New AI-Native Biotechs: 2026 Watchlist
Just a few days ago, the FDA formally established a one-pivotal-trial default for drug approval, explicitly emphasizing mechanistic, real-world, and model-based “confirmative evidence” around that study.
Under the updated framework, sponsors are expected to supplement a single pivotal study with mechanistic data, validated biomarkers, external controls, real-world evidence, and other forms of “confirmative evidence.” In practice, this elevates the importance of computational modeling, synthetic comparators, in-silico simulations, and high-dimensional data analysis—areas where AI methods are already deeply embedded.
At the same time, a new generation of biotech startups is being built around exactly those capabilities. For example, Yale spinout CellType, which just joined Y Combinator, is developing large-scale foundation models of human biology coupled with task-specific AI agents to simulate drug response before first-in-human studies, and its team reported that at least one cancer-related prediction from its virtual screening framework was subsequently identified and validated in living cells. Its premise is that more faithful in-silico modeling of patient biology can reduce attrition.
Below is a look at several newly launched companies embedding AI across RNA design, biomolecular modeling, regulatory DNA engineering, data imputation, lab automation, and structure-based drug discovery.
Blank Bio
RNA foundation models for therapeutics and trials
Blank Bio is a YC-backed techbio company founded in 2025 by Philip Fradkin, Jonny Hsu, and Ian Shi, and is based in San Francisco, California. The company entered Y Combinator’s Summer 2025 batch and positions itself as building next-generation RNA foundation models to support RNA-based therapeutics, biomarkers, and diagnostics across the R&D pipeline.
The platform is centered on RNA-specific foundation models. These models analyze different versions of RNA molecules, their mutations, and how strongly they are expressed in cells to predict how stable they are and how much protein they might produce, effectively simulating many wet-lab assays in silico. Based on these predictions, the system can help optimize mRNA sequences for better performance, analyze cell-free RNA in blood samples, identify biomarkers at the level of specific RNA variants, and discover new RNA-based drug targets. It can also support the design of RNA-targeting therapies. The company reports that these models may also help group patients more precisely and design more efficient clinical trials.
Boltz PBC
Open biomolecular foundation models for drug discovery
Boltz PBC is a public-benefit AI research company focused on biomolecular foundation models for drug discovery. In early 2026 the company publicly launched with a reported $28 million seed round led by Andreessen Horowitz and other investors, described as an MIT CSAIL spin-out built around the “Boltz” model family. Boltz is led by CEO Gabriele Corso, with co-founders Jeremy Wohlwend and Saro Passaro, all with technical roots in machine learning work at MIT.
The platform takes a set of large AI models that “understand” proteins and other biomolecules in 3D and turns them into tools that working scientists can use for design and prediction. The open Boltz-1 model (released before the company launch) was introduced as achieving AlphaFold-3-level accuracy on protein complex structure prediction, followed by Boltz-2 for joint structure and binding-affinity prediction and BoltzGen for generative protein design. These models are integrated into Boltz Lab, a hosted environment with small-molecule and protein-design “agents” that automate workflows like proposing binders, ranking them by predicted affinity, and filtering for synthesizability. Boltz’s collaboration with Pfizer involves further training of these foundation models on Pfizer’s internal data to create exclusive versions for structure prediction, small-molecule affinity estimation, and biologics design, while the general-purpose models remain available to a wider community of users.
Origin
Regulatory DNA design for safer cell and gene therapies
Origin is a San Francisco-based startup founded in 2025 by Yash Rathod (CEO) and Malhar Bhide (CTO), and is part of Y Combinator’s Winter 2026 batch. The company focuses on making cell and gene therapies safer by designing regulatory DNA “switches” (promoters and enhancers) that control where and how strongly therapeutic genes are turned on in the body. In October 2025, Origin publicly introduced Axis, an AI model trained on large regulatory DNA datasets from ENCODE that can both predict how existing regulatory sequences behave and generate new ones.
Origin uses AI as a designer and tester of gene-control switches. Axis reads regulatory DNA and learns how transcription factors bind and affect gene activity, then generates millions of candidate sequences tailored to specific disease-relevant cell types and ranks them by predicted strength and specificity. The company reports that Axis outperforms DeepMind’s AlphaGenome on benchmarks of regulatory element activity prediction and that “high-affinity” prompts can enrich desired transcription factor motifs by up to about ninefold, which they position as a way to tune gene therapies toward target cells while reducing unwanted activity elsewhere.
