Mandrake Bio has raised Rs 16 crore, or Rs 160 million, to develop a generative artificial intelligence platform designed to create new gene-editing enzymes for agricultural and medical applications. The Bengaluru-based company is seeking to move AI-enabled biological research beyond prediction and target discovery by designing the molecular tools that directly modify DNA.

The financing was co-led by Activate, the early-stage investment firm founded by technology entrepreneur Aakrit Vaish, and Antler India. Spectrum Impact and venture investment firm DeVC also participated, alongside a group of individual investors that included biotechnology veteran Vijay Chandru, entrepreneur Paras Chopra, Sanjiv Rangrass and Vatsal Dusad.

Mandrake said the new capital would be used to increase the size of its artificial intelligence and biophysics teams and accelerate laboratory validation of enzyme candidates generated by its computational platform. Financial terms beyond the total amount raised were not disclosed, including the company’s valuation, ownership dilution or the formal stage assigned to the round.

The financing remains small compared with the sums typically required to bring agricultural biotechnology products or gene therapies to market. Its importance instead lies in the technical thesis being funded: that generative models trained on biological sequences can create gene-editing enzymes tailored to specific functions, rather than forcing researchers to adapt a limited set of tools discovered in nature.

Mandrake was founded in 2025 by chief executive Tanay Lohia and Kutubuddin Molla, a scientist associated with the Indian Council of Agricultural Research. The company has assembled a team of about 10 people spanning artificial intelligence research, computational biology and molecular science, according to information reported alongside the funding announcement.

The startup’s approach begins with protein language models, a class of AI systems that process amino-acid sequences using methods conceptually similar to those used by language models to identify patterns in text. Proteins are represented as sequences whose order influences their structure and function. By learning statistical relationships across large biological datasets, the models can propose sequences that may have desired characteristics.

Mandrake uses open-source protein language models and fine-tunes them with its own metagenomic database. Metagenomic data can contain genetic material collected from complex environmental samples, offering access to biological diversity that may not appear in conventional databases focused on well-studied organisms. The company’s system then generates or prioritizes potential enzymes for experimental evaluation.

That computational stage is intended to narrow a vast search space. Instead of synthesizing and testing thousands of proteins without strong prior selection, Mandrake’s models can rank candidates according to predicted editing characteristics. Researchers can then manufacture a smaller number of proteins and assess their performance in laboratory systems.

The process illustrates why wet-lab validation is the most consequential near-term milestone for the company. A model may generate a sequence that appears plausible based on training data, structural predictions or simulated interactions, but biological activity must still be demonstrated experimentally. Candidate editors must cut or modify DNA at the intended location, operate efficiently in relevant cells and avoid unacceptable activity elsewhere in the genome.

Mandrake expects to obtain its first laboratory validation results within approximately two months of the funding announcement. Those results could provide an initial indication of whether its generated enzymes are functional, but they would not by themselves establish commercial readiness. Replication, optimization, delivery testing and evaluation in increasingly realistic biological systems would follow.

The company is initially prioritizing agriculture. Gene editing can be used to develop crops with characteristics such as improved disease resistance, tolerance to drought or heat, better nutrient profiles and more efficient use of water or fertilizer. Unlike traditional breeding, targeted editing can make defined genetic changes, although development timelines remain long because promising edits must still be tested across generations, environments and regulatory settings.

Lohia has said existing workflows can require seven to eight years to produce an improved crop variety. Mandrake believes enzymes built for particular crops or genetic targets could potentially reduce that period to roughly two years. That estimate should be viewed as a development objective rather than an established outcome, since seed multiplication, field trials, regulatory review and commercialization would remain outside the enzyme-design process.

Even so, reducing the earliest discovery and optimization stages could carry economic value. Agricultural biotechnology companies frequently face constraints when a standard editing enzyme performs poorly in a plant species, cannot access the desired DNA sequence or creates unintended changes. A broader library of editors could allow researchers to select enzymes with properties suited to different crops, cell types and environmental conditions.

Mandrake’s proposition differs from companies that apply established CRISPR systems to individual products. The startup is attempting to work at the enabling-technology layer by redesigning the editor itself. If successful, it could supply or license molecular tools to crop developers, pharmaceutical companies and research institutions rather than bearing the full cost of developing every downstream product.

A platform-oriented model could also allow the company to operate across multiple industries. Gene editing is sometimes described as a horizontal technology because similar underlying tools can be used before agricultural and medical development paths diverge. Both a disease-resistant plant and a therapeutic intervention may begin with a need to make a precise change in DNA, even though their delivery systems, testing requirements and regulators are different.

The medical opportunity is potentially larger but also substantially more demanding. Gene-editing therapies must meet high standards for safety, manufacturing consistency and clinical efficacy. The enzyme must reach the appropriate cells, perform the intended edit at a sufficient rate and minimize off-target changes that could create harmful effects. Regulators may also require long-term patient monitoring.

