Blinded checks of biomarker tests on banked Indian patient samples
TL;DR Cohortmark checks biomarker signatures and diagnostic tests blind, on banked and consented Indian patient samples, and nobody involved has a stake in the answer. Diagnostics firms, pharma teams and academic spin-outs pay per study. It works because finding a biomarker is no longer the weak step - proving it holds up on new patients is.
Most biomarkers look great in the first study and then fall apart on fresh patients. A 2026 review in Signal Transduction and Targeted Therapy lists the usual suspects - batch effects, platform drift, overfitting when samples are few, and AI models that quietly learn a site's quirks instead of the disease[1].
Indian work shows the same pattern. A 2026 study from a private university and AIIMS reported an AI-checked lupus flare signature built on 16 patients, and its own authors say it needs larger cohorts before anyone relies on it[2]. So what happens next? Usually nothing, because few Indian groups have a cheap and neutral way to run that second study and the promising signal just sits there.
What it does, in regulatory terms: research-use validation studies only. Cohortmark does not sell a diagnostic or report results to patients. When a client's test is meant for clinical sale, the study is designed to feed the client's own CDSCO performance-evaluation filing under the Medical Devices Rules 2017[3].
Here's how it works. A client hands over a locked model or assay protocol and Cohortmark freezes it, then runs it on a sample panel the client never sees, drawn from partner hospital biobanks with consent and ethics approval. Before anything is unblinded, software checks the model for leakage, site effects and calibration drift. The client gets a signed report with confidence intervals by age, sex and region.
The seed pitched an AI platform that finds biomarkers. We turned it around, because discovery is crowded and hungry for data while independent validation is the step the literature keeps saying is missing. Selling that one step is narrower and a lot easier to price.
Indian IVD makers preparing CDSCO files pay, and so do pharma teams checking a companion marker on Indian patients and university spin-outs that need credible data before a seed round. Hospitals supplying samples earn a fee per use and co-authorship. My estimate is ₹15–60 L per study depending on sample count and assay. I couldn't find a public price list for comparable Indian services.
CDSCO licensing reached Class C and D devices on 1 October 2023, so more IVD makers need Indian performance data[3]. DBT's Indian Biological Data Centre now runs controlled-access genomic portals[4], which makes compliant data handling much easier to set up. And AI-built signatures are multiplying faster than anyone can check them[1].
There's no public figure for India's biomarker validation services. The firmest signal is regulatory - every new IVD sold in India now needs a licence, and new tests need clinical performance evaluation approved through Form MD-24 and MD-25[3]. As an adjacent number, one large genomics lab reports more than 600,000 exomes and genomes sequenced and partnerships with 8,000+ hospitals[5], which tells you how much sample flow already exists.
That same genomics lab lists multiomics research services for pharma[5], but it also sells its own tests, and that limits its neutrality. Indian assay makers publish real-world data on their own panels, such as one cancer panel in a 2026 journal paper[6]. Global CROs run validation too, just not on Indian banked samples at Indian prices. The gap is a neutral validator with Indian cohorts.
| Difficulty | Hard (55/100) |
| Starting capital | ₹1.2–2.2 Cr |
| Time to launch | 9–12 months |
The software is the easy part. The hard part is a sample network, which means hospital biobanks with consent forms that allow commercial reuse, clean clinical labels and stored samples that are still usable. You also need lab partners whose assays give the same answer twice and a reputation for saying no to clients, which is harder to build than it sounds.
You don't need a diagnostic licence for research-use studies, but your clients do, and your studies have to stand up in their filings.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
AI shouldn't decide whether a test is clinically fit. That call belongs to the client's regulatory filing and to CDSCO, and your statistician owns every number in the report.
Want to work out the AI layer for Cohortmark? Talk to Yash →
Here's what can go south -
I'd back this for a founder from clinical research or a diagnostics lab who already knows biobank heads by name. A pure ML team will struggle, honestly. Before you raise money, prove two things - that two hospitals will sign a sample-sharing agreement and that one client will pay for a pilot study.
Plan for roughly ₹1.2–2.2 Cr, mostly on biobank partnerships, sample costs, statisticians and the analysis platform.
Research-use studies don't need a diagnostic licence, but each study needs ethics approval and proper consent. Clients selling a test still need CDSCO approval.
Yes, independent validation is needed everywhere. You'd work with local biobanks and ethics committees and design studies to fit your regulator's rules for diagnostics.
The free Cohortmark validation kit has the line-by-line capital breakdown, the effort scoring, the CDSCO, ethics and data-protection checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for diagnostics firms and biobanks, and 5 assumptions to test. Download the kit →
Thinking about building Cohortmark or something close to it? The make-or-break work is signing hospital biobanks on consent terms that allow paid studies, and that's the kind of thing I help founders sort out. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.