APPENDIX CANCER AI BIOMARKER RESEARCH, IN PLAIN LANGUAGE
A new appendix cancer AI biomarker concept is making its way through the research world. A preprint from MD Anderson, posted in May 2026, describes a deep learning model called TmSNet…
A new preprint from researchers at MD Anderson, posted in May 2026, describes a deep learning model called TmSNet. The model predicts something called Tumor-specific total mRNA expression, or TmS for short. TmS is a single number that captures how much RNA a tumor’s cells are producing compared with normal tissue. Tumors that produce more RNA tend to behave more aggressively. They divide faster, adapt to treatment more easily, and recur more often.
The study did not include AI biomarkers in appendix cancer. It was trained on twelve cancer types from public databases, and validated on breast cancer cohorts. So nothing in this paper applies to our community today. But the idea behind it is one of the more interesting concepts I have seen in a while, and it points at a problem appendix cancer patients know well: pathology and mutations do not always predict who does well and who does not. AI biomarkers can help.
The problem TmS is trying to solve
Two patients can walk into clinic with the same pathology report. Same grade. Same subtype. Same surgical findings. One stays stable for years. The other progresses fast. That gap, between what the pathologist sees and what the cancer actually does, is one of the hardest problems in oncology. It is especially hard in appendix cancer, where the disease spans indolent mucinous tumors all the way to aggressive signet ring and goblet cell carcinomas, and where the same label can hide very different biology underneath. That gap is exactly the kind of problem researchers hope an appendix cancer AI biomarker could eventually help solve.
The authors of this study argue that mutations alone do not explain this. Two tumors with the same KRAS mutation can behave completely differently. What might explain it, they say, is the overall transcriptional state of the tumor. Transcription is the step where DNA gets copied into RNA, which is then used to build proteins. Aggressive tumors tend to ramp transcription up across the board, not just on a single gene. TmS measures that overall activity as a continuous score.
| Mutation-centric frameworks fail to capture the system-level transcriptional heterogeneity of evolving tumors. |
Why this AI biomarker concept matters for appendix cancer.
Appendix cancer has very few validated biomarkers. CEA, CA19-9, and ctDNA help in some situations but all have real limits. Many appendiceal tumors have quiet genomes. KRAS and GNAS mutations show up often but do not fully predict outcomes. And the grading system itself has known gaps, including the ongoing problem of moderately differentiated mucinous adenocarcinoma being collapsed into a single high-grade bucket in much of the research literature.
A continuous biologic score, one that integrates RNA, methylation, immune context, and microRNA signals, is the kind of tool appendix cancer needs. The authors describe TmS as lineage-agnostic, meaning it is not tied to one tissue of origin. In theory, that is exactly the kind of metric that could help stratify a disease as heterogeneous as ours, where tumors range from slow-moving mucinous disease to rapidly recurring carcinomatosis.
| What TmS was associated with in earlier work Activation of metabolic programs that support rapid growth, including the pentose phosphate pathway and glycolysisTranscriptional amplification tied to MYC signaling, a well-known driver of aggressive cancerStem-like, less differentiated cell states and epithelial-to-mesenchymal transitionIn triple-negative breast cancer, distinct patterns of immune cell infiltration versus stromal exclusion |
Those features overlap heavily with what appendix cancer specialists recognize clinically. Aggressive signet ring disease, rapidly recurring peritoneal disease, chemotherapy-resistant tumors, and high-grade dedifferentiated appendiceal cancers all share traits with the high-TmS profile the authors describe. The overlap is conceptual, not proven. But what AI biomarkers tells us is striking.
How the AI part works, briefly
Calculating TmS the original way requires matched DNA and RNA sequencing and a computationally heavy pipeline. Most clinical samples do not come with all of that. What TmSNet does is predict the TmS score from data that is easier to get: RNA expression, DNA methylation, microRNA, and immune cell proportions. It uses a deep learning architecture that combines feature selection with two types of neural networks to handle the high-dimensional inputs in AI biomarkers.
