Together, step by step, we’re making great strides — the best is yet to come!🚀💪

#LiquidBiopsy #CfRNA #RNASeq #Transcriptomics #Biomarkers #Bioinformatics #PrecisionMedicine #Preprint #CancerDiagnostics #Flomics #MethodsMatter #Reproducibility #Genomics #Diagnostics

And a huge thanks to our incredible team:

@julienlag.bsky.social, @ctuni.bsky.social, @fokkerplanck.bsky.social, @asoleee.bsky.social, Pablo Monteagudo, Elena Rusu, Lluc Cabus, Lucila González, Laura Sánchez, Beatriz Nieto & Phil Sanders who made this possible!

🙏 Thanks to @upf.edu, @crg.eu, @idibaps.bsky.social, @cienciagob.bsky.social, @leitat.bsky.social and other collaborators for their invaluable support and contribution!

We believe this resource will be very useful to researchers, clinicians, and companies in the liquid biopsy space. We welcome constructive feedback and discussion!

And now for a not-so-humble brag…
We were thrilled when the analysis revealed our own in-house Flomics libraries, using a rigorous double-DNase digestion protocol, ranked as the highest-quality dataset among all whole-cfRNA-Seq methods! 🎉💪

But we’re not just here to point out problems!
Our work provides a data-driven benchmark of what works, and a roadmap for more robust studies.

Beware of confounders!
We highlight how technical factors are often mixed with patient phenotype, making true, generalizable biomarker discovery a huge challenge.

Protocols are not created equal. We show how different workflows give you vastly different pictures of the transcriptome.

This technical noise is loud!
The variation we found is greater than the biological variation across 29 different human tissues. Let that sink in.

It’s the method, not the biology.
We found that inter-lab batch effects, genomic DNA contamination, and library diversity explain the main differences between samples.

We had a hunch, so we conducted a massive cross-study analysis of >2,300 samples from 15 studies, plus libraries generated in-house at Flomics. We used a uniform bioinformatics pipeline to see what was really going on.

Results were… illuminating.

But we’re not just here to point out problems! Our work provides a data-driven benchmark of what works, and a roadmap for more robust studies.

Beware of confounders!
We highlight how technical factors are often mixed with patient phenotype, making true, generalizable biomarker discovery a huge challenge.

Protocols are not created equal.
We show how different workflows give you vastly different pictures of the transcriptome.

This technical noise is loud!
The variation we found is greater than the biological variation across 29 different human tissues. Let that sink in.

It’s the method, not the biology.
We found that inter-lab batch effects, genomic DNA contamination, and library diversity explain the main differences between samples.