My name is Naomi Chen, and I work in biomedical informatics, building machine learning models that flag early biomarkers for pancreatic cancer from routine blood panels. It is a field where the technical bar is high and the writing bar somehow gets overlooked, which I did not fully understand until my own manuscript got turned away twice in under six months.

I am putting this account together because I remember scrolling through forums at 1 a.m. looking for someone to tell me plainly what actually goes wrong at this stage, instead of another polished success story with no specifics. So here are the specifics from my own case, for whoever is searching the way I was.

Two Rejections, Neither One About the Model

My first submission went to a computational biology journal that regularly publishes diagnostic modeling work. The rejection came after six weeks, without full review, and the note said the clinical relevance of the model's performance metrics was not made clear enough for a non-computational readership. I revised the framing and sent a second version to a different journal about two months later.

That second rejection arrived faster, in under four weeks, and raised a completely different concern. This editor felt the paper leaned too heavily on model architecture detail relative to how it discussed the actual patient population and clinical implications, and suggested the balance needed rethinking before resubmission anywhere in that space.

Neither editor questioned my model's accuracy, my dataset, or my validation approach. Both rejections were, in different ways, about how the work was explained rather than what the model actually did. I did not connect those two notes at the time. It felt safer to treat each one as an isolated editorial quirk than to admit that two unrelated reviewers landing on a similar structural concern was probably telling me something real.

What Reading Both Rejections Side by Side Actually Showed Me

After the second rejection, I sat down and read both editorial letters together instead of reacting to each one separately as it came in. Looking at them side by side made the overlap harder to explain away. My introduction spent most of its length on model architecture and almost none on why a clinician should care about the specific biomarkers I was flagging. My results section reported performance metrics precisely but never translated what those numbers meant for actual early detection outcomes.

Part of the problem was that I had spent close to two years deep in model iteration, so I had lost the ability to judge what a first-time reader, particularly a clinically minded one, would actually take away from the paper. A former postdoc mentor, who had gone through something similar with an earlier diagnostics paper, suggested getting an outside manuscript review before submitting a third time without more information. I had held off earlier over cost, but two rejections in, that hesitation stopped making sense.

That is how I ended up reaching out to Harvard Publication Hub, after reading through several accounts from researchers in adjacent computational field’s describing a similar cycle of rejection without ever getting feedback specific enough to act on.

What the Sample Review Caught That I Had Missed

Before agreeing to a full review, I sent over just my introduction and results section to test whether the feedback would be genuinely useful. What came back was more precise than I expected. It pointed out that my abstract claimed the model "outperformed existing screening methods" while the results section only supported a more modest, comparable performance claim under specific conditions, a mismatch a careful reviewer would likely flag as overstatement. It also noted that two of my figures were referenced out of order relative to how they appeared in the text, which was a small thing but the kind of small thing that makes a reviewer question the paper's overall care.

I had genuinely missed both issues. After two rounds of editorial feedback that never named a specific problem, having someone point to an exact sentence and an exact figure, and explain clearly why each mattered, was the first feedback that gave me something concrete to fix.

The Actual Scope of the Revision

Once I moved forward with the full review, the changes went well beyond what I had expected walking in:

●        Rewrote the abstract and introduction so performance claims matched exactly what the results section actually supported

●        Reordered and relabeled figures so they appeared in the same sequence they were referenced in the text

●        Added a clinical interpretation paragraph after the results section explaining what the flagged biomarkers meant for early detection timelines in practical terms

●        Trimmed a lengthy architecture description down to what was relevant for a mixed clinical and computational readership

●        Corrected citation formatting across roughly twenty-eight references that had drifted between two citation styles across drafts

●        Rebuilt the conclusion so it addressed clinical relevance directly rather than restating model performance alone

None of my underlying model, training data, or validation results changed through this process. What changed was whether the paper read as an overstated technical showcase or an appropriately framed clinical contribution, which turned out to be the actual issue behind both earlier rejections. The revision took a little over six weeks, and more than once a clarifying question made me realize I could explain the clinical significance out loud far more clearly than I had managed to write it down.

Matching the Paper to the Right Journal This Time

Before resubmitting, I had an actual conversation about journal fit rather than defaulting to the two or three names I already knew. We compared how recent issues balanced computational depth against clinical framing, typical review timelines, and how that lined up with a fellowship application deadline I was working against.

I ended up submitting to a journal I had originally passed over, assuming it was too clinically oriented for a paper this model-heavy. It turned out to be exactly the right audience for research connecting machine learning outputs to early detection practice. Neither of my first two submissions had gone anywhere close to the right fit, regardless of how carefully the manuscript had been written.

Getting Through Peer Review

This journal sent the manuscript to three reviewers, and comments came back roughly seven weeks later. One reviewer wanted a clearer explanation of how the training and validation cohorts differed demographically, and requested a supplementary sensitivity analysis. A second raised a fair concern about how the model would generalize outside the original patient population. The third had mostly minor comments but pushed back on one clinical claim in the discussion that the data only partially supported.

Responding well took genuine effort. I got help structuring the response letter so each point was addressed directly, conceding where the critique was fair and explaining my reasoning clearly where I saw it differently, without sounding dismissive or rewriting the entire paper around a single reviewer's preference.

Being Upfront About Time and Cost

From the first sample review to the version I finally resubmitted, the process took about eight weeks, not counting the seven weeks peer review added afterward. It was a genuine expense, and I weighed it seriously against how much a peer-reviewed publication would matter for an upcoming fellowship cycle. For my situation, it was worth the cost, though I recognize that calculation depends heavily on someone's career stage and funding source.

The Acceptance Email I Had Stopped Expecting

The acceptance notice arrived on a Friday afternoon while I was running an unrelated set of model diagnostics. I read it once quickly, then again more slowly to make sure "accept with minor revisions" meant what I thought it did. After two rejections, I had genuinely stopped expecting that email to show up at all.

Looking back at both earlier rejections now, neither one feels like a verdict on the model or the underlying science. They were both pointing at the same fixable problem the entire time. I just did not have anyone name it clearly until the third attempt.

What I Would Tell Someone in the Same Position

If you have been rejected more than once and the feedback feels technically vague or inconsistent each time, my honest takeaway is this. It is rarely the model or the underlying research, and it is rarely bad luck with reviewers either. It is almost always something specific and fixable in how the claims and framing are presented on the page, and it usually takes someone outside the project, someone who reads manuscripts professionally, to name exactly what that something is. I wish I had asked for that kind of read after my first rejection instead of my second.