Research software · Bioimage analysis
Biodock
A cloud platform that lets a biologist train an image-analysis model without writing code. The strongest evidence for it is not a press release. It is other people’s methods sections.
Three minutes and fifty-nine seconds
A cystic fibrosis laboratory had a counting problem. Every mouse experiment ended with a stained slide of lung lavage fluid and a person at a microscope sorting cells into three piles: neutrophils, macrophages, lymphocytes. Two to three hundred cells per mouse, then the next mouse.
In 2025 the group published what happened when they handed the job to software. They uploaded 189 training images, drew 18,259 neutrophil labels, 3,317 macrophage labels and 1,133 lymphocyte labels, trained five versions of a model, then ran a fresh test set of 86 images from 20 mice it had never seen.
“The analysis took 3 min and 59 s to complete.”
The counts held up. Agreement was checked with a Bland Altman plot, a Spearman correlation and a t test. The authors concluded that the model “can accurately produce the percentages that correlate well and have no bias or significant difference with manually differentiated percentages performed by trained professionals in a fraction of the time and without human bias.”
Nobody on that team wrote code. That is the product.
The bottleneck nobody funds
Microscopes got cheap and fast. Analysis did not. A single plate of organoids holds thousands of objects somebody has to find, classify and measure, usually the graduate student who generated the data.
Free tools exist and are good. ImageJ and Fiji, CellProfiler, Icy, Ilastik and QuPath all handle serious work, and Biodock’s own blog catalogues them at length rather than pretending they do not. The gap is stated flatly in that 2025 paper: “Knowledge of programming and costs bar these tools from being used on a larger scale.”
Biodock sells the missing middle: upload images, label objects, click train, click run, get a table with area, perimeter, solidity, eccentricity, position and per-channel intensity for every object. The documentation says the architectures “are usually based off of vision transformer bases.”
Two Stanford labs across a hallway
Nurlybek Mursaliyev won medals at national biology olympiads in Atyrau, Kazakhstan, then left for James Madison University and a Stanford PhD in cell and molecular biology. His thesis question was how obesity affects embryo development, and he built a device that grades embryo quality by density. He had already co-founded SmartLens, a contact lens that measures intraocular pressure in glaucoma, and stepped back from it to finish the degree.
The other founder was Michael Y. Lee, a deep learning researcher at Stanford and a co-author on CellSeg, an open source nucleus segmentation tool from Garry Nolan’s lab. The two worked across from each other. As Lee put it in the round announcement: “As a PhD student, my cofounder Nurlybek spent hours manually counting through lipid droplets in microscope images of embryonic tissues.”
The choice of what to build was deliberate. Mursaliyev had priced the alternatives: “Based on my past experience, I knew building a therapeutic or diagnostic company would take years and require massive amounts of capital. So I focused on software for pharmaceutical companies. It’s a faster path to revenue.”
One loud year, then a quiet one
The minimum viable product came together in the summer of 2020 and a first client signed within months. Y Combinator took them for the Winter 2021 batch, and the money arrived faster than the program did: “Yes, and actually, most of the round was closed in just a couple of days, even before Demo Day.” They skipped Demo Day and later called that a mistake.
On March 30, 2021 the company announced $2.1 million in pre-seed funding led by Andreessen Horowitz, with TQ Ventures, Soma Capital and angels including John Curtius and Zach Weinberg. That is the last funding announcement on the record. In January 2026 Mursaliyev put the five year total at $3.8 million, leaving $1.7 million with no public paper trail.
