Biotech
5 Early-Stage AI Biotech Startups Rethinking Drug Discovery
DealBio tracks early-stage AI biotech startups across Europe and North America. Here are 5 applying AI to different layers of biology, from microbial communities and RNA to therapeutic DNA, molecular interactions and protein design.
- AI Health
- Drug Discovery
- Protein Design
- Precision Medicine
AI biotech is becoming more specialized. The early wave of the sector often grouped very different platforms under the same drug-discovery label, but the underlying models are increasingly being built around specific biological data and specific R&D decisions.
That specialization matters commercially. A microbiome model, an RNA foundation model and a protein-design system solve different problems, require different validation and fit into different pharmaceutical workflows.
The five startups below show AI moving deeper into biology and into specific R&D workflows. Their value will depend on whether model performance can translate into better experiments, better molecules or better clinical-development decisions.
At a glance
- Anto BiosciencesUnited States2025Microbial communities
- Blank BioUnited States2025RNA and tumor transcriptomes
- Houdini BioUnited Kingdom2025Therapeutic DNA design
- Lemna BioSwitzerland2025Molecular interactions
- ProtoVelaGermany2025Protein developability
Building foundation models for microbial communities and microbiome drug discovery.
- Country
United States- Founded
- 2025
- Sector
- Biotech
- Funding stage
- Seed or earlier
Why Anto Biosciences stands out
Anto Biosciences is applying foundation models to microbial communities, a data layer that is still largely separate from conventional drug-development models. The founders combine research at MIT, the Broad Institute and Harvard Medical School with prior startup experience.
The startup joined Y Combinator’s Fall 2025 batch. Co-founder David de Gruijl also co-authored peer-reviewed work on microbiome-aware drug development, giving the product thesis a direct scientific foundation.
What Anto Biosciences is building
Anto Biosciences is building a prediction platform for pharmaceutical R&D rather than its own therapeutic. The models are designed to estimate how microbiome variation affects drug efficacy and toxicity across patient populations.
The initial focus is oral medicines, where gut bacteria can interact with a compound before absorption. That gives the platform a defined entry point into drug design and development rather than a broad microbiome analytics proposition.
Training RNA foundation models to guide oncology trial patient selection.
- Country
- United States
- Founded
- 2025
- Sector
- Biotech
- Funding stage
- Seed or earlier
- Founders
Why Blank Bio stands out
Blank Bio raised a $7.2 million oversubscribed seed round in May 2026 and partnered with PacBio to generate long-read RNA-sequencing data from up to 100 patient tumor samples across multiple cancer indications.
Its underlying Orthrus RNA foundation model was published in Nature Methods in 2026. The startup also joined Y Combinator’s Summer 2025 batch, while the founding team brings experience across RNA biology, machine learning and oncology.
What Blank Bio is building
Blank Bio is training RNA foundation models to extract patient-level information from tumor transcriptomes. Its two initial applications are predicting disease trajectories and identifying patients most likely to respond to a therapy.
That positions the technology around expensive clinical-development decisions. Its value lies in using RNA data to improve trial design, patient selection and biomarker strategy.
Using machine learning to redesign therapeutic DNA and reduce gene silencing.
- Country
United Kingdom- Founded
- 2025
- Sector
- Biotech
- Funding stage
- Seed or earlier
- Founders
Why Houdini Bio stands out
Houdini Bio raised £1.5 million in a pre-seed round led by SCVC / Science Creates in June 2026. It has achieved four-to-five-times higher protein output per dose after redesigning a haemophilia A gene-therapy construct in liver cells.
CEO Jonathan Cohen-Gold identified the DNA design rules behind the platform during his Cambridge PhD. Co-founder Lee Dunham spent seven years at the Cell and Gene Therapy Catapult, and the team later completed the KQ Labs programme at the Francis Crick Institute.
What Houdini Bio is building
Houdini Bio’s ARCANA platform uses machine learning to redesign therapeutic DNA so cells are less likely to silence it while preserving the encoded protein. The same principle is intended for gene therapy, cell therapy and biomanufacturing.
This is a narrow but important AI biotech problem: improving expression without changing the therapeutic payload. If the platform can generalize across constructs, it could affect dose, durability and manufacturing economics.
Building structural AI models to decode molecular interactions for drug discovery.
- Country
Switzerland- Founded
- 2025
- Sector
- Biotech
- Funding stage
- Seed or earlier
Why Lemna Bio stands out
Lemna Bio already reports signed letters of intent and active biopharma contract discussions, giving its research platform an early commercial test. It has also secured CHF 400,000 in non-dilutive support.
The science is unusually visible for an early startup. PeSTo and CARBonAra were both published in Nature Communications, and CEO Moustafa Houmani previously co-founded a Y Combinator-backed protein-engineering startup. Scientific co-founder Matteo Dal Peraro leads EPFL’s Laboratory for Biomolecular Modeling.
What Lemna Bio is building
Lemna Bio develops physics-grounded AI models that work directly from molecular structure. PeSTo predicts interaction interfaces, while CARBonAra designs protein sequences in the context of a three-dimensional backbone and surrounding molecules.
The commercial strategy spans fee-for-service work, early biopharma partnerships and eventual platform access. That gives Lemna Bio several routes to monetize before attempting to build a fully integrated drug pipeline.
Designing AI-optimized protein variants with improved stability and expression.
- Country
Germany- Founded
- 2025
- Sector
- Biotech
- Funding stage
- Seed or earlier
- Founders
Why ProtoVela stands out
ProtoVela focuses AI protein design on developability rather than de novo drug discovery. Founder Kornelius Zeth brings more than 30 years of protein-structure experience, and the startup has published prior industrial design results in which all eight selected enzyme variants expressed successfully.
ProtoVela was selected into BioPark Regensburg’s BioPark Jump accelerator. Its operating model is deliberately asset-light: computational design is done in-house, while clients or laboratory partners carry out experimental validation.
What ProtoVela is building
ProtoVela redesigns client proteins using structural biology, physics-based modelling, bioinformatics and machine learning. The goal is to improve properties such as stability, expression, solubility and manufacturability before larger experimental programs begin.
ProtoVela operates as an R&D service layer for biotech and pharmaceutical teams. The platform is designed to narrow experimental search space and prioritize candidates for validation, making repeat client use the key commercial test.
Deal Sourcing for Investors
Looking for More Startups Like These?
DealBio helps investment teams discover, research, and monitor early-stage healthcare and life sciences startups aligned with their investment criteria.
AI biotech is becoming more biology-specific
These five startups are not competing to build the same AI drug-discovery platform. Anto Biosciences models microbial communities, Blank Bio works from RNA, Houdini Bio redesigns therapeutic DNA, Lemna Bio reasons over molecular structure, and ProtoVela focuses on protein developability.
The common pattern is specialization. AI is becoming a layer inside particular biological workflows, with the model architecture, training data and validation strategy shaped by the scientific problem rather than by a generic software category.
For investors, that shifts attention toward proprietary or defensible data, experimental validation and customer integration. The strongest AI biotech startups will be those that can show not only that a model predicts well, but that its predictions change R&D decisions and improve downstream outcomes.
Methodology
DealBio selected these startups from its coverage of early-stage health and longevity startups, focusing on those where artificial intelligence is central to a biotech R&D workflow rather than a general software feature. The selection spans microbiome modelling, RNA foundation models, therapeutic DNA design, structural biology and protein engineering.
DealBio builds its Startup Briefs using publicly available startup materials, ecosystem sources, market references and professional profiles.