Y_{0j}\) be the baseline disease burden in tissue \(j\), \(Y_{kj}^{\mathrm{on}}\) the burden during the prespecified modulation of candidate target \(k\), and \(Y_{kj}^{\mathrm{off}}\) the burden at the end of a fixed recovery period after modulation is removed. Assume lower burden is desirable and \(Y_{0j}>0\). A relative benefit is

For example, \(B_{kj}=0.25\) means the modeled burden is 25% lower than baseline during modulation. A negative value indicates worsening, not benefit. Use a matched baseline and the same intervention duration for all candidate comparisons.

A useful descriptive recovery fraction is

If the outcome returns exactly to baseline, \(R_{kj}=1\); if it remains as far from baseline as it was during modulation, \(R_{kj}=0\). Values below zero indicate that the endpoint is farther from baseline after recovery than during modulation. Values can exceed one if the endpoint crosses baseline and ends closer to it than the on-treatment endpoint. Report the endpoints as well as this summary: the absolute-distance formula does not show whether recovery undershot or overshot baseline, and it does not describe the full time course.

The calculation assumes the baseline is a suitable reference, the same outcome is measured at all three times, and the on-treatment change is large enough for the denominator to be meaningful. If the on-treatment change is near zero, the fraction is unstable; report the raw trajectory and treat recovery as not estimable or not informative for that tissue rather than interpreting a noisy ratio.

Represent whether a therapy can reach the required modulation

Reach should be assessed for a specified therapeutic concept, not assigned to a target in the abstract. For each tissue, define \(d_{kj}^{\mathrm{req}}\) as the magnitude of target modulation used to produce the modeled benefit, and \(d_{kj}^{\mathrm{max}}\) as a justified estimate of the modulation that the proposed approach could achieve in that tissue. Both quantities must use the same scale and direction. A simple reach fraction is

This bounded measure is 1 when the estimated feasible modulation meets or exceeds the modeled requirement, and below 1 when it falls short. It is a screening summary, not a substitute for a mechanistic exposure-response simulation. Its inputs may come from modality, tissue-delivery, or target-engagement evidence; uncertainty in those inputs should be made explicit. If no credible estimate of feasible modulation exists, label reach as unknown rather than silently setting it to 1.

A transparent tissue-weighted ranking

A practical screening score can combine positive predicted benefit with reach and recovery. Choose tissue weights \(w_j\) before examining candidate rankings, with \(w_j\ge 0\) and \(\sum_j w_j=1\). For a candidate with estimable benefit and recovery in the tissues being scored, define

This score rewards benefit only where the therapy is estimated to reach the required modulation and where the modeled response recovers after modulation ends. The maximum function prevents a worsening tissue outcome from being counted as positive benefit. It does not make worsening harmless: report negative \(B_{kj}\) values separately, along with any other modeled adverse responses. A material harmful effect can rule out a candidate even when its score is high elsewhere.

The score is a transparent decision aid, not a universal biological quantity. It depends on the outcome, weights, perturbation size, recovery horizon, and feasibility assumptions. Preserve the component results beside the total so a ranking can be audited. In particular, a candidate with strong benefit in one tissue and poor reach in another should not be described simply as “highly effective.”

Justifying the recovery summary and checking the simulation

The recovery fraction has a simple interpretation because it compares the residual distance from baseline with the original on-treatment displacement. Let \(\Delta_{\mathrm{on}}=Y^{\mathrm{on}}-Y_0\) and \(\Delta_{\mathrm{off}}=Y^{\mathrm{off}}-Y_0\). When \(\Delta_{\mathrm{on}}\ne0\), the fraction of displacement remaining is \(|\Delta_{\mathrm{off}}|/|\Delta_{\mathrm{on}}|\). Subtracting that remaining fraction from one gives

This derivation justifies \(R=1\) for a return to baseline and \(R=0\) when the distance from baseline is unchanged. It does not justify interpreting \(R\) as a probability, nor does it establish that every internal state has recovered just because one output has.

Before accepting a ranking, inspect the full simulated trajectories and the implementation of the perturbation and its removal. Check that target modulation is actually withdrawn at the specified time, that events and solver settings do not leave an unintended input active, and that the system is evaluated over a sufficiently long, biologically justified recovery period. Confirm that the baseline is stable or that the comparison accounts for untreated disease progression. If different candidates are simulated from different starting conditions or with different intervention durations, the score does not provide a fair comparison.

Alternatives, limitations, and a common pitfall

The combined score is useful when a compact, prespecified summary is needed, but it is not the only appropriate presentation. A tissue-by-tissue profile preserves where benefits, limited reach, or poor recovery occur. A gate-based approach can be preferable when a minimum benefit or minimum reach in a critical tissue is mandatory: candidates that fail the gate are not rescued by strong performance elsewhere. A Pareto comparison is useful when no defensible single weighting exists; it identifies candidates that cannot be improved on one component without sacrificing another, while leaving trade-offs visible. These approaches answer different decision needs and should not be presented as interchangeable.

Important limitations remain. A single recovery endpoint compresses a time course and can miss slow recovery, overshoot, or persistent changes in unmeasured states. A modeled feasible-modulation ceiling is uncertain and may not represent delivery or target engagement in every relevant cell type. Tissue weights encode priorities, not objective biological truths. Finally, a simulation ranking is conditional on model structure, parameterization, and the chosen intervention representation; it does not establish clinical efficacy or guarantee that a therapy can achieve the modeled effects.

Common pitfall: “Candidate A has the largest predicted benefit, so it is the best target.” This ignores whether the benefit can be reached in the relevant tissues and whether the response behaves as required after modulation ends. The correction is to report benefit, recovery, and reach separately, examine adverse tissue responses, and use a prespecified score or decision rule only as a transparent summary of those components.