Across the twelve cancer types tested, the model performed well internally, with concordance correlation coefficients up to 0.93 in head and neck cancer and 0.92 in bladder cancer. Performance dropped in external validation, especially when DNA methylation data was not available. That is a real limitation in the study of AI biomarkers. So is the fact that mucinous tumors, which dominate the Appendix Cancer community, were not in the training data and likely have biology the model has never seen.
What this is not
This is a preprint. It has not been peer reviewed. It has not been tested in any appendiceal cancer cohort. The model was built and validated mostly on common cancers like breast, lung, colorectal, and bladder. Pseudomyxoma peritonei, LAMN, goblet cell adenocarcinoma, and signet ring cell carcinoma were not part of this work. None of this is ready to inform clinical care for appendix cancer patients.
Multi-omic testing is also expensive. Most appendix cancer patients do not have access to the kind of RNA sequencing, methylation arrays, and microRNA profiling this approach requires. Datasets for our disease are small, scattered, and often incomplete. Any future application to AI biomarkers in appendix cancer would require new training data, validation in appendiceal cohorts, and infrastructure that does not exist yet. No validated appendix cancer AI biomarker exists today, and TmSNet is not one.
| Cancer aggressiveness may be measurable as a whole-system transcriptional state, not just a mutation list. |
Why I am writing about it anyway
The reason I am sharing this is the AI biomarker concept, not the model. The idea that tumor behavior is a system-level property, something that emerges from the combined activity of many genes, methylation patterns, and immune signals working together, is the direction biomarker research needs to move for rare cancers like ours. We do not have the cohort sizes to find single-gene biomarkers the way more common cancers do. What we might have, eventually, is access to integrated scores that combine many weak signals into something clinically useful. Artifical Intelligence makes this much faster.
For patients living with appendix cancer right now, this does not change anything. For the research field, it is a useful signal of where things are headed. If a future appendix cancer AI biomarker is trained on appendiceal cohorts, if multi-omic profiling becomes more accessible, and if the score holds up in mucinous and signet ring disease, then we could be looking at a tool that helps with surveillance decisions, surgical timing, chemotherapy selection, and clinical trial matching. That is a long road. But it is a road worth watching. Again, using AI in the study of Appendix Cancer biomarkers make it much faster.
That’s why contributing your data to the Global AI Appendicure Patient-Led Data registry is so important. We need to be included in these studies.
| Questions to ask your doctor Is multi-omic profiling, including RNA sequencing or methylation analysis, available or being studied for my type of appendiceal cancer?Are there any clinical trials open to appendix cancer patients that use integrated biomarkers like RNA expression signatures to guide treatment?How are decisions about surveillance frequency and adjuvant chemotherapy currently being made for my specific subtype and grade?Is my tumor tissue being banked or made available for future biomarker research? |
Glossary
TmS (Tumor-specific total mRNA expression): A score that measures how much RNA tumor cells are producing overall, compared with normal cells. Higher scores have been linked to more aggressive behavior in several cancer types.
Multi-omic data: Information collected from several layers of biology in the same tumor, including DNA, RNA, methylation, and immune cell makeup.
Methylation: A chemical modification of DNA that affects which genes are turned on or off without changing the underlying sequence.
Preprint: A research paper that has been posted publicly but has not yet been peer reviewed or published in a journal.
Lineage-agnostic: Not tied to a specific tissue of origin. A lineage-agnostic biomarker could in principle apply across many cancer types.
Source
Paul A, Lal JC, Ji S, et al. Tumor cell specific total mRNA expression informed neural networks predicts cancer progression. bioRxiv preprint, posted May 6, 2026. doi: 10.64898/2026.05.01.722212. Not peer reviewed. AI Biomarkers.
| Share this with your community If this resonates with you, or if you know someone navigating appendix cancer who would find this useful, please share it. Appendicure exists because patients and caregivers in this community keep pushing for better information, better research, and better care. You can find more resources, the latest research summaries, and ways to connect at appendicure.com. |
Keep Reading: ctDNA is changing how we think about surgery in appendix cancer
Leave a Reply