Manas AI
AI-native end-to-end drug development for oncology and rare diseases
Manas AI is a full-stack, AI-native biopharmaceutical company headquartered in New York City, founded in 2025 by oncologist and author Siddhartha Mukherjee and LinkedIn cofounder Reid Hoffman; later, former Google engineering leader Ujjwal Singh joined as co-founder and CTO. The company publicly launched with a $24.6M seed round co-led by General Catalyst and Hoffman, and later announced a $26M dollar seed extension. Manas AI positions itself as developing medicines for cancer and rare diseases, with early public statements highlighting initial programs in breast, prostate, and lymphoma cancers.
The platform is designed to support drug development from early target discovery through candidate design, laboratory testing, and clinical studies. It combines AI systems with input from biologists, chemists, and clinicians to guide decisions at each stage. It can be applied to several types of therapies, including antibodies, small molecules, RNA-based drugs, and combination treatments, and works with external partners to help run and manage clinical trials. Manas describes its core models as neuro-symbolic, science-based foundational models and runs them on Microsoft’s Azure cloud infrastructure, with the stated aim of shortening discovery timelines and lowering the cost of bringing new oncology and rare-disease drugs into development.
Strand AI
Foundation models to impute missing omics and clinical data
Strand AI is a San Francisco-based startup founded in 2025 by Yue Dai (CEO) and Oded Falik (CTO). It is part of Y Combinator’s Winter 2026 batch, with a focus on life-science data infrastructure. The company describes itself as building “multimodal patient datasets for life sciences,” with an emphasis on drug discovery and clinical development use cases. Strand also provides an open-access “1000 Genomes VariantFormer Predictions Explorer,” an interactive tool that shows AI-predicted RNA expression from DNA variants across tissues and populations and allows users to browse and download the underlying dataset.
Strand AI’s platform uses foundation models to “fill in” missing biological measurements in human studies. It can reconstruct omitted or unmeasured modalities in clinical trial datasets so patients with incomplete data are still usable, predict expensive readouts like proteomic or transcriptomic profiles from cheaper routine data, and impute unmeasured molecular markers across an entire cohort to speed up biomarker discovery without repeating wet-lab assays. For rare disease studies with few patients and missing data, you can synthesize the missing measurements so it becomes possible to train models that otherwise couldn’t be built.
Medra AI
Autonomous AI-robotic lab automation for drug discovery
Medra is a San Francisco-based startup building what it calls “Physical AI Scientists” for life-science labs. It was founded by Michelle Lee, a Stanford-trained roboticist and former NYU professor, who began working on the company in 2021. Medra came out of stealth in September 2025 with the launch of its Continuous Science Platform, then raised a $52M Series A in December 2025 led by Human Capital with participation from Lux Capital, Neo, NFDG, Catalio, Menlo Ventures, 776, Fusion Fund and others, bringing total funding to about 63M dollars. Medra is trying to close the loop between in silico drug discovery models and wet-lab validation by giving AI systems the ability to physically run and iterate on experiments in an always-on, autonomous lab.
Medra’s platform joins two main components: Physical AI, a fleet of general-purpose lab robots, and Scientific AI, a reasoning layer built on large language and vision models. Physical AI systems can plug into standard instruments, read screens, press buttons and handle labware, and have been described as able to automate roughly 70% of equipment in a typical lab. Scientific AI proposes hypotheses, designs protocols, and interprets results; the robots then run those experiments end-to-end and feed detailed measurements (down to pipette angles and timing) back into the models to improve the next round of experiments.
Resonate Bio
AI-powered NMR platform for structure-based drug design
Resonate Bio is a Vienna-based biotech spin-out from the University of Vienna, founded in 2025 by Darryl B. McConnell, Gerald Platzer, and Robert Konrat. The company was formally launched in July 2025 as a university spin-out licensing its AI-NMR software and later received an Austria Wirtschaftsservice DeepTech pre-seed grant to advance the platform. Its origins lie in years of work in a Christian Doppler Laboratory jointly run by the University of Vienna and Boehringer Ingelheim, where the underlying AI-NMR technology was developed.
Resonate Bio’s core platform combines nuclear magnetic resonance (NMR) experiments in solution with machine-learning models to generate atomic-level “movies” of drug molecules binding to highly flexible, disease-relevant proteins. Instead of a single static structure, the system produces ensembles of protein–ligand conformations and their relative populations, which the company calls “Resonate Structures”. These ensembles are intended to support structure-based drug design against dynamic targets that are often considered “undruggable” using traditional crystallography or cryo-EM alone. In practical terms, AI models learn from NMR data to infer likely 3D binding modes for many compounds in a high-throughput way, giving medicinal chemists richer structural input when designing and optimizing small-molecule drugs.
Topic: Biotech Ventures