Mandrake has argued that better-designed editors could eventually contribute to lower treatment costs. Existing one-time genetic medicines can carry prices measured in millions of dollars, reflecting years of research, clinical trials, specialized manufacturing, small patient populations and complex delivery procedures. Improving the editing enzyme may reduce some technical costs, but it would not remove the broader expenses associated with clinical development and regulated production.

The company’s stated ambition of making gene-editing therapies available at significantly lower prices therefore depends on factors beyond protein design. Delivery technologies, intellectual-property rights, hospital infrastructure, manufacturing scale and reimbursement systems can all have a larger influence on final treatment prices than the cost of discovering an enzyme.

The financing nevertheless reflects a widening investor interest in AI systems built for scientific and industrial discovery. Much of the first wave of generative AI investment focused on consumer applications, enterprise software and content generation. Capital is increasingly moving toward models designed for chemistry, biology, materials science and engineering, where successful outputs can become patentable physical products or research platforms.

Biology presents a more difficult validation environment than software. A software product can often be tested and revised rapidly after deployment. Biological candidates must be synthesized, handled under controlled conditions and evaluated through experiments that may take weeks or months. Failures can arise because models do not fully capture protein folding, cellular conditions, molecular interactions or evolutionary constraints.

That creates a feedback-loop advantage for companies that combine AI teams with laboratory operations. Each experiment can produce proprietary data showing which computational designs succeeded and which failed. Those results can be used to refine models, improve ranking systems and guide subsequent generations of candidates. Over time, the experimental dataset may become a more defensible asset than the underlying open-source model.

Mandrake’s use of an internal metagenomic database is therefore central to its competitive strategy. Open protein models are widely available, which lowers the barrier to entering computational biology but also makes model access alone difficult to defend. Proprietary sequence collections, carefully labeled experimental results and specialized laboratory protocols can provide differentiation that competitors cannot easily reproduce.

The company will still face competition from global biotechnology startups, academic laboratories and established life-sciences groups developing new nucleases, base editors, prime editors and other genome-engineering systems. Some competitors have larger datasets, experienced drug-development teams and partnerships with multinational pharmaceutical or agricultural companies.

Mandrake’s location could provide both advantages and constraints. India has a large pool of software engineers, life-sciences researchers and agricultural scientists, while research costs may be lower than in major U.S. biotechnology centers. The country also offers a large agricultural market with varied crops, climates and disease pressures that could support application-specific development.

At the same time, deep-biotechnology companies require specialized equipment, reliable research materials, experienced laboratory managers and access to international scientific networks. Scaling from computational design into advanced biological testing may require partnerships with universities, contract research organizations, seed companies and medical research institutions.

Regulation will also shape the commercial path. Agricultural rules differ by country and may depend on whether an edited crop contains foreign genetic material or could have been produced through conventional breeding. Medical products face a more standardized but lengthy sequence of preclinical studies, clinical trials and manufacturing reviews. A versatile editor may be scientifically useful across both sectors, but each downstream product must satisfy its own approval framework.

Intellectual property is another material consideration. The gene-editing industry has been shaped by extensive patent disputes covering CRISPR systems, delivery methods and therapeutic applications. Mandrake will need to establish freedom to operate for any commercial editor and determine whether AI-generated sequences can be protected through patents, trade secrets or licensing arrangements.

Safety governance may become increasingly important as models make biological design easier. Systems capable of generating functional enzymes can support beneficial research, but they also require controls around data access, candidate screening and laboratory use. Investors and commercial partners are likely to examine whether Mandrake has procedures for biosafety, cybersecurity and review of potentially sensitive designs.

For the current funding round, the most important measure of progress will be scientific rather than financial. Investors will be watching whether the company can produce enzymes that work as predicted, whether performance can be reproduced and whether generated candidates demonstrate advantages over existing editors on clearly defined targets.

Subsequent milestones could include validation in plant cells, partnerships with crop-science companies, expansion of the proprietary dataset and evidence that the platform can generate distinct classes of editors. Medical development would likely require additional capital and collaborations with organizations experienced in delivery, toxicology, clinical trials and regulated manufacturing.

The Rs 16 crore investment gives Mandrake enough resources to test the foundation of its thesis but not to eliminate the considerable technical and commercial risks. The company must show that generative AI can do more than propose biologically plausible proteins. It must consistently produce gene-editing tools that scientists can manufacture, control and use more effectively than existing alternatives.

If those results emerge, Mandrake could become part of a new class of Indian companies building AI systems for physical science rather than conventional software services. If validation falls short, the round will still illustrate the limits of applying language-model techniques to complex biological functions. The funding is therefore best viewed as an investment in experimental proof: converting computational designs into working molecular machinery.