Wednesday, October 14, 2026

Science Is Supposed to Change Its Mind

One of the easiest ways to misunderstand science is to imagine it as a warehouse of facts. A result enters the warehouse, acquires the label scientifically proven, and remains there forever.

That is not how science works.

The National Academies report Reproducibility and Replicability in Science presents a much more interesting picture. Scientific knowledge develops through discovery, confirmation, criticism and correction. New evidence can strengthen an existing explanation, reveal its limits, or occasionally force scientists to replace it.

The report captures this beautifully in three words: scientific knowledge is “durable and mutable.”

Those two qualities might seem contradictory.

Durable means that well-supported scientific knowledge does not evaporate every time a new paper appears. Newtonian mechanics did not become useless when relativity was developed. Existing knowledge survives because it has accumulated evidence behind it.

Mutable means that no scientific claim receives immunity from new evidence.

Better instruments can reveal things that older instruments could not detect. Larger datasets can uncover patterns hidden in small samples. New experiments can identify conditions under which an apparently universal relationship no longer applies.

That ability to change is not a defect in science. It is one of its central strengths.

The report's preface uses changes in nutritional and medical advice to illustrate the danger of speaking too confidently about what we currently believe to be true. Scientific conclusions generally arrive with some degree of uncertainty, and repeated confirmation increases confidence rather than magically transforming a claim into absolute truth.

This is why a paper that overturns an earlier result does not necessarily mean that someone previously failed at science.

Sometimes the new study is wrong.

Sometimes the original study was wrong.

Sometimes both studies are correct under different conditions.

And sometimes the disagreement reveals a variable nobody previously knew mattered.

The scientific process is designed to sort among these possibilities.

The deeper lesson is therefore surprisingly optimistic: science progresses not because scientists never make mistakes, but because scientific claims remain exposed to interrogation.

A scientific result is not carved into stone.

It enters a conversation.

########################################################################

Further reading: This post is based on Reproducibility and Replicability in Science, a consensus study report of the National Academies of Sciences, Engineering, and Medicine. National Academies of Sciences, Engineering, and Medicine. 2019. Reproducibility and Replicability in Science. Washington, DC: The National Academies Press. DOI: 10.17226/25303.

Tuesday, October 13, 2026

Is It Still Science If It Isn't New?

We often talk about science as if its defining feature were novelty.

A scientist discovers something nobody has seen before. A new species is found. A new gene function is reported. A new mechanism is proposed. A new theory explains an old observation.

And because scientific papers are expected to make a "novel contribution," it is tempting to equate science with discovering something new.

But that raises a deceptively simple question:

If an experiment tells us nothing new, is it still science?

And an even more interesting one:

If an experiment is perfectly reproducible, but the interpretation of its results is wrong, is it still science?

The answer to both questions is essentially yes—but with important distinctions.

Science is not the same thing as novelty. Novelty describes a contribution to knowledge. Science describes a way of generating, testing, communicating, and correcting knowledge about the world.

That distinction changes how we think about scientific research.

Science is more than discovery

Imagine that a researcher performs an experiment and obtains a result that has already been reported ten times.

At first glance, it might seem that nothing scientific has happened. There is no discovery. No new phenomenon. No new fact.

But suppose the researcher carefully repeats the experiment with a different population, laboratory, instrument, organism, or environmental condition.

The result is again obtained.

The experiment has not necessarily produced a novel fact. But it has done something scientifically valuable: it has provided confirmation.

Science does not consist only of asking:

"What has nobody seen before?"

It also asks:

"Is what we think we know actually reliable?"

That is why replication matters.

A replicated result may be completely unsurprising, yet still contribute to the reliability of scientific knowledge.

Discovery is science—but science is not only discovery

The U.S. National Academies describe scientific knowledge as developing through discovery, confirmation, and correction.

That three-part description is useful because it immediately takes us beyond the popular image of the lone scientist making a spectacular discovery.

Consider three researchers.

Researcher 1: Discovery

They investigate a poorly studied organism and discover that a particular gene is completely absent from its genome.

Nobody had previously reported this.

If the observation is supported by appropriate genomic evidence, this is a scientific discovery.

It adds something previously unknown to our knowledge of nature.

Researcher 2: Confirmation

Another scientist investigates the same gene in independently generated genomes and finds the same absence.

They haven't discovered a new fact.

But they have strengthened the evidence that the original observation was genuine.

That is also science.

Researcher 3: Correction

A third scientist discovers that both studies overlooked a highly divergent copy of the gene.

The apparent "gene loss" was actually an annotation problem.

Again, science has happened.

The third study has not merely failed to discover something new. It has corrected an existing scientific claim.

And correction is not an embarrassment that sits outside science.

It is one of the ways science works.

What about a wrong interpretation?

This is where things become particularly interesting.

Suppose an experiment is beautifully designed.

The measurements are reliable.

The experiment is independently replicated.

Everyone agrees on the observed result.

But the researchers interpret the result incorrectly.

Does that make the experiment non-science?

No.

We need to separate the observation from the interpretation.

Consider this simplified chain:

Experiment → Data → Observation → Interpretation → Conclusion

Each step makes a different claim.

Suppose an experiment shows that treatment X causes gene Y to become more highly expressed.

That observation may be completely reproducible.

