Artificial intelligence (AI) is becoming increasingly common in drug discovery. But in 2026, the more interesting question may no longer be whether a biotech company uses AI.
What matters is where AI is being applied—and what kinds of biological problems it is making possible to tackle.
A new generation of startups is moving beyond familiar applications such as virtual screening and molecular generation. They are using AI to explore previously underexplored biology, build computational models of cellular states, characterize complex protein therapeutics, reconstruct patient data that was never measured, and even predict how humans may respond to drugs.
Y Combinator's (YC) Drug Discovery and Delivery portfolio offers an interesting lens through which to examine this shift.
For this article, I selected five companies from YC's 2026 batches that represent different ways AI is being applied across drug discovery and development. This is an editorial selection, not a ranking of the companies.
Together, these five startups illustrate how the role of AI in drug development is expanding.
No. 01Ditto Biosciences
Discovering autoimmune therapies from parasite biology
dittobio.com →Batch — Winter 2026
Focus — Autoimmune disease, protein therapeutics, evolutionary biology
Sources — Official website / YC Profile / Publications
Many medically important drugs have their origins in nature. Ditto Biosciences is looking at a particularly unusual—and still relatively underexplored—source.
Parasites.
Viruses, ticks, worms, and other parasites have spent millions of years evolving ways to survive by manipulating their hosts' immune systems. In doing so, they have evolved proteins capable of suppressing or redirecting immune responses.
Those biological mechanisms could potentially be useful in treating diseases in which the immune system becomes overactive.
Ditto uses AI to explore this vast protein space created by evolution and identify candidates with potential applications in autoimmune disease.
The company combines computational and experimental approaches in a design–build–test–learn cycle. Its founding team brings expertise spanning host–pathogen interactions, evolutionary biology, computational biology, AI, and molecular biology.
According to Ditto, the company has already identified thousands of parasite-derived proteins with potential therapeutic applications. The diseases it is exploring include rheumatoid arthritis, inflammatory bowel disease, psoriasis, systemic lupus erythematosus, multiple sclerosis, type 1 diabetes, and other autoimmune and inflammatory conditions.
Ditto represents a somewhat different direction for AI-enabled drug discovery. Rather than using AI primarily to generate new molecules, the company is using it to search a vast biological space that already exists in nature.
Parasites can, in a sense, be viewed as having conducted millions of years of evolutionary experiments against the immune system. Ditto is betting that AI can help identify which of those evolutionary solutions might be translated into therapeutics—an example of AI used not only for molecular design, but to systematically explore therapeutic biology that has historically been difficult to search at scale.
CellType
Building a “biological world model” for drug development
celltype.com →Batch — Winter 2026
Focus — Biological foundation models, single-cell biology, translational research
Sources — Official website / YC Profile / Cell2Sentence (ICML 2024) / C2S-Scale
What if an AI model could learn a representation of human biology itself—and help predict which experiments, drug responses, or preclinical signals are most likely to matter in patients?
That is the direction CellType is pursuing.
The company describes its platform as a “biological world model for drug development.” Its foundation models learn from cells, perturbations, molecules, and patients, with the goal of capturing biological state within a shared computational representation.
One of the scientific foundations behind this work is Cell2Sentence (C2S), presented at ICML in 2024. The approach transforms single-cell gene-expression profiles into sentence-like sequences that can be processed by large language models.
The approach has since been scaled substantially. C2S-Scale uses a 27-billion-parameter model trained on more than one billion biological tokens.
Importantly, the work has gone beyond computational benchmarking. In research involving scientists from Yale and Google, C2S-Scale generated a new hypothesis about cancer cell behavior. The prediction was subsequently tested experimentally using human cell models that were not used to train the model.
CellType is now applying this underlying technology to areas including translational biology, patient signals, and virtual experiments. Rather than simply analyzing completed experiments, the company aims to use its models to help determine which experiment should be run next.
Much of AI-enabled drug discovery has focused on predicting molecular properties or generating new compounds. CellType is aiming at something more abstract: learning computational representations of biological state itself.
If such models can generalize across cells, perturbations, molecules, and patients, they could help address one of the central challenges in drug development—whether promising preclinical biological signals will translate into humans. The experimental validation in cancer research is particularly interesting: rather than simply reproducing known biology, the model generated a testable hypothesis that led to a new experimental observation.
10x Science
Making protein characterization AI-native
10xscience.com →Batch — Winter 2026
Focus — Protein characterization, mass spectrometry, AI-native proteomics
Sources — Official website / YC Profile
As drug discovery becomes capable of producing increasingly complex protein therapeutics, finding a promising molecule is only part of the challenge. Researchers also need to understand what they actually made.
10x Science is developing an AI-native molecular characterization platform, with a particular focus on complex protein therapeutics. Its software analyzes mass-spectrometry data across top-down, middle-down, and targeted workflows, and is designed to support proteoform identification, characterization of post-translational modifications (PTMs) such as glycosylation, de novo sequencing, unknown-modification searching, and peptide mapping.
This matters because biologic drugs are not defined by amino acid sequence alone. Post-translational modifications, unexpected sequence variants, and different proteoforms can influence the properties and performance of therapeutic proteins.
10x Science is applying AI and machine learning to automate and accelerate these analytical workflows, for protein-based therapeutics including antibodies, enzymes, bispecific molecules, and peptides.
As AI-enabled protein design continues to advance, characterization of candidate molecules remains critical—if design becomes faster, bottlenecks may simply emerge elsewhere, since the molecules produced still need to be experimentally characterized.
10x Science represents another side of AI-enabled drug discovery: rather than using AI to propose the next drug candidate, it applies AI to the analytical infrastructure researchers use to understand what they have actually produced.
Strand AI
Predicting the patient data that was never measured
strandai.com →Batch — Winter 2026
Focus — Multimodal AI, patient biology, biomarker discovery
Sources — Official website / YC Profile / Lattice
Modern drug development can generate remarkably diverse types of patient data: pathology images, genomics, transcriptomics, proteomics, and clinical data. In practice, however, few patient cohorts contain every modality for every patient.
Some assays are too expensive to perform at scale. Historical cohorts may contain only a subset of the measurements that could be collected today. Patients may drop out of studies or miss assessments. And in rare diseases, there may simply not be enough patients to build comprehensive multimodal datasets.
Strand AI is developing models designed to fill these gaps. Its platform aims to predict biological modalities that were never measured, using patient data that already exists.
Imagine, for example, that a pharmaceutical company has H&E images and genotype data for a patient cohort but lacks transcriptomic or proteomic measurements. Rather than running new assays across every sample, Strand aims to computationally predict the missing modalities—rescuing incomplete cohorts, expanding sparse rare-disease datasets, reducing dependence on expensive assays, and identifying biomarkers that were not originally measured. The company has also introduced Lattice, a model for predicting spatial proteomics.
AI in drug development is no longer only about molecular design. Strand is targeting another fundamental problem: the biological data researchers wish they had measured—but did not.
Strand illustrates an emerging use of multimodal AI—not simply analyzing the data that exists, but computationally reconstructing parts of the biological information that are missing. That could be particularly valuable in translational research, biomarker discovery, patient stratification, and rare diseases, where samples and patient populations are inherently limited.
Atlas Discovery
Predicting human drug response
atlasdiscovery.bio →Batch — Summer 2026
Focus — Human drug response, translational biology, clinical prediction
Sources — Official website / YC Profile / Research Preprint
One of the biggest problems in drug discovery appears after a promising molecule has already been found. A drug can perform well in cell models or animals and still fail when tested in humans.
Atlas Discovery is approaching this translational gap by developing models designed to predict how humans will respond to drugs. The company starts from the idea that much of the information needed to make these predictions may already exist—the problem is that biological and clinical data often sit in disconnected datasets that do not directly link a patient's biological state to drug response.
Atlas aims to computationally integrate these signals, rather than relying solely on animal models or isolated cell systems to infer what might happen in humans. According to the company, its research has been presented in connection with ICLR 2026, Cold Spring Harbor Laboratory, and ICML 2026.
The central translational problem in drug discovery remains: will a drug that looks promising preclinically actually work in patients? Atlas is targeting this problem directly—part of a broader shift toward applying AI to difficult downstream questions, not only the steps leading up to candidate selection.
If patient biology and drug-response data can be linked with sufficient reliability, such approaches could inform patient stratification, drug repurposing, clinical-development decisions, and the reevaluation of previously unsuccessful therapeutic programs—moving beyond predictions about molecules toward the question that ultimately matters most: what happens in humans?
What These Five Startups Tell Us About Drug Discovery in 2026
At first glance, all five companies could be grouped under the broad label of “AI drug discovery startups.” But the problems they are trying to solve are remarkably different.
AI is moving both deeper into biology and more broadly across the drug-development process—and that changes the question we should ask when evaluating an AI biotech company.
Increasingly, it is not “Does this company use AI?”
but rather: “What unique biological problem, dataset, experimental system, or scientific workflow does its AI make possible?”
Field Note
Another Perspective from the Field
In genomics and disease research, there has also been a broader shift from analyzing genomic data alone toward integrating multiple omics layers—including transcriptomics and proteomics—with clinical data and real-world data (RWD). AI and machine learning are increasingly being used to identify complex patterns across these heterogeneous data types.
Among the five companies discussed here, CellType, Strand AI, and Atlas Discovery in particular reflect this broader direction: connecting different types of biological and clinical information through computational models and using them to support decisions in drug development.
At the same time, many of the companies in this space are likely to depend on pharmaceutical and biotech companies as important customers, partners, or routes to commercialization. That raises another important question beyond the technology itself: how will these technologies actually be commercialized, who will recognize their value, and who will pay for it?
As AI becomes more deeply embedded as a computational layer within drug discovery, technical performance alone may not be enough. Commercialization and partnership design—connecting a scientific capability to real R&D workflows, business models, and decision-makers—may become increasingly important.
This is where I believe expertise in business development and partnership building can play a meaningful role. Understanding the technology makes it possible to ask a broader question than simply, “Which pharmaceutical company could buy this?” Depending on the underlying capability, there may be opportunities across diagnostics, healthcare platforms, research infrastructure, or entirely different partnership models.
This perspective also connects closely to my own experience working at the intersection of science and business. My scientific training taught me to continually break down a problem and ask: “Are we actually asking the right question?” Research is not simply about finding answers—it also involves challenging assumptions, identifying the underlying problem, and redesigning the question from a different angle.
After working in startups and business, I realized that this way of thinking applies far beyond research. Understanding the science makes it possible to challenge existing assumptions about commercialization and start again with a more fundamental question.
“What problem can this technology actually solve—and for whom?”
I believe that asking and answering that question is one of the most valuable roles for people working at the intersection of science and business.
Working on Japan strategy for a science-led company?
I help international biotech and healthtech companies understand where their science fits, who they should talk to, and how to build the right relationships—in Japan and beyond.
Discuss Your Japan Entry