Scientific due diligence: how to test a compelling story before you invest

A coherent mechanism is not yet a de-risked asset

Science-led companies often begin with a persuasive chain: a biological mechanism is associated with a condition; an intervention affects that mechanism; early data show movement in a useful direction; therefore the product has a credible path to value. Every link may be plausible while the chain remains fragile.

Scientific due diligence tests the chain at its weakest points. It asks not only whether published evidence exists, but how directly it supports the commercial claim, whether the result has survived independent testing, whether the model resembles the intended use, and which assumptions still carry most of the risk.

Translate the pitch into testable claims

Start by separating the story into claims that could be wrong. A useful claim ledger might include the mechanism, target engagement, effect size, safety, generalisability, manufacturability, measurement validity and the path from a surrogate outcome to a meaningful benefit.

This matters because broad questions invite broad reassurance. ‘Is the science sound?’ is difficult to answer and easy to perform. ‘Does the cited evidence show that modulating target T changes clinically meaningful outcome O in the intended population?’ creates a reviewable task.

For an investment or partnership, a commissioned scientific due diligence review applies this claim-by-claim scrutiny to the materials behind the deal.

Five layers of evidence to inspect

• Mechanistic credibility: is the proposed causal chain supported, or mainly inferred from association?
• Model relevance: do cell, animal or simulation models capture the features that matter in humans or real-world use?
• Measurement validity: are the endpoints reliable, meaningful and resistant to analytical flexibility?
• Reproducibility and independence: has the signal appeared outside the originating group, dataset or analysis pipeline?
• Translation: what must remain true as the product moves from controlled experiments to manufacturing, clinical use or deployment?

None of these layers supplies a verdict alone. Together they expose whether the opportunity rests on several independent supports or on one result repeated in different forms.

Read negative and missing evidence correctly

A failed replication is informative, but so is the absence of an expected study. If a mechanism has been prominent for years yet independent teams have not produced the predicted result, that absence belongs in the risk picture. It is not proof of failure. It is a reason to ask what incentives, technical barriers or publication patterns might explain the gap.

Likewise, a lack of direct human evidence can be acceptable early in development. The diligence question is whether the remaining uncertainty is recognised, whether the next experiment can reduce it, and whether the valuation already assumes success.

Build an uncertainty register, not a literature dump

The most useful output is a ranked register of assumptions. For each one, record the supporting evidence, contrary evidence, directness, quality, consequence if false, and the fastest way to test it. This makes research operational. It tells an investment committee which risks are fundamental, which are manageable, and which can be converted into milestones.

A concise memo can then separate three categories: claims that are well supported; claims that are plausible but under-tested; and claims that currently outrun the evidence. The language should preserve those distinctions. ‘Promising’ is not a scientific category unless the promise is specified.

What good diligence changes

Good scientific diligence does not merely produce a yes or no. It may change the structure of the decision: stage capital against a replication, request access to raw data, narrow the initial indication, revise the market model, recruit a missing specialist, or stop before sunk costs make stopping harder.

The value of the review lies in making the dependency structure visible. Once you know what must be true for the opportunity to work, you know what to verify next.