Hypothesis-Driven vs. Discovery Aims: How to Frame Exploratory Science for NIH Reviewers

Why Reviewers Keep Asking for a Hypothesis

Most study section members trained in an era when a grant meant a hypothesis, a prediction, and an experiment that could prove the prediction wrong. That habit hasn't gone away just because sequencing got cheap. When a reviewer reads "we will profile X across Y conditions," the unspoken follow-up is "and then what?" If the application doesn't answer, the critique writes itself: descriptive, lacks a clear hypothesis, risk of generating data nobody can interpret.

There's a second reason, and it's less about taste. A hypothesis does real work for a reviewer. It tells them what outcome would count as failure, which lets them judge feasibility, statistical power, and rigor in a concrete way. A discovery aim that doesn't supply a substitute for that work leaves the reviewer guessing. Guessing reviewers tend to guess conservatively.

I can't give you a number for how often this critique sinks an application, because it isn't tracked anywhere public. What I can say from reading summary statements is that it shows up disproportionately in applications built around a screen, a cohort profiling effort, or a new computational method, and that it often appears next to a perfectly good Significance score.

Three Kinds of Aims, and What Each One Owes the Reader

It helps to be honest about which type of aim you're writing, because each one has a different obligation to the reviewer.

Mixing types inside one aim is where applications get muddy. If Aim 1 is a screen and also quietly tests three hypotheses, the reviewer can't tell which obligations you're accepting. Split them, or state clearly which part is which.

Give Every Discovery Aim a Decision Rule

The most useful fix I know is to replace the missing hypothesis with a pre-specified decision rule. A decision rule says: here is the threshold at which we call something a hit, here is the number of hits we consider a success, and here is what happens next at each outcome. It's the exploratory equivalent of a falsifiable prediction.

For example, instead of "we will perform a genome-wide screen to identify regulators of drug tolerance," try something with teeth: hits are defined by a fixed effect-size and false discovery threshold, the top candidates go to orthogonal validation in two cell models, and any validated hit that also shows up in patient-derived data advances to Aim 2. Notice that this doesn't claim you know the answer. It only claims you know what you'll do with whatever comes back.

Include the null scenario too. If the screen returns few or no validated hits, what's the fallback? A sentence on this does more for your Approach score than a page of background, because it shows you've thought about failure before a reviewer had to ask. This ties directly into the risk-versus-ambition judgment reviewers make, which we cover in our piece on overly ambitious R01 scope.

The Hybrid Structure: Discovery Feeding Testable Claims

For many R01s the strongest layout is a hybrid. Aim 1 is the discovery or resource aim, where the decision rule does the work a hypothesis would. Aim 2 takes the most likely outputs of Aim 1 and tests a specific mechanistic hypothesis. Aim 3, if you have one, tests whether the mechanism holds in a different context, like an animal model or a clinical cohort.

The risk with this pattern is the pipeline trap. If Aim 2 can only proceed with Aim 1's specific results, reviewers will flag dependency, and they'll be right to. The workaround is to anchor Aim 2 in a hypothesis you already have preliminary data for, so it stands on its own, and then state that Aim 1 will broaden or refine it. If Aim 1 comes back empty, Aim 2 still runs. That independence is the same principle covered in our guide to the Specific Aims page.

One caution. Some applicants bolt on a hypothesis they don't believe just to satisfy the critique. Experienced reviewers can smell it, particularly when the experiments proposed couldn't actually distinguish the hypothesis from its alternative. A candid exploratory aim with a decision rule is better than a decorative hypothesis.

Rigor Looks Different When You Don't Have a Single Hypothesis

Discovery work creates its own rigor problems, and reviewers who care about this will look for them. The big ones are multiple testing, batch effects, sample size justification, and the gap between a statistical hit and a biological one. Name each of these and say how you'll handle it. A sentence on how many biological replicates per condition, how you'll control the false discovery rate, and which hits get independent confirmation is usually enough in the Aims and gets expanded in the Approach.

Power is trickier than it is for a two-group comparison. If you can't run a formal calculation, say so, and point to pilot variance or published effect sizes from a similar system. Reviewers tolerate honest uncertainty far better than a power statement that clearly wasn't calculated. Our overview of writing the rigor and reproducibility sections goes through the other required elements, including authentication and biological variables.

Finally, think about data and code. Exploratory aims produce large datasets, and the reviewer will wonder whether anyone but you can use them. Pointing to your Data Management and Sharing Plan, and naming the repository you'll deposit in, turns a potential weakness into a visible strength.

Before-and-After Aim Language

Here are three patterns I see often, with the kind of rewrite that tends to help. The wording is illustrative, so adapt it to your field.

Frequently Asked Questions