Company
Jindu Biotech / OxTium (津渡生科)
A Shenzhen cfRNA foundation-model startup pitching itself as an agent-orchestrated autonomous lab.
1. Core Product / Service
Shenzhen Jindu Biomedical Technology (English name OxTium Technology) sells a three-layer stack.
- GeneLLM: a multi-omics model pretrained directly on raw sequencing reads, with no alignment to a reference genome. It tokenizes ~150 bp RNA reads into 7-mers. The company reports 1.5B parameters trained on 3.5T bases, with a 30B "XLarge" version in training.
- BioFord: a cloud bioinformatics platform.
- BioFord Agent / Harness: five agents (literature, design, reasoning, analysis, scheduling) plus a "universal instrument abstraction layer" that drives sequencers, liquid handlers, PCR/plate readers, flow cytometers and LC-MS. The company says it connects to "30+ instruments."
Jindu builds no hardware. The peer-reviewed output is cfRNA liquid-biopsy diagnostics, published in three journals:
- multi-cancer classification in Nature Communications (2026; colorectal, gastric, liver and lung cancer; reported AUC ≈0.98)
- ovarian-cancer chemoresistance prediction in Advanced Science (2026)
- preterm-birth prediction, a BGI-led study, in npj Digital Medicine (2025)
2. Target Users & Pain Points
- Hospitals and genomics labs that want cheaper cfRNA diagnostics. The company claims AUC above 0.8 at 1 Gb shallow sequencing versus the usual 6 Gb, which it says cuts sequencing cost by ~83%.
- University and institute labs that want literature-to-experiment loops across instruments from different vendors. The company claims design cycles fall from months to about a week, with up to 40% less manual work. No named lab backs these figures.
- Per Dealroom, revenue today comes from bioinformatics compute services and custom algorithm work, not platform subscriptions.
3. Competitive Landscape
| Player | Positioning | Vs. Jindu |
|---|---|---|
| dp-technology | AI4S platform with the open-source Uni-Lab-OS instrument layer | Overlaps directly with Jindu's instrument-abstraction pitch, with far more resources and an open-source distribution edge |
| xtalpi | Listed AI drug discovery company with large robotic labs | Owns physical lab capacity and pharma contracts; Jindu has neither |
| megarobo | Lab-automation integrator | Delivers working hardware plus scheduling software; Jindu sits only at the software and orchestration layer |
| hetan-ai | "Robot scientist" built around a purification robot | Narrower, more verifiable claim (a named pharma pilot) |
| FutureHouse / Biomni / Lila Sciences | Global science-agent and autonomous-lab players | Set the narrative Jindu borrows; Lila alone has raised ~US$550M |
| GRAIL / Exai and Chinese cfDNA/cfRNA screening firms | Liquid-biopsy early detection | The market Jindu's actual publications address |
4. Unique Observations
- Narrative drift. Every peer-reviewed result is a cfRNA diagnostic model. The "embodied autonomous research / physical AI" framing appeared in 2025–26 alongside the fundraising wave, and no paper, named customer or install count supports it yet.
- Genuine technical kernel. Pretraining on raw reads to work at shallow sequencing depth is a defensible, clinically relevant idea. The co-authors are real institutions (BGI/MGI, Beijing Hospital, Peking University First Hospital, Fudan Cancer Center). The Nature Communications paper was first posted as a 2024 preprint, so the work predates the agent pitch.
- Instrument abstraction is not a moat. Cross-vendor drivers are being built by DP Technology, the integrators and LIMS vendors alike. Jindu's edge, if any, is the omics model, not the scheduler.
- Inconsistent numbers. Sources disagree on the founding date (2022 vs 2023), the round count ("4 rounds in 2025" vs "6 rounds in 14 months"), round labels, and AUC figures (preterm 0.85 vs 0.89; multi-cancer 0.98 vs 0.9962). A claimed "~1M overseas registered users" has no stated definition.
- Founder background. The CEO's Oxford PhD under Hagan Bayley was in 3D tissue bioprinting, not AI or automation. The modelling work is led by the CTO and CSO.
5. Financials / Funding
- Revenue / orders: not disclosed.
- Rounds (mostly aggregator data, amounts in RMB):
| Round | Date | Amount | Lead / investors |
|---|---|---|---|
| Seed | Mar 2023 | not disclosed | Xiantong Capital (single source) |
| Angel | Nov 2024 | not disclosed | Leaguer Venture (single source) |
| Angel+ | Oct 2025 | tens of millions | Sequoia China Seed Fund |
| Pre-A | Oct 2025 | ~10M+ | Cowin East (创东方) lead, Jiangxi Financial Holdings |
| Pre-A+ | Nov 2025 | ~10M+ | Nanshan Strategic Emerging Industry Fund |
| Series A | Jun 2026 | nearly 100M | Gaotejia (高特佳) lead |
- A Shenzhen municipal grant of roughly RMB 10M+ is reported by a single source.
- Dealroom lists a US$55M post-money valuation for the Series A. This is likely an estimate; no primary source confirms it.
6. People & Relationships
- Founder & CEO: 金泳成 (Jin Yongcheng). Oxford DPhil in biomedical engineering (2022) in Hagan Bayley's lab; also heads a medical-intelligence centre at the Research Institute of Tsinghua University in Shenzhen.
- CSO: 沙磊 (Sha Lei). Professor at Beihang University's AI institute and former Oxford research associate.
- CTO: 邓司伟 (Deng Siwei). Oxford computational-biology DPhil and first author of the GeneLLM papers.
- COO: 周天尧 (Zhou Tianyao).
- Partners: BGI / MGI (a joint medical-AI centre at China National GeneBank, reported Jun 2024) and PUMCH Cancer Center. Clinical co-authors include Beijing Hospital, Peking University First Hospital, Sir Run Run Shaw Hospital, Fudan University Shanghai Cancer Center and Fujian Maternity & Child Health Hospital.
- Peers: dp-technology, xtalpi, megarobo, hetan-ai, shuye-tech.
Last compiled: 2026-10-12