The researchers then conclude:

"Therefore, gene Y causes cell proliferation."

But their experiment never manipulated gene Y. It only measured its expression.

The causal interpretation may therefore be wrong.

The experiment can still be scientifically sound.

The data can still be scientifically useful.

The conclusion can still be wrong.

And, importantly, discovering that the conclusion was wrong can itself become part of science.

Reproducibility does not make an interpretation true

This distinction is easy to miss.

We sometimes talk about reproducibility as though it were a guarantee of truth.

It isn't.

Imagine that ten laboratories independently reproduce an experimental observation.

That gives us increasing confidence that the observation is real.

But it does not necessarily establish that we understand what the observation means.

You can therefore have:

Highly reproducible data + highly reproducible analysis + an incorrect interpretation.

Consider a diagnostic assay that consistently produces a particular molecular signal.

The reproducibility tells us that the signal is reliably generated.

But we still have to establish what the signal represents.

Is it:

the presence of a particular protein?
a cross-reacting molecule?
a degradation product?
a related paralog?
an artifact of the assay?
or genuinely the biological entity being claimed?

Reproducibility answers one question:

"Does this observation reliably occur?"

It does not necessarily answer:

"Why does it occur?"

And certainly not:

"Does our preferred explanation for it have to be true?"

An experiment can be scientific even when its conclusion is wrong

This is an uncomfortable but fundamental feature of science.

Science is not a magical procedure that converts experiments into truth.

Scientists can make mistakes.

They can misunderstand mechanisms.

They can overinterpret correlations.

They can mistake artifacts for biological signals.

They can construct explanations that later turn out to be wrong.

What distinguishes science is not that scientists are always correct.

It is that scientific claims are supposed to remain open to empirical challenge and correction.

A wrong scientific hypothesis can therefore be perfectly legitimate science.

The problem arises when the evidence is presented as establishing something it does not establish, particularly when alternative explanations are ignored or contradictory evidence is suppressed.

The experiment may remain scientific.

The claim about what the experiment demonstrates may not.

Does science require novelty?

This is where scientific research and scientific publishing sometimes get confused.

There are at least two different questions:

Question 1

Is this science?

Question 2

Is this a sufficiently novel contribution to publish in this particular journal?

They are not the same question.

A replication study can be science even if a journal decides that it lacks sufficient novelty.

A careful reanalysis of existing data can be science.

A correction of an existing result can be science.

A negative result can be science.

A methodological validation can be science.

A failure to reproduce an influential result can be science.

A careful measurement that confirms something we already strongly suspected can be science.

None necessarily requires discovering a previously unknown fact.

Novelty is therefore a property of the contribution, not the defining property of science itself.

What kinds of science can lack novelty?

Consider several possibilities.

1. Replication

"We performed the experiment again and obtained the same result."

No new phenomenon may have been discovered.

But the reliability of the existing observation has been tested.

Science? Yes.

2. Confirmation

"Using an independent method, we obtain evidence consistent with the existing conclusion."

Again, there may be little or no novelty in the phenomenon itself.

But independent evidence can strengthen scientific knowledge.

Science? Yes.

3. Falsification or refutation

"We tested a previously accepted prediction and found that it does not hold."

This may not discover a new phenomenon.

Instead, it removes support from an existing explanation.

Science? Absolutely.

4. Correction

"The earlier observation was real, but the proposed explanation was incorrect."

The new study may not discover anything fundamentally new.

It may simply establish that our interpretation of existing evidence needs revision.

Science? Yes.

5. Methodological validation

"We tested whether this experimental technique actually measures what researchers think it measures."

The study may produce no new biological discovery whatsoever.

But it can establish whether a tool is reliable.

Science? Yes.

6. Negative results

"We predicted that X would happen, but under these conditions it does not."

The absence of an expected effect can be scientifically informative.

Indeed, a scientific framework in which only positive results count would systematically distort our understanding of nature.

Science? Yes.

7. Reanalysis of existing data

Suppose all the data already exist.

A researcher develops a new statistical analysis and asks whether the original conclusion follows from the data.

No new experiment is performed.

No new organism is collected.

No new molecule is discovered.

Yet the analysis may reveal that the original conclusion was unsupported—or that an overlooked pattern exists in the existing data.

Science? It can be.

Science is not defined by whether new data were personally generated. It is defined by what is being done with evidence and how claims about the world are evaluated.

But then what is novelty for?

If novelty isn't what makes something science, why do scientists care so much about it?

Because science is not only about producing reliable knowledge.

It is also about extending the frontier of knowledge.

A research paper generally needs to answer:

"What does this work add?"

That addition might be a new observation.

Or a new explanation.

Or a new method.

Or stronger evidence.

Or a correction.

Or a previously unrecognized limitation.

Or a demonstration that a supposedly general phenomenon does not generalize.

Thus, novelty is enormously important for scientific progress and publication, even though it is not the defining characteristic of scientific activity.

A useful distinction is:

Science asks whether a claim is empirically justified.

Scientific novelty asks whether the resulting knowledge adds something not already known.

Those are different axes.

And what about an incorrect scientific paper?

This brings us to perhaps the most uncomfortable conclusion.

A paper can be:

scientifically conducted,
carefully executed,
reproducible,
transparently reported,

and nevertheless contain a wrong interpretation.

