Hi all, I’m looking for feedback on whether this type of work is realistically publishable as a speculative, hypothesis-generating study, rather than as definitive biological truth. We would be extremely conservative in our claims and explicitly frame this as proposing a mechanistic hypothesis rather than proving one.

Background

I’m studying a historically rare but increasingly frequent subtype of liver cancer that appears resistant to the standard drug used for more common liver cancers. The original goal was to identify candidate pathways that might plausibly explain this resistance and then validate them experimentally.

We initially planned to conduct cell culture and qPCR validation, but funding cuts eliminated this possibility. The available human bulk microarray cohorts and TCGA data are so poorly annotated that meaningful clinical validation isn’t possible. I contacted a group with semi-annotated data, but legal restrictions prevented further data sharing.

Despite this, my PI would like to pursue publication, specifically as a computational, hypothesis-generating paper, rather than a validation study. I'm the only computational person in the lab, and most of what I do is beyond her scope, so she's given me some time to brainstorm and figure out a viable path forward.

Analysis overview

Because human datasets for the rare cancer are extremely limited, I used mouse model scRNA-seq datasets, which have been shown in the literature to closely resemble human liver cancer transcriptional programs and are commonly used as stand-ins when human data are unavailable.

1. Ortholog mapping & cell selection

• Mouse genes were mapped to human orthologs using orthogene.
• Cell types were annotated, and the analysis was restricted to hepatocytes.

2. Cross-species integration

• Mouse and human scRNA-seq datasets were integrated using scANVI (semi-supervised) on the top 6,000 HVGs.
• This produced a corrected counts matrix.
• Correlation and PCA analysis on raw versus corrected counts showed a broadly similar structure, supporting preservation of the biological signal.

3. Pseudobulk DE and pathway analysis

• Hepatocyte-only pseudobulk DE was performed using limma-voom, followed by GSEA.
  (Hepatocytes are of particular interest to the lab as key resistance drivers and are the most easily validatable with cell culture at a later date.)
• The corrected counts matrix was used. The intent here was not to claim definitive DE, but to identify candidate pathways that differ between conditions on a comparable expression scale.

4. Internal consistency/support analyses

• To test whether the identified resistance pathways showed preferential activation (and whether known drug-target pathways were suppressed), FDR-corrected Spearman correlations were computed between pathway gene signatures and pseudobulk-aggregated raw hepatocyte counts within each original dataset.
• Genes outside the 6,000 HVGs could still emerge if they showed significant correlation with the pathway signature.
• Strong negative correlations aligned with known drug-action pathways.
• GSEA on FDR-significant genes ranked by signed correlation coefficients further supported the internal coherence of the hypothesized resistance program.

5. Biological plausibility

• Key regulators of this pathway are known to be mutated specifically in the rare cancer subtype, but their downstream transcriptional effects have not been explored.
• No direct DE comparison between these cancer subtypes has been published.
• A prior microarray meta-analysis reported upregulation of a broad pathway class consistent with our findings, although it did not explicitly identify this pathway.

What I’m asking

• Is a clearly labeled, hypothesis-generating, cross-species scRNA-seq study like this publishable at all without wet-lab or clinical validation?
• Are there aspects of this approach (e.g., ortholog mapping, scANVI correction, pseudobulk DE) that reviewers are likely to reject even for a speculative paper?
• Would this be better framed as a brief report / computational hypothesis / methods-forward paper, or is the lack of validation still likely to be a hard stop?

I’d really appreciate honest, even blunt, feedback so I can decide whether to proceed or pivot while there’s still time.