Venomics AI Bets $2.7M That Nature's Venom Is the Next Drug Library

Fewer than 0.1% of an estimated 20 million venom peptides have ever been studied. The rest sit in public databases that hold, between all of them, roughly 40,000 to 60,000 entries — scattered across repositories that do not talk to each other.

Venomics AI is a $2.7 million bet that the next drug library will not be designed from scratch. It will be read out of evolution, one wet-lab label at a time.

The Bengaluru company spun out of the Indian Institute of Science and closed its first institutional round on 5 November, co-led by Speciale Invest and South Park Commons, with Operators Studio joining.

The library evolution already validated

Venom peptides evolved to hit precise targets, fast. Most of those targets are ion channels — the gates that control electrical signalling in every cell and govern pain, heartbeat, and insulin release. Small-molecule drugs bind deep inside these receptors, where subtypes look nearly identical, so they often hit the wrong one. Venom peptides sit on the outside, where the subtypes differ.

The distinction is not academic. Blocking Nav1.7 silences pain. Blocking its near-identical cousin Nav1.5 can stop the heart. Tarantula and centipede peptides have shown more than 1,000-fold selectivity between the two.

Three drugs already prove the mechanism at commercial scale. Captopril, inspired by bradykinin-potentiating peptides in the Brazilian pit viper, became the first ACE inhibitor on the market in 1981. Exenatide, built on exendin-4 from Gila monster saliva, became the foundational GLP-1 receptor agonist. Ziconotide, derived from a cone snail toxin, is a non-opioid analgesic with roughly 1,000 times the potency of morphine.

What has been missing is throughput. For decades, searching venom meant one painstaking experiment at a time.

What the money is actually for

Venomics AI is building the Venom Atlas: a proprietary, multi-omics map of Indian venom peptides, each tied to experimentally measured activity. Every venom it collects is sequenced, characterised, and screened against human drug targets in its own wet lab.

The models then do the first pass. Built on open foundations such as ESM and AlphaFold, they rank hits, flag likely toxicity, and eventually redesign venom scaffolds to be safer and more selective. Each screen sharpens the models, and the models choose the next screen.

As we wrote in October, the hit rate in AI drug discovery comes from validated labels, not from generating more molecules. Reflector Bio's 80% hit rate made the same point from the opposite direction: a small, well-measured dataset beat a large, unvalidated one.

That is the difference between asking a model to find a needle in a haystack of billions of theoretical molecules and handing it needles that evolution has already tested under the harshest conditions available.

The numbers that keep the claim honest

The strongest evidence for the approach does not come from the startup. It comes from a 2025 Nature Communications study by Guan and colleagues, which built an antimicrobial discovery pipeline on venom data.

The team mined 16,123 venom proteins and generated 40,626,260 venom-encrypted peptides. From those, they identified 386 candidates structurally and functionally distinct from known antimicrobial peptides. Of the 58 peptides selected for experimental validation, 53 showed potent antimicrobial activity against clinically relevant strains. In a murine model of Acinetobacter baumannii infection, lead peptides reduced bacterial burden without observable toxicity.

Source: Venomics AI / Speciale Invest, 2026

The caveats are just as specific. Roughly 60% of the identified peptides were predicted to modulate ion channels, including some tested in vivo — a toxicity signal that demands electrophysiology work. The APEX model relies on discrete minimum inhibitory concentration values recorded in multiples of two and on AAindex features only, which limits both accuracy and generalisability. Its predictions are also hard to interpret: it does not say which sequence features drive activity.

In other words, the filter is not solved. It is improving.

The permit moat

India is one of the richest biodiversity regions on Earth, and barely a sliver of it has been harnessed for medicine. Access is also tightly governed. Speciale Invest notes that under India's amended Biological Diversity Act, foreign-controlled entities need prior approval to use Indian biological resources, including isolated peptides. Add the permits and field networks that take years to assemble, and the raw material becomes something even a well-funded drug company cannot simply buy.

That is the asset the round is really securing access to. The software is reproducible. The field expeditions, species identification, and venom sampling are not.

Venom is one of nature's oldest targeting systems. It gave us Captopril and GLP-1s, and until now we've only been able to search one molecule at a time. AI changes that.— Prateek Mehta, general partner, South Park Commons

What $2.7M buys, and what it does not

Venomics AI was conceptualised in January 2026 and founded in May by Kartik Sunagar, a professor at IISc's Centre for Ecological Sciences, and Vinay Prabhu, a Carnegie Mellon PhD who previously built AI systems at scale. Sunagar's record is unusually deep for a first-time founder: 97 publications, 6,211 citations, and a 39 h-index, with antivenom work that has reached the cover of Science Translational Medicine.

The company says it raised about $3 million in April 2026 to establish the platform, joined the NVIDIA Inception programme, and won a place among the top five finalists of the 2026 Indo-Boston Biotech Bridge Award. Its research partner, the Evolutionary Venomics Lab at IISc, received an Anthropic AI for Science grant in August.

The new capital is earmarked for the Venom Atlas, AI and laboratory screening capacity, and hiring across venom biology, computational science, and drug discovery. There is no mention of an IND filing, a lead candidate, or a clinical timeline.

The field is filling in around it. Bengaluru-based Peptris Technologies raised Rs 70 crore in a Series A co-led by IAN Alpha Fund and Speciale Invest in February 2026, using AI-driven molecular discovery plus experimental validation. Snaayu Lifesciences, incubated at C-CAMP, is characterising analgesic compounds from cone snail venom. Differentiate on data provenance, or not at all.

The bet is not that venom contains a drug. It is that the fastest way to find one is to stop guessing and start labelling. The $2.7 million is small enough to be honest about that — and large enough to buy the one thing that compounds: measurements nobody else has.