The company left the Bay Area for Austin, and the founder is unsentimental about it. “The main reason was the high cost of living in Silicon Valley,” he said, before adding the part most founders leave out: “But now, I wouldn’t recommend doing the same.” By 2026 the Y Combinator page lists one active founder, Mursaliyev, and the open roles sit in Astana, Kazakhstan. In August 2026 the platform was rebuilt with an AI assistant for querying results, mobile access, Python scripting on outputs, and Public Models that let labs share trained models.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Independent methods citations | A Europe PMC search on Sep 23, 2026 returned 57 records, 50 indexed in PubMed. Of 33 open access full texts retrieved, 29 describe using the platform. Journals include PNAS, Nature Neuroscience, Nature Communications, Molecular Cell, Cell Stem Cell, eLife and Acta Neuropathologica. | Public record |
| Reach into clinical tissue | Acta Neuropathologica, 2026. Whole slide images of human post mortem brain were uploaded to Biodock and beta amyloid deposits classified by morphology. | Peer reviewed |
| Paid commercial work | Human Vaccines and Immunotherapeutics, 2023. The disclosure reads: “ML and NM are paid employees of Biodock Inc, which carried out work for the study on a contract basis paid by Vaxxas Pty Ltd.” | Public record |
| Name collisions in the citation count | Four of the 33 full texts refer to something else: three to a UVP BioDock-It 2 gel imager from Analytik Jena, one to an older docking program called BioDock. Any tally that skips this filter overstates adoption. | Public record |
| Installed base | January 2026: “Currently Biodock is used by over 5,000 labs worldwide”. In March 2021 it was hundreds of institutions. Neither figure is auditable. | Company-stated |
| Named enterprise customers | Genentech and CSL Behring appear on a third party startup directory. No release, filing or paper confirms either. Vaxxas is the only customer named in a primary source. | Unverified |
| Federal grant | The founder says he and his co-founder received a National Science Foundation grant. The NSF awards database returns no award under the company name, and NIH RePORTER returns none. It may sit under a university or personal record. | Differs from record |
| Accuracy claim | The 2021 round announcement says the modules are “up to 50% more accurate than other software solutions”; the Y Combinator listing says “30-50% more accurate analysis”. No published benchmark against a named competitor was found. | Company-stated |
Read plainly: the adoption is real and it is other people’s adoption. Nearly thirty independent groups have put Biodock in a methods section where a reviewer could challenge it, and one of them tested it against human counting and published the comparison. What is not on any record is the business: one announced round in five years, no Form D, no audited customer list, and two vendor revenue estimates that disagree by a factor of two. The science is checkable. The company is not.
What to watch
- Whether a second round ever appears, and whether a Regulation D notice is filed when it does.
- The citation curve. The verified papers cluster in 2025 and 2026, which suggests adoption is still compounding rather than flattening.
- Whether any enterprise customer beyond Vaxxas is confirmed by a primary source.
In their words
“We’re proud that the Biodock platform is AICPA SOC 2 Type II certified, the gold standard for cloud security.”
Biodock security page, accessed 2026 · Company
“Through multiple rounds of training and refinement, we have created a tool that is as accurate as manual review of slide images while removing the subjectivity and making the process mostly hands off, saving researcher time for other tasks and improving core turnaround for experiments.”
Williams, Faber and Kelley, Journal of Pathology Informatics, 2025 · Peer reviewed
“Our workflow demonstrates how to implement computer vision using limited computational resources and minimal programming expertise by leveraging cloud-based, user-friendly ML tools (e.g., Biodock) integrated into local scripts.”
Bell and colleagues, Scientific Reports, 2025 · Peer reviewed
“Currently Biodock is used by over 5,000 labs worldwide, including large pharmaceutical companies and biotech companies, mostly in the US, Europe, Southeast Asia, and Australia.”
Nurlybek Mursaliyev, Digital Business interview, January 2026 · Company-stated
“But now, I wouldn’t recommend doing the same. If you’re building a startup, it’s better to stay in the Bay Area.”
Nurlybek Mursaliyev, on leaving Silicon Valley for Texas, Digital Business, 2026 · Interview
Related companies
Sources
- Public recordEurope PMC literature search for Biodock, 57 records
- Public recordAI image processing versus manual leucocyte differential
- Public recordAutomated computer vision for an ex vivo precision medicine platform
- Public recordTGFBR2 coordinates the endometrial response to estrogen
- Public recordChoroid plexus apocrine secretion shapes CSF proteome
- Public recordAlzheimer pathology with R47H and R62H TREM2 variants
- Public recordEndometrial cells with high ALDH activity
- Public recordSelf-administration potential of high-density microarray patches
- Public recordCellSeg segmentation software, competing interests statement
- Public recordSenescence biomarkers in a motor neuron disease mouse model
- Public recordDeep blueprint, a guide to automated image classification
- Public recordEDGAR full text and company search, no filings found
- Public recordNSF awards database search, no awards found
- IndependentKazakh scientist moves to the US and builds a biotech startup
- InterviewNurlybek Mursali on starting Biodock
- IndependentPhysicians should build their own machine-learning models
- CompanyAnnouncing Biodock’s pre-seed round
- CompanyThe new Biodock is live
- CompanyHome, security and documentation pages
- CompanyCompany page and open roles
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
