The Indian Canvas #41: India funds AI drug candidates. Who makes the cell data the models lack?

On 12 October 2026 India's Technology Development Board signed Rs 86 crore of support for Bengaluru's Peptris to move four AI-discovered drug candidates towards the clinic. The same week, Paris-based Rivercell launched with USD 25 million to do something else: generate its own time-resolved cell data to train a model of how cells respond to treatment. India is not empty on AI for drugs. What I did not find is anyone selling that kind of experimental cell data to Indian AI teams. That gap is a hypothesis, not a finding.

India is paying for the model-to-clinic step

A report of the agreement says the Technology Development Board (TDB), under the Department of Science and Technology, signed Rs 86 crore of support for Peptris Technologies from the Research Development and Innovation fund, in the form of optionally convertible debt. The project, called VIGOR, costs Rs 172 crore in total and aims to take four drug candidates from technology readiness level 4 to 6, through preclinical work and preparation for trials.

That is money for development after a molecule is designed. Peptris's own site says its platform designs molecules while "significantly reducing the need for wet lab testing". I did not see any part of this grant going to generating new biological training data, but I have only read the announcement, not the agreement.

Stock photo of a researcher with a sample. Not Rivercell, Peptris or any Indian lab.

Rivercell's bet: the missing piece is data

Alven's announcement of 13 October says Rivercell raised a USD 25 million seed led by HV, with Alven, HCVC and Bpifrance's digital venture arm. HV, HCVC and Bpifrance Digital Venture each posted about the round too. The company's launch release is on Business Wire.

Per Alven, Rivercell is building a platform that generates multimodal, time-resolved data on how cells change under treatments and genetic changes, and uses it to train an "AI Virtual Cell", a model that predicts cell responses on a computer. The founders are Yann Fleureau, who co-founded Cardiologs and led it through its acquisition by Philips, and Eric Durand, formerly of Novartis and Owkin. All of this is company-reported. The release does not say how big the dataset is or when a model will be ready, and I did not find either elsewhere. It is seed stage and unproven.

India is crowded at the model layer

I checked the Indian players on their own sites. Peptris sells AI-led molecule design. Cellworks says it models how an individual patient will respond to a drug by combining mechanistic biology, clinical guidelines and more than 300 million patient records, which is simulation on existing records, not new lab data. Aganitha says it applies AI and deep tech to biopharma research, with more than 100 collaborations over seven-plus years.

On the lab side, the C-CAMP high-throughput screening and imaging facility offers cell-based assays, siRNA and small-molecule screens and a high-content imaging platform to academia and industry. So screening services exist. What I did not find, in a short search, is an Indian company running automated experiments over time to build large cell-response datasets for model training. A stealth company or a private pilot would not show up. "I did not find one" is the limit.

Who pays, and where it could fail

My hypothesis: the buyer is an Indian AI drug-discovery team that has models but few proprietary cell measurements, paying per plate or per perturbation set for data made to its design. The missing piece would be a contract lab built around time-resolved readouts and clean metadata. I have not surveyed these companies, so I do not know whether they want bespoke data or reuse public datasets.

It could fail on cost, since Rivercell has USD 25 million to build its own instruments and an Indian seed company would not. It could also fail on demand, since Rivercell's own premise that a cell model needs data that does not yet exist is still unproven. A smaller start would be a contract service on rented instruments. Any per-plate price is a test assumption, not a market price.

What I would do first

Ask three Indian AI-drug teams whether they have ever paid for experimental data, and what they would need to see. Ask C-CAMP what time-resolved imaging runs they have done. If nobody has paid for data, the idea starts as a services pilot with one customer, not a platform.

Sources and photo credits

TDB-Peptris report, Alven announcement, Business Wire release, HV post, HCVC post, Bpifrance Digital Venture post, C-CAMP facility page, company sites linked above. Photos on Unsplash with no stated location: cover by Adam Bezer (@bezer), scientist uses pipette in a laboratory; body by Faustina Okeke (@thefourthwxll), scientist examining a small sample in a lab. Free to use under the Unsplash License. This is a research view, not a proven business.