Finding a relevant public dataset is only the beginning. With Trove, you can compare studies, check whether they’re actually compatible, and look for signals shared across datasets. Public proteomics data, put back to work.

What if better data matters as much as a bigger model? Our new preprint: a 70M-parameter model trained with proteomics matched or exceeded RNA-only models up to 50× larger. 48,843 samples · 440 studies. Why we’re building Trove: put existing proteomics data back to work. tinyurl.com/yubucze5

Personalized cancer vaccines are advancing. So is the need to identify the tumor antigens that matter. See how Tesorai Search supports immunopeptidomics with greater sensitivity and at scale. bit.ly/4gp3kRF

Public proteomics data is abundant. Reusing it is another story. In our analysis of 439 datasets, 84% had essential metadata outside the data file. With Trove, we’re working to make public proteomics data easier to find, understand, and reuse. #Proteomics

What biological signals hold up across independent studies? Tesorai Trove helps researchers evaluate and compare public proteomics datasets through a conversational workflow—making it easier to find results reproduced across studies. Here’s what that looks like in 23 seconds.

Good data can outlive the question that generated it. Public proteomics datasets can be revisited with new hypotheses, new analyses, and new comparisons. That’s the idea behind Tesorai Trove: making the evidence behind published research easier to explore and reuse.

Describe the visualization you need. Tesorai builds it from your proteomics data. Here, a natural language request becomes a custom, interactive plot of protein CV distributions, with options to download the figure or inspect the code. What would you visualize?

We're proud to share our latest peer-reviewed research, now published in the Journal of Molecular Biology. Thanks to our collaborators, especially @mpibiochem.bsky.social and @jmolbiol.bsky.social. 📄 www.sciencedirect.com/science/arti...

Differential expression tells you which proteins changed. Protein interaction networks help explain how those proteins work together, revealing the biological processes behind the results. What's your favorite follow-up analysis after differential expression?

Better hit identification creates more opportunities for successful drug discovery. In a recent biotech collaboration, Tesorai Search identified 11 additional candidate hits, resulting in >30% more new hits than the existing workflow. Reference customer available for qualified prospective partners

Volcano plots are great at identifying what's different between samples. They're not designed to explain why those differences exist. Biology starts after differential expression.

Same public Parkinson's disease dataset. Same statistical threshold. Different search pipeline. Original analysis: 2 statistically significant proteins. Tesorai Search reanalysis: 9. What datasets deserve another look?

What if quality control started with a question instead of a workflow? Here's a short demo of asking Tesorai for proteomics QC guidance in natural language. What workflow should we show next? #Proteomics #MassSpectrometry #AI

The biology was always there. Better computational methods made it possible to find. In collaboration with UCSF, Tesorai identified 7,998 protein-verified RNA editing sites—an 18× increase over previous reports. Congrats @dlswaney.bsky.social, @petercim.bsky.social, and colleagues!

Which proteins are changing between my experimental groups? Ask the question in natural language. Tesorai performs the statistical analysis and generates the volcano plot. What's the slowest step in your Olink analysis workflow?

Tesorai completed the analysis in ~7 hours of compute time, enabling the work to be completed before an ARPA-H proposal deadline. "These analyses dramatically accelerated our time from preliminary hypothesis to in silico antigen confirmation." — Nathan Salomonis, PhD @nathansalomonis.bsky.social

A collaborator needed to reprocess the public ImmunoVerse immunopeptidomics collection: • ~4,000 raw MS files (~4 TB) • Custom FASTAs with ~100,000 proteins • Search for non-canonical antigens Conventional workflows would have taken weeks to months.

Some scientific opportunities can't wait weeks for analysis. 👇

In a collaboration with researchers at Notre Dame and NYU, Tesorai Search identified 136% more high-confidence peptide identifications than standard search engines. Resulting analysis contributed to a pan-cancer atlas of 28,446 tumor-specific antigens. Publication: www.biorxiv.org/content/10.1...

In immunopeptidomics, a missed peptide isn't just another data point. It could be a tumor-specific antigen that never gets investigated. That's why sensitivity matters.

Some discoveries start with a single missed peptide. More peptide identifications don't just improve a metric—they can reveal biology that would otherwise remain hidden. 👇

Many analysis workflows require exporting differential expression results into another tool for pathway analysis. With Tesorai, differential expression and GO enrichment happen in the same workflow—so you can move from proteins to biology without leaving the platform.

GO enrichment groups proteins into biological functions and pathways. Instead of looking at individual proteins, you can begin identifying the biological processes underlying your experiment. That's often where interpretation begins.

A volcano plot might identify hundreds of significant proteins. But researchers rarely care about a list of proteins. They want to understand: • Which pathways are changing? • Which biological processes are enriched? • What mechanisms may be driving the phenotype?

A list of proteins isn't the answer. Biology is. Differential expression tells you what's changing. GO enrichment helps explain what those changes mean. In this example, inflammatory and cytokine signaling emerge as some of the most enriched biological processes. 👇

I’ve been re-analyzing a public proteomics dataset and noticed that cancer profiles can look more similar to other cancers than to normal tissue profiles. Is that expected biology, or an artifact of analysis choices?

Large-scale proteomics analysis shouldn't take weeks. Working with frankligy.bsky.social and Mark Yarmarkovich, Tesorai processed 2,500 immunopeptidomics files in under 5 hours—a workflow that previously required more than 4 weeks. Publication: www.biorxiv.org/content/10.1...

Proteomics datasets have grown dramatically in size and complexity. Analysis infrastructure needs to evolve alongside the data.

Honored to see our work highlighted by Dr. Yu-Ju Chen (Academia Sinica) in her #HUPO2025 plenary talk yesterday. We’re thrilled to support her and her team's groundbreaking research in proteomics and lung cancer.

Incredible energy at #HUPO2025! Huge thanks to everyone who has been stopping by the Tesorai booth to see our live demos in action! We’re thrilled by the excitement, thoughtful conversations, and momentum around what we’re building. Grateful to be part of such a passionate proteomics community!