insitro has announced the acquisition of CombinAbleAI and launched its new TherML platform, a modality-agnostic, AI-powered system that integrates small molecules, oligonucleotides, and biologics into a unified therapeutic design stack. The move signals a strategic attempt to industrialize AI-native drug discovery across modalities while reducing late-stage attrition risks.

What makes this acquisition different from typical AI-platform bolt-ons in biotech?

At first glance, insitro’s acquisition of CombinAbleAI might appear like another incremental expansion of an AI biotech’s capabilities. However, the company is making a broader claim: that TherML, its newly announced therapeutic machine learning platform, can meaningfully unify the design space across all major drug modalities—including complex biologics—within a single AI framework. That vision has been promised before by many players in the space. Few have actually delivered a stack that integrates AI with wet lab automation across modalities while maintaining high-fidelity feedback loops.

The addition of CombinAbleAI gives insitro access to a physics-informed, AI-driven engine for protein design trained on over 100,000 molecular dynamics surrogates. This platform complements insitro’s existing work in small molecules and oligonucleotides, creating a full-stack system capable of designing multi-specific antibodies, T-cell engagers, siRNAs, and small molecule drugs with comparable precision and performance benchmarks.

Unlike other acquisitions that plug a new asset into a legacy workflow, the integration here appears to be foundational. CombinAbleAI’s models are not being tacked on to insitro’s platform—they are forming a critical pillar of the TherML architecture. This creates a structure where drug-likeness, manufacturability, and binding affinity are co-optimized from the outset, rather than tackled sequentially. In doing so, insitro is trying to solve one of drug discovery’s most persistent headaches: the misalignment between in silico hits and real-world manufacturability.

What this reveals about insitro’s strategic evolution in the AI drug discovery space

Founded by Daphne Koller, insitro has long positioned itself as a leader in applying causal biology and machine learning to uncover therapeutic targets. The company’s earlier focus leaned heavily on small molecules and siRNA design using high-resolution adaptive chemical libraries and automation-driven experimental loops. The acquisition of CombinAbleAI represents an intentional pivot to become truly modality-agnostic.

This move mirrors a growing consensus among industry strategists that single-modality AI platforms—no matter how sophisticated—will eventually hit ceiling effects in both deal value and pipeline optionality. The future belongs to platforms that can move fluidly between modalities depending on biological context, disease tractability, and commercial potential. By launching TherML with integrated capabilities across oligonucleotides, biologics, and small molecules, insitro is placing itself in that category of next-gen discovery platforms that can credibly claim end-to-end therapeutic design agility.

The acquisition also formalizes insitro’s move toward becoming a full-stack AI therapeutics company with deeper control over experimental data loops. The company already runs automated labs that feed back into its AI models, creating a continuous learning cycle. Now, with CombinAbleAI’s biologics engine, those loops extend into multi-specific antibodies and T-cell engagers—modalities with high clinical relevance but steep manufacturability hurdles.

How TherML tries to rewire the therapeutic design workflow

What’s genuinely new here is not just the coverage of multiple modalities but the restructuring of the optimization process itself. Traditional workflows often separate the goals of biological activity (e.g., potency, selectivity) from developability factors (e.g., stability, manufacturability, immunogenicity). Molecules are optimized for one, then retrofitted for the other, often resulting in clinical setbacks and costly redesigns.

TherML aims to collapse that dichotomy. Its AI models co-optimize for binding affinity, selectivity, stability, and safety profile in tandem with manufacturing constraints—using surrogate models, in silico simulation, and real-time lab validation to navigate the trade-offs. For biologics, the CombinAbleAI engine brings predictive capabilities for protein structure, folding, and flexibility. For oligonucleotides, the platform uses automation to industrialize siRNA design. For small molecules, dense local training data from adