01Separate new observations from a repeat analysis

The National Academies uses reproducibility for obtaining consistent computational results from the same data and procedures, and replicability for consistency across studies using new data. Terminology varies between fields, so this article uses that stated convention rather than assuming every paper does.National Academies — Reproducibility and Replicability in Science, chapter 3 (opens in a new tab)

Re-running a peptide dataset through the original analysis can reveal whether the computational result is obtainable. Collecting another dataset asks whether the finding is supported by additional observations. Both are useful, but they address different possible weaknesses.

A different laboratory can add information about transfer between settings. A new experiment by the original team can also constitute new-data replication; the affiliation alone does not define what was repeated.

02Read the claim the replication was designed to test

A broad statement that a peptide affects a pathway is harder to replicate precisely than a defined comparison in a stated model with a specified readout. Look for the original result that the follow-up selected as its target.

The new study should explain which features were retained and which changed. Material identity, biological model, outcome definition and analysis can all matter to whether the comparison addresses the same claim.

An original hypothetical follow-up might use the same cellular model and outcome but obtain material from a different batch. That supplies a test with one changed material context. A study using a different organism and a different outcome asks a broader transfer question.

03Compare estimates rather than counting positive papers

Suppose one study reports a positive estimated change with substantial uncertainty and a second reports a smaller positive change with similarly substantial uncertainty. A significance threshold might be crossed in only one study even though their estimates do not provide clear evidence of disagreement.

The reverse problem also occurs: two studies can both have statistically significant results while estimating effects of materially different sizes. Shared labels do not establish close agreement.

Read the direction, magnitude and precision in relation to the replication target. If the question concerns a biologically meaningful effect, consider whether the evidence can distinguish that effect from changes too small to matter in the stated model.

Avoid defining replication success only after inspecting the new result. A stated comparison criterion helps readers understand what the investigators intended to evaluate and prevents several different questions being traded against one another unnoticed.

04Use disagreement to locate the remaining uncertainty

A follow-up that does not support the original result warrants scrutiny of both studies. It does not, by itself, establish fraud, prove the effect is impossible or identify which methodological difference caused the disagreement.

Useful reporting makes alternative explanations inspectable: uncertain original estimates, changed materials, differences in the biological setting, measurement limitations or analytical choices. These are questions to investigate, not a list of excuses that automatically preserves the original claim.

Likewise, successful replication under similar conditions does not establish that a result applies everywhere. Agreement strengthens a bounded evidence claim; transfer to substantially different settings remains a separate question.

A good research summary therefore identifies the replicated comparison, the new observations and the conditions under which results agreed or differed. It gives the follow-up its proper evidential role without reducing a developing literature to a tally of successful and failed papers.