That does not necessarily mean the work was "not science."

Instead, we should ask which component is wrong.

Perhaps the measurement was wrong.

Perhaps the experiment was incapable of distinguishing competing hypotheses.

Perhaps the statistical inference was inappropriate.

Perhaps the causal interpretation went beyond the evidence.

Perhaps the biological mechanism was misunderstood.

Perhaps a later study revealed an overlooked alternative explanation.

These distinctions matter because otherwise we fall into a binary way of thinking:

Correct = science
Incorrect = not science

But science doesn't work that way.

If it did, there would be no need for scientific correction.

The deeper idea: science is a process, not a truth machine

Perhaps the most useful way to think about science is not:

Science = discovering new facts.

Nor:

Science = producing correct answers.

Rather:

Science is a systematic, evidence-based process through which claims about the natural world are generated, tested, challenged, communicated, confirmed, refined, and corrected.

Within that process, discovery is one activity.

Replication is another.

Confirmation is another.

Refutation is another.

Correction is another.

And explanation is another.

Some produce novelty.

Some don't.

Some ultimately produce correct conclusions.

Some don't.

Yet all can participate in science.

The paradox

There is therefore a fascinating paradox at the heart of science:

A study can be scientifically valuable without discovering anything new.

And:

A study can discover something genuinely new while getting its interpretation wrong.

The first tells us that novelty is not necessary for something to be science.

The second tells us that being new is not sufficient for something to be scientifically established.

What matters is the relationship between evidence and claims.

A new observation requires evidence.

An old observation can be tested again.

An interpretation must be justified.

A conclusion must not exceed what the evidence supports.

And when we get something wrong, science has a mechanism—at least in principle—for finding out that we were wrong.

Perhaps that is a more useful definition of science than the popular image of scientists constantly discovering spectacular new things.

Science is not simply the business of finding what nobody knew before.

It is the collective process of figuring out what we have good reason to believe about the world—and continuously finding out where we were wrong.

Sunday, October 11, 2026

Forgotten Crops May Be Part of Agriculture's Future

The final section of the article turns archaeology toward the future.

This is perhaps its most important message.

Studying ancient agriculture is not merely about discovering where wheat came from.

The archaeological record represents thousands of years of experiments in agricultural resilience.

And many of those experiments have been forgotten.

Diversity repeatedly rises and falls

Early domestication often incorporated variation from multiple populations.

But isolation can gradually erode that diversity.

The same process occurs at the level of farming systems.

Agricultural communities once cultivated many crop combinations that later disappeared.

Some species became marginal crops.

Others vanished from large parts of their former ranges.

A crop can go from important to nearly invisible

The paper highlights Triticum timopheevi.

This wheat was once cultivated across a large region.

Archaeological evidence places it in parts of Turkey before 6000 BC.

It eventually spread from central Europe toward Turkmenistan.

Yet during later periods its range contracted dramatically.

By the twentieth century it had become largely associated with Georgia.

The crop did not cease being biologically viable.

Agricultural history moved around it.

“Orphan crops” are evolutionary libraries

Many under-studied crops may contain useful characteristics:

drought tolerance,

disease resistance,

adaptation to poor soils,

nutritional qualities,

unusual perennial growth,

or compatibility with local farming systems.

Modern agriculture is heavily concentrated around a relatively small number of globally dominant crops.

That concentration creates efficiency, but also vulnerability.

A changing climate makes those vulnerabilities harder to ignore.

Archaeology can become agricultural research

Usually we imagine modern agricultural science helping us understand ancient farming.

The authors propose reversing the direction.

Ancient agriculture can help us imagine future agricultural systems.

Archaeobotany tells us which crops survived particular climatic fluctuations.

Genomics reveals where ancient populations acquired new genetic diversity.

Archaeology documents agricultural systems that persisted for centuries or millennia.

Historical crop distributions reveal species that were once important but later abandoned.

Together, these records form a gigantic natural experiment conducted across thousands of years.

The final lesson

The history of domestication is not a story of humans progressively perfecting a few plants.

It is a history of experimentation, hybridisation, ecological engineering, accidental selection, migration, loss and renewal.

Agricultural diversity has repeatedly contracted.

But it has also repeatedly been rebuilt when people moved crops, exchanged seeds and brought previously separated populations together.

The future of agriculture may therefore depend partly on recovering something that the past possessed in abundance:

options. 🌾

###################################################################################

Further reading: This post is inspired by and draws on Dorian Q. Fuller, Tim Denham, and Robin Allaby, “Plant domestication and agricultural ecologies,” Current Biology 33, no. 11 (2023): R636–R649, doi:10.1016/j.cub.2023.04.038. Read the original article in Current Biology
For a classic and highly readable perspective on why plant domestication arose in some regions but not others, see Jared Diamond, Chapter 8, “Apples or Indians: Why Did Peoples of Some Regions Fail to Domesticate Plants?”, in Guns, Germs, and Steel: The Fates of Human Societies (W. W. Norton, 1997), pp. 131–156. Read “Apples or Indians” View Guns, Germs, and Steel on Google Books

###################################################################################

Saturday, October 10, 2026

Why Some Crops Conquered Continents While Others Stayed Home

Some domesticated plants spread astonishingly far.

Wheat and barley moved across huge portions of Eurasia.

