Sid Sijbrandij: collecting the data first

The situation

In November 2022, Sid Sijbrandij was running GitLab, the public software company he co-founded. A pain near his chest started during a bench press and did not go away. A scan at the emergency room found a six-centimeter tumor growing out of his vertebrae. The diagnosis was high-grade osteosarcoma of the thoracic spine¹, at the T5 vertebra².

He went through the standard treatment. Surgeons removed the mass and fused his spine. He had radiation and chemotherapy intensive enough that he needed four blood transfusions¹. He also received an investigational targeted chemotherapy from the startup Shasqi. It was given through a single-patient IND filed with the FDA, the mechanism by which the agency allows one named person to receive an unapproved drug¹. The cancer went into remission.

Then it came back. The timeline on his data site records a recurrence at the Th4 vertebra in June 2024. Another followed at Th4–Th5 in January 2025³. His oncologist told him there was nothing left he could recommend, and no clinical trial fit an adult with his cancer¹. Sijbrandij has written that during his first round of care he left the analysis and decisions to others. The recurrence ended that arrangement⁴. At the end of 2024 he moved from CEO to executive chair of GitLab to focus on treatment².

What he measured

His doctors worked from a familiar rule: order a test when you already know what you will do with the result. After the recurrence he reversed it. In his words, the team “collected data first and figured out what to do with it after”¹.

The list of what they collected is long. Tumor and blood samples were profiled with whole-genome sequencing and whole-exome sequencing, which covers the protein-coding fraction of the genome. They also ran bulk RNA-seq and single-cell RNA-seq. Bulk RNA-seq measures average gene activity across a whole sample, and single-cell RNA-seq measures it cell by cell. That was done at four tumor timepoints. They were the original resection at UCSF in December 2022, two biopsies at UCLA in June 2024 and January 2025, and an excision at Memorial Sloan Kettering in April 2025³. Tissue from the first surgery went through H&E and immunohistochemistry, the standard stains a pathologist reads. It also went through 18-plex immune imaging and Xenium spatial transcriptomics, which maps gene activity to positions within the tissue. His imaging record includes nine whole-body FDG PET/CT scans. It also includes tracers aimed at two specific proteins, EphA2 and B7-H3³. Every month he takes blood-based minimal residual disease (MRD) tests, which look for traces of tumor DNA circulating in blood. A lab at Baylor tracks his immune cells with monthly flow cytometry and single-cell T-cell receptor sequencing¹. His team also tested drugs against organoids, small tissue cultures grown from his own tumor cells¹².

Collecting all of this was mostly a logistics problem. Hospitals routinely preserve tissue in formalin and paraffin. That suits a pathology report but destroys much of the RNA and DNA that sequencing needs. Getting samples frozen instead was a struggle at every institution. So was getting the hospital to release them to him². He calls access to his own tumor tissue one of the hardest parts of the whole effort¹. He hired Jacob Stern, a former 10x Genomics director, to run his care full time. His team kept a running document of every test and meeting. It passed 1,000 pages for 2025 alone².

What the data showed

The finding that changed his treatment came from RNA. Bulk RNA sequencing of the recurrent tumor flagged unusually high expression of FAP, fibroblast activation protein. That alone was ambiguous. FAP is normally a marker of the fibroblasts that surround a tumor rather than of the tumor itself. Single-cell sequencing settled the question. The expression was in the tumor cells¹. Computational biologists who clustered his single-cell data found that his putative tumor cells over-expressed several fibroblast markers, including FAP².

That mattered because a drug existed that targets FAP. Around the same time, one of his concierge medical services surfaced an experimental FAP-targeted radioligand therapy. It was offered by Richard Baum at CURANOSTICUM in Germany¹². A radioligand pairs a molecule that binds a target with a radioactive isotope. Before any treatment dose, the same ligand can carry an imaging isotope to show where it binds. His tumor lit up on that scan. Together with the single-cell result, that gave him the confidence to go ahead².

The data surfaced other candidates too. His team is tracking B7-H3, which has no approved therapy in osteosarcoma. It is the target of antibody-drug conjugates in development for other cancers. They are also running a binder discovery campaign against PANX3, a channel protein expressed on his tumor cells¹.

What he did with it

He went to Germany and had the FAP-targeted treatment twice. He reports 60% necrosis and 20% shrinkage, enough that surgeons could remove the bulk of the tumor¹. He was not taking one drug at a time. Alongside the radioligand he was on dual checkpoint blockade and NK cell therapy. He was also taking an IL-15 superagonist and an oncolytic virus. Earlier he had taken a personalized peptide neoantigen vaccine¹.