Rice spread from China across much of Asia.

Maize moved far beyond Mesoamerica.

Pearl millet expanded beyond West Africa.

Other crops remained geographically restricted.

Why?

The article highlights two especially important variables:

time and productivity.

Older crops had longer to travel

A crop domesticated several thousand years earlier simply had more opportunities to spread through migration and exchange.

But age is not enough.

Agricultural productivity influences human demography.

Productivity changes population movement

A highly productive crop can support relatively high population densities.

Under lower productivity, communities may reach the effective carrying capacity of a local environment more quickly.

This can encourage group fission.

Some people establish new settlements.

Crops travel with them.

Paradoxically, under certain circumstances a less productive agricultural system might therefore promote faster territorial expansion.

Carrying capacity is not fixed

The paper wisely treats carrying capacity as “notional.”

A landscape does not possess one permanent human carrying capacity.

It changes with:

technology,

labour,

crop choice,

soil management,

climate,

water availability,

and cultivation strategy.

Humans continually modify the ecological ceiling they are approaching.

Some forgotten crops may simply have missed history's transport network

A crop remaining local does not necessarily mean it is agronomically inferior.

Historical circumstances matter.

Enset in Ethiopia, oca in the Andes and browntop millet in India are examples discussed in the paper of crops whose geographical spread remained limited.

Their unrealised agricultural potential may therefore be considerable.

History's winners are not automatically biology's best candidates.

Sometimes they were simply the crops standing nearest the door when human populations began moving.

###################################################################################

Further reading: This post is inspired by and draws on Dorian Q. Fuller, Tim Denham, and Robin Allaby, “Plant domestication and agricultural ecologies,” Current Biology 33, no. 11 (2023): R636–R649, doi:10.1016/j.cub.2023.04.038. Read the original article in Current Biology
For a classic and highly readable perspective on why plant domestication arose in some regions but not others, see Jared Diamond, Chapter 8, “Apples or Indians: Why Did Peoples of Some Regions Fail to Domesticate Plants?”, in Guns, Germs, and Steel: The Fates of Human Societies (W. W. Norton, 1997), pp. 131–156. Read “Apples or Indians” View Guns, Germs, and Steel on Google Books

###################################################################################

Friday, October 9, 2026

Domestication Did Not Necessarily Begin with a Genetic Bottleneck

A classic genetic model of domestication predicts a bottleneck.

A small subset of a wild population is selected.

Those individuals become the founders of the crop.

Genetic diversity collapses.

Strong selection then fixes important domestication alleles.

This certainly can happen.

But the review argues that it is not an adequate general model for early plant domestication.

Archaeology predicts weak selection

If domestication took thousands of generations, selection could have been relatively weak.

Under weak selection, more individuals remain within the reproducing population.

That preserves diversity.

Genomes reveal multiple contributors

Ancient and modern genomic data increasingly indicate that several crops incorporated genetic material from multiple wild populations.

Barley, emmer, maize and bananas all provide examples of complex histories.

Instead of:

wild population → bottleneck → crop,

the history may resemble:

multiple wild populations ↔ cultivated populations ↔ migration ↔ hybridisation → crop metapopulation.

Gene flow can rescue crops

One of the most important examples discussed in the paper concerns sorghum.

Ancient sorghum genomes from Nubia suggest that genetic load increased through time.

Isolation can allow harmful variation to accumulate or useful diversity to disappear.

But contact with other sorghum populations can introduce new variation.

This process is known as genomic rescue.

Hybridisation, far from contaminating a pure crop lineage, can therefore restore resilience.

Purity may be the wrong metaphor

Modern breeding often emphasises uniformity.

Ancient domestication frequently appears to have benefited from mixture.

The history of successful crops may therefore be partly a history of repeatedly reconnecting populations.

Agricultural resilience sometimes grows from genetic messiness.

That lesson has obvious consequences for modern crop conservation.

Wild relatives, traditional landraces and geographically isolated crop populations should not be viewed simply as museum pieces.

They represent reservoirs of evolutionary options.

###################################################################################

Further reading: This post is inspired by and draws on Dorian Q. Fuller, Tim Denham, and Robin Allaby, “Plant domestication and agricultural ecologies,” Current Biology 33, no. 11 (2023): R636–R649, doi:10.1016/j.cub.2023.04.038. Read the original article in Current Biology
For a classic and highly readable perspective on why plant domestication arose in some regions but not others, see Jared Diamond, Chapter 8, “Apples or Indians: Why Did Peoples of Some Regions Fail to Domesticate Plants?”, in Guns, Germs, and Steel: The Fates of Human Societies (W. W. Norton, 1997), pp. 131–156. Read “Apples or Indians” View Guns, Germs, and Steel on Google Books

###################################################################################

How to Suggest Reviewers for a Scientific Manuscript

A practical guide to identifying appropriate reviewers, avoiding conflicts, using AI responsibly, and predicting who is most likely to accept

Suggesting reviewers is one of the most delicate parts of a journal submission.

It can be tempting to treat it as a strategic exercise:

"Who will give us a favourable review?"

That is the wrong starting point.

The appropriate question is:

Who is sufficiently knowledgeable, sufficiently independent, and sufficiently likely to evaluate this manuscript fairly?

A good reviewer list helps the editor find qualified experts quickly. A poor list can raise concerns about conflicts of interest, reviewer manipulation, or attempts to influence the editorial process.

The ideal reviewer-suggestion process therefore combines scientific expertise, independence, diversity, availability and editorial judgment.

1. What is the purpose of suggesting reviewers?

When a journal asks authors to suggest reviewers, it is generally not asking:

"Who do you want to review your paper?"

It is asking:

"Who could competently and independently evaluate this work?"

The editor remains responsible for choosing the reviewers.

Your suggestions are therefore recommendations, not nominations.

A strong reviewer list gives the editor several useful options.

2. The first principle: choose expertise, not friendliness

The most important rule is:

Never start by asking who likes your work. Start by asking who understands your work.

Suppose your paper contains:

comparative genomics,
phylogenetics,
gene-loss analysis,
transcriptomics,
evolutionary modelling,
and functional interpretation.

You may need reviewers with different strengths.

For example:

Reviewer	Primary expertise
A	Comparative genomics
B	Molecular evolution
C	Gene-loss/pseudogene evolution
D	Transcriptomics
E	Functional genomics

You don't necessarily need five people who work on exactly the same gene.

You need a group capable of evaluating the different claims made by the paper.

3. Start by identifying the manuscript's "reviewer needs"

Before searching for names, make a list of the expertise required.

For example:

Scientific question

Who studies the biological phenomenon?

Methodology

Who understands the main analytical approach?

Data type

Who understands the underlying dataset?

Interpretation

Who works on the broader biological/evolutionary question?

Specialized methods

Who understands the unusual method or model?

This produces a reviewer expertise matrix.

For example:

Manuscript component	Required expertise
Gene loss	Comparative genomics
Orthology	Genome evolution
Phylogeny	Molecular phylogenetics
Selection tests	Evolutionary modelling
Transcriptomics	Gene-expression analysis
Functional interpretation	Molecular biology

Now search for people who collectively cover these areas.

4. Where should reviewer candidates come from?

There are many legitimate sources.

A. Recent papers

Probably the best starting point.

Search for researchers who published papers addressing the same scientific problem within approximately the last 3–5 years.

Look at:

authors of key papers
authors of recent methodological papers
authors of competing interpretations
authors of high-quality reviews

5. Follow the citation network

Start with your manuscript's most important references.

For each important paper:

Who wrote it?
Who cited it?
Who subsequently challenged or extended it?
Who publishes repeatedly in this area?
Who developed the methods you use?

This creates a much better reviewer pool than simply searching Google for:

"gene X expert."

6. Look at recent papers rather than reputation alone

A famous scientist may be an excellent expert but a poor practical reviewer candidate.

Possible problems:

no longer active in the field
too senior and overloaded
moved into a different research area
does not review manuscripts anymore
lacks expertise in the specific methods

A researcher publishing three relevant papers in the last two years may be a better candidate than a famous scientist who published one relevant paper ten years ago.

7. Examine the authors' recent publications

For each candidate, ask:

How recently did they publish on the topic?

How often do they publish in the area?

Are they methodologically relevant?

Do they publish independently?

Are they likely to understand the manuscript?

Are they currently active?

This is much more informative than citation count alone.

8. Build a reviewer candidate spreadsheet

A lab should ideally maintain one.

For example:

Candidate	Institution	Expertise	Recent relevant paper	Conflict?	Likely expertise	Availability	Recommendation
A	Univ X	Gene evolution	2025	No	Very high	High	Yes
B	Univ Y	Phylogenomics	2026	No	High	Medium	Yes
C	Univ Z	Functional biology	2024	Recent collaborator	High	Unknown	No
D	Univ Q	Gene loss	2026	Competitor	Very high	Unknown	Maybe

This is far better than producing five names at the last minute.

9. Conflict of interest is the critical filter

Before suggesting anyone, investigate conflicts.

Potential conflicts include:

Recent collaboration

Someone who recently co-authored a paper with you is generally inappropriate.

Same institution

A researcher at your institution is usually unsuitable.

Recent student-supervisor relationship

Avoid former supervisors, students and close academic relationships where the journal's policies consider them conflicts.

Family/personal relationships

Obviously avoid these.

Financial relationships

Avoid people with relevant financial interests.

Direct professional conflicts

If there is an ongoing dispute, competition or formal conflict, be cautious.

Manuscript overlap

A person working on an extremely similar unpublished project may present a competing-interest problem.

10. Scientific disagreement is not automatically a conflict

This distinction is important.

Suppose your manuscript challenges Professor X's published interpretation.

Should Professor X automatically be excluded?

Not necessarily.

If Professor X is an independent expert and the journal permits author-suggested reviewers, they may actually be an excellent reviewer.

But there is a difference between:

"This researcher has a different scientific interpretation."

and:

"This researcher is involved in an active personal/professional dispute with us."

The former can be scientifically valuable.

The latter may constitute a genuine conflict.

11. Competitors require careful judgment

Direct competitors are complicated.

A direct competitor may be:

highly qualified,
completely independent,
scientifically ideal,

but may also have:

unpublished competing work,
a strong personal stake in the outcome,
access to your ideas before publication.

Do not automatically exclude competitors.

Instead ask:

Would a reasonable editor consider this person capable of providing an independent review without a significant competing interest?