The tissue from that surgery let the team measure his immune response directly. When the cancer returned, 19% of the immune cells infiltrating the tumor were T cells. After the radioligand therapy and surgery, 89% were². Sijbrandij is careful about what this means. With several treatments running at once and a single patient, he writes, no one can say with certainty which one worked¹. The longitudinal data does support hypotheses. One analysis of his single-cell data points to neutrophils that shifted toward pro-tumor signaling during immunotherapy and partly reversed after the FAP-targeted therapy¹.

Since then, the goal has been to keep the cancer from coming back. He received a personalized mRNA vaccine designed around neoantigens from his tumor, the mutated proteins unique to his cancer. It was built by a team at Houston Methodist. He was the first patient in an investigator-initiated trial, and it took six months from the first conversation to the first dose¹⁵. Houston Methodist announced the work in May 2026 as the first personalized mRNA cancer vaccine for osteosarcoma. It was given under a compassionate use allowance from the FDA⁶. A personalized TCR T-cell therapy has been manufactured and a CAR T-cell therapy is in development. Both are held in reserve¹.

As of August 2026, he reports no evidence of disease for more than a year. That is based on negative monthly MRD tests and regular CT and biomarker scans. He does not describe himself as cured¹.

In January 2026 he began publishing his data at osteosarc.com⁴. The dataset is listed on the AWS Registry of Open Data under a CC0 public-domain license⁷. The site’s file catalog describes about 25 TB³. By August he put the total at more than 30 TB¹.

What others can take from it

Most of what Sijbrandij did is out of reach for almost anyone. He says so himself. He credits money, access, and luck for part of his result¹. The treatments are his and his doctors’ decisions. The habits around the data are the part worth copying.

Get the raw data along with the report. A report is one lab’s reading of a file. His team reanalyzed some of his off-the-shelf MRD results and found critical errors in the reports². That is only possible if the underlying data sits somewhere you control.

Look at the same question more than one way. The FAP signal showed up in bulk RNA. Single-cell data narrowed it to tumor cells. Imaging then confirmed it in his body before anyone committed to treatment. Each layer could have contradicted the one before it.

Measure over time, starting before you need to. The 19% to 89% comparison only exists because tissue was profiled at recurrence and again after treatment. His monthly immune data is read against healthy donors¹. A single snapshot tells you where you are. A series tells you what changed.

Plan for the sample before the procedure. Whether tissue is frozen or fixed is decided in the operating room and pathology lab. The patient often does not know a choice exists. Once it is in paraffin, most of the sequencing value is gone.

Write everything down in one place, and be plain about what the data cannot tell you. His team’s notes tie every test to every decision. His own writing is just as clear that one patient on parallel treatments produces hypotheses, and that proof takes more than one person¹.

Footnotes

1. Sid Sijbrandij. From “default dead” to living the future of cancer care. The Cancer Letter, 2026. https://cancerletter.com/guest-editorial/20260814_2/ ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵ ↩¹⁶ ↩¹⁷ ↩¹⁸ ↩¹⁹ ↩²⁰ ↩²¹ ↩²²

2. Elliot Hershberg. Going Founder Mode On Cancer. The Century of Biology, 2026. https://centuryofbio.com/p/sid ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

3. Sid Sijbrandij. Sid Sijbrandij’s Osteosarcoma Data. osteosarc.com, 2026. https://osteosarc.com/ ↩ ↩² ↩³ ↩⁴

4. Sid Sijbrandij. I’m going Founder Mode on my cancer. Sid’s Substack, 2026. https://sijbrandij.substack.com/p/im-going-founder-mode-on-my-cancer ↩ ↩²

5. Sid Sijbrandij. Cancer care updates. Sid’s Substack, 2026. https://sijbrandij.substack.com/p/cancer-care-updates ↩

6. Houston Methodist. Houston Methodist develops first personalized mRNA cancer vaccine for osteosarcoma. Houston Methodist Newsroom, 2026. https://www.houstonmethodist.org/newsroom/houston-methodist-develops-first-personalized-mrna-cancer-vaccine-for-osteosarcoma/ ↩

7. Registry of Open Data on AWS. Sid Sijbrandij’s osteosarcoma dataset. Registry of Open Data on AWS, 2026. https://registry.opendata.aws/sid-osteosarc/ ↩