If the answer is uncertain, disclose the issue rather than attempting to manipulate the reviewer pool.

12. How many reviewers should you suggest?

Follow the journal's instructions.

If it asks for:

3 reviewers

provide three strong candidates.

If it asks for:

5–8

provide a broader but carefully screened pool.

Do not submit 20 names merely to increase the chance that your preferred person is selected.

Quality matters more than quantity.

13. Don't suggest only your friends

This is one of the most obvious warning signs.

A suspicious list might consist of:

former collaborators,
close colleagues,
former supervisors,
people from your academic network,
researchers who cite you frequently.

Even if none technically violates a conflict rule, the pattern can look inappropriate.

14. Don't suggest only famous scientists

The opposite mistake is also common.

A list of five Nobel-level or field-leading researchers may look impressive but be practically useless.

Very senior researchers may:

receive enormous numbers of invitations,
have heavy administrative responsibilities,
decline most reviews,
delegate some reviewing,
have limited time.

A mid-career researcher with exactly the right expertise may be much more likely to accept.

15. Which reviewers are most likely to accept?

There is no reliable way to predict an individual's decision.

However, reviewer acceptance tends to be more plausible when the candidate:

Is actively publishing

Someone currently publishing in the area is more likely to see the manuscript as relevant.

Has reviewed before

Researchers who regularly review for journals are accustomed to the process.

Is at an appropriate career stage

Mid-career researchers and established early-career investigators may have both expertise and motivation, although this varies enormously.

Has a close topical match

A manuscript directly aligned with their research is more attractive than a peripheral connection.

Has the methodological expertise

Researchers are more likely to accept when they can confidently evaluate the work.

Is not overloaded

A person with a huge administrative workload may decline even if they are the perfect scientific match.

Is not conflicted

Obvious conflicts reduce the likelihood of a legitimate invitation.

16. A useful way to think about reviewer acceptance

You can conceptualize reviewer availability as:

Expertise × relevance × independence × availability

rather than:

fame × citation count.

A candidate with moderate fame but excellent topical fit may be a much better reviewer than a superstar.

17. Early-career researchers can be excellent reviewers

Do not restrict your list to professors.

A postdoctoral researcher, assistant professor, research scientist or newly independent investigator may have:

extremely current expertise,
detailed methodological knowledge,
time and motivation,
experience with the exact analytical techniques.

However, check whether the journal permits the person and whether the invitation should instead go to their PI or supervisor.

Never assume that someone should review simply because they are a coauthor's former student.

18. Geographic diversity

If appropriate candidates exist, avoid selecting everyone from:

one country,
one institution,
one academic network.

A geographically diverse reviewer pool can help reduce network bias.

However, expertise should come first.

Do not select a weaker reviewer merely to satisfy geographic diversity.

19. Gender and other diversity considerations

Where multiple equally qualified candidates exist, diversity can be a useful consideration.

But diversity should not become a substitute for expertise.

The priority remains:

competence
independence
relevance
availability

with diversity considered among otherwise appropriate candidates.

20. Search systematically

A robust search strategy might look like this:

Step 1

Identify 10–20 relevant recent papers.

Step 2

Extract authors.

Step 3

Remove:

your authors,
recent collaborators,
institutional colleagues,
obvious conflicts.

Step 4

Rank the remaining candidates by expertise.

Step 5

Check recent publication activity.

Step 6

Check methodological relevance.

Step 7

Check current affiliation.

Step 8

Check whether they are plausibly independent.

Step 9

Select the strongest candidates.

Step 10

Verify contact details from authoritative sources.

21. Use multiple discovery routes

Do not rely on one database.

Useful sources can include:

PubMed
Google Scholar
Web of Science
Scopus
Crossref
ORCID
institutional websites
journal articles
conference programs
field-specific databases

The objective is not to find the biggest list.

It is to verify the identity, expertise and independence of candidates.

22. AI can help—but it should not make the final decision

AI can be very useful for generating a candidate pool.

For example, give an AI system:

title
abstract
keywords
methods
major references
research question

and ask it:

Identify researchers who have published peer-reviewed work in the last five years on the specific scientific questions and methods represented in this manuscript.

This can produce useful candidates.

But the output should be treated as:

a discovery tool, not a reviewer recommendation.

23. A good AI reviewer-search workflow

A robust workflow is:


MANUSCRIPT
     ↓
AI identifies scientific themes
     ↓
AI generates candidate researchers
     ↓
Human verifies publications
     ↓
Human checks conflicts
     ↓
Human checks current affiliation
     ↓
Human evaluates independence
     ↓
Human ranks candidates
     ↓
Final reviewer suggestions

The human verification stage is essential.

24. What AI is particularly good at

AI can help with:

Identifying subfields

For example:

comparative genomics + molecular evolution + pseudogenization

Finding expertise gaps

AI may recognize that your manuscript actually requires:

phylogenetics expertise,
statistical modelling,
functional genomics,
genome annotation.

Expanding candidate pools

It can identify researchers you may not have encountered through your immediate academic network.

Comparing expertise

You can ask AI to summarize:

What aspects of this manuscript could researcher X evaluate particularly well?

Detecting obvious conflicts

AI can flag potential coauthorship relationships—but these must be independently verified.

25. What AI should NOT be trusted to do

Do not blindly accept AI-generated:

researcher names,
affiliations,
email addresses,
publication histories,
conflicts,
claims about expertise.

AI systems can hallucinate.

A nonexistent paper or incorrect affiliation can easily make its way into a reviewer list.

Therefore:

Every suggested reviewer should be verified against primary or authoritative sources.

26. Don't ask AI "Who will give me a favourable review?"

That is a problematic objective.

A better prompt is:

"Identify independent researchers who are highly qualified to evaluate the scientific claims and methodology of this manuscript."

The distinction is important.

The goal should be appropriate review, not favorable review.

27. Don't use AI to predict personal bias

Be particularly cautious with prompts such as:

"Which researchers are likely to agree with our conclusions?"

or:

"Find reviewers who will be sympathetic to our paper."

This moves the process away from identifying independent expertise and toward attempting to influence peer review.

Instead ask:

"Which researchers have published evidence supporting, challenging, or independently investigating the scientific question?"

That produces a more balanced reviewer pool.

28. Use AI to identify both sides of a scientific debate

This can actually improve reviewer selection.

For example:

Identify researchers who have independently published evidence supporting the prevailing model and researchers who have published evidence challenging it.

Then:

verify the candidates,
remove conflicts,
assess expertise,
let the editor choose.

This is particularly useful when your manuscript challenges an established interpretation.

29. AI can help build a reviewer matrix

For example:

Candidate	Expertise	Methods	Recent work	Independence	Potential conflict	Overall
A	Very high	High	2026	High	None	Excellent
B	High	Very high	2025	High	None	Excellent
C	Very high	Medium	2026	Low	Competitor	Avoid
D	Medium	High	2026	High	None	Good

This is a much better use of AI than asking:

"Give me five reviewers."

30. Verify every candidate manually

For every final candidate, check:

Identity

Is this actually the person you think they are?

Current affiliation

Has the person moved?

Expertise

Do their recent papers actually match your manuscript?

Recent publication

Have they published recently in the area?

Conflicts

Have they coauthored with your authors?

Institution

Are they independent?

Relationship

Do you have a close professional relationship?

Contact details

Is the email address from an authoritative source?

31. Be careful with email addresses

Do not obtain reviewer emails from questionable databases.

Prefer:

university profile
laboratory website
ORCID-associated information
journal publication
institutional directory

And if the journal can identify the reviewer from their name and affiliation without requiring an email, do not invent one.

32. What about people who cited your work?

They can be appropriate.

But citation alone does not make someone a good reviewer.

A person who cited your paper once in passing is not necessarily an expert.

Conversely, someone who never cited you may be an excellent reviewer.

Expertise should be determined from the body of relevant work, not citation relationships.

33. What about authors of papers you cite?

They are often excellent candidates.

But there are two possibilities:

They are independent experts.

Excellent.

Your manuscript directly attacks their published conclusion.

Potentially still appropriate, but consider whether there is a genuine conflict and whether the editor should know about the relationship.

Scientific disagreement itself is not a reason to manipulate the reviewer pool.

34. What about reviewers who reviewed your previous paper?

Potentially excellent candidates if:

they are independent,
their expertise remains appropriate,
there is no conflict,
the journal allows it.

But do not attempt to reconstruct anonymous reviewers through clues.

If a reviewer was anonymous, you should not try to identify them simply because you liked or disliked their review.

35. What about suggesting the same reviewers to multiple journals?

There is nothing inherently wrong with recommending an appropriate expert to another journal if the manuscript is subsequently submitted elsewhere.

But don't maintain a rigid list of "my five favourite reviewers."

The appropriate reviewer pool should be rebuilt based on:

the manuscript,
the target journal,
the current literature,
current conflicts.

36. A particularly useful concept: reviewer complementarity

Instead of asking:

Who are the five best experts?

ask:

What five perspectives would provide the strongest independent evaluation?

For example:


Reviewer 1 → Biological question
Reviewer 2 → Evolutionary theory
Reviewer 3 → Computational methodology
Reviewer 4 → Statistical analysis
Reviewer 5 → Functional interpretation

This can be much stronger than five people who all work on exactly the same biological system.

37. The ideal reviewer profile

A particularly good candidate often has:

recent publications in the topic,
direct methodological expertise,
sufficient independence,
active research,
experience reviewing,
no obvious conflicts,
an appropriate career stage,
a professional connection to the scientific problem rather than to the authors.

That combination is much more valuable than prestige alone.

38. A practical ranking system

A lab could score candidates:

Criterion	Score
Scientific expertise	0–5
Methodological expertise	0–5
Recent activity	0–3
Independence	0–5
Relevance to manuscript	0–5
Availability likelihood	0–3
Conflict risk	−5 to 0
Overall	—

But do not turn this into an algorithm that automatically selects reviewers.

The scoring system is a decision aid.

39. An important warning: don't optimize for acceptance probability

You might be tempted to build:

"Reviewer X = 80% chance of accepting invitation."

That is not really knowable from public information.

Instead, use qualitative categories:

Likely to be feasible

active researcher
strong topical fit
regular publication activity
not obviously overloaded

Possibly difficult

extremely senior
major administrative role
exceptionally high-profile
already serving on many editorial boards

Poor candidate

conflict
weak expertise
inactive in field
insufficient independence

40. Never contact a reviewer asking them to review your paper

Unless the journal explicitly permits pre-submission reviewer contact, do not send someone:

"We have suggested you as a reviewer. Please accept if invited."

That undermines the independence of the process.

Similarly, do not send them the manuscript privately with the expectation that they will review it.

Let the journal handle the invitation.

41. Don't ask colleagues to create fake reviewer identities

This is an obvious but important ethical boundary.

Never:

create fake email accounts,
provide personal addresses pretending to be institutional,
nominate yourself under another identity,
ask a friend to review under a different identity,
manipulate reviewer databases.

These practices can constitute serious publication misconduct.

42. Don't suggest people who cannot reasonably review the manuscript

For example, don't recommend:

a graduate student who has no independent reviewing role unless the journal permits it,
a retired researcher who has completely left the field,
someone working in an unrelated field simply because they are famous,
a person whose expertise is only tangentially related.

43. Reviewer suggestions should help—not control—the editor

The healthiest attitude is:

"Here are several people who could competently evaluate this manuscript."

not:

"Here are the people we want you to select."

The editor may select none of them.

That is completely normal.

44. A recommended laboratory workflow

For every manuscript:

Stage 1 — Identify expertise

Create the reviewer-needs matrix.

Stage 2 — Generate candidates

Use:

recent papers,
citation networks,
databases,
editorial boards,
conferences,
AI-assisted discovery.

Stage 3 — Create 10–20 candidates

Do not immediately select five.

Stage 4 — Verify candidates

Check:

publications,
affiliation,
conflicts,
independence.

Stage 5 — Rank

Identify the strongest candidates.

Stage 6 — Create a balanced final list

Provide the number requested by the journal.

Stage 7 — Record the rationale

Maintain the internal reviewer-selection record.

45. Keep a reviewer-selection record

For your lab's publication archive, I would create:


Reviewer_Selection/
├── Candidate_List.xlsx
├── Expertise_Assessment.xlsx
├── Conflict_Check.xlsx
├── AI_Assisted_Search_Record.txt
└── Final_Reviewer_List.pdf

This is particularly useful if a question later arises about why a reviewer was suggested.

46. A useful AI prompt

A responsible AI-assisted workflow might use a prompt like:

"Analyze the following manuscript abstract, key claims, methods and references. Identify the scientific subfields and methodological expertise required to independently evaluate the manuscript. Then generate a candidate pool of researchers who have published recent peer-reviewed work in those areas. For each candidate, explain which aspect of the manuscript they are qualified to assess. Do not rank candidates according to whether they are likely to support the manuscript's conclusions. Do not infer personal relationships or conflicts without evidence. Clearly mark any information that requires independent verification."

Then provide:

title
abstract
key claims
methods
major references

After AI produces the candidate list, independently verify every candidate.

47. An even better AI workflow

Instead of asking AI for names immediately:

Prompt 1

Identify the five most important scientific expertise areas needed to review this manuscript.

Prompt 2

Identify the major scientific debates relevant to the manuscript.

Prompt 3

Identify researchers who have published recent work representing the different perspectives in those debates.

Prompt 4

Compare these candidates by topical expertise and methodological relevance.

Prompt 5

Identify information that must be independently verified before any candidate is suggested to a journal.

This reduces hallucination and confirmation bias.

48. The final reviewer-selection checklist

Before submitting the names:

Expertise

Can this person genuinely understand the manuscript?
Have they published recently in the relevant area?
Do they understand the key methodology?

Independence

No recent collaboration?
No institutional conflict?
No close personal/professional relationship?
No inappropriate financial relationship?

Scientific balance

Are different relevant perspectives represented?
Is the list dominated by one academic network?
Have we avoided selecting people simply because they are likely to agree?

Practicality

Are they currently active?
Is their current affiliation correct?
Are contact details verified?
Is the suggested number consistent with journal instructions?

AI

Was AI used only as a discovery/organization aid?
Were all candidates independently verified?
Were no AI-generated claims about conflicts accepted without verification?
Was AI used to identify expertise rather than favorable reviewers?

49. The central principle

The best reviewer-selection strategy is not:

"Find people who will like our paper."

It is:

"Find people who would be capable of finding the weaknesses in our paper, while being sufficiently independent to evaluate those weaknesses fairly."

That mindset has an interesting advantage.

If your manuscript survives scrutiny from people who genuinely understand its weaknesses, the paper is usually stronger.

50. The final reviewer-selection philosophy

A good reviewer list should make an editor think:

These are credible, independent experts who understand the science and can evaluate this manuscript properly.

It should not make the editor think:

These authors appear to be trying to engineer a favourable review.

The strongest reviewer-selection process therefore follows a simple sequence:

Understand the manuscript → identify the expertise required → find active experts → verify their work → eliminate conflicts → ensure appropriate breadth → suggest the strongest independent candidates.

AI can make the search and organization much faster.

It should not replace the scientific and ethical judgment involved in deciding who is an appropriate reviewer.

And perhaps the most important rule of all:

Choose reviewers you would still consider appropriate if you had no idea whether they would recommend acceptance or rejection.

That is a good test of whether the reviewer-selection process is genuinely about obtaining an independent scientific assessment rather than attempting to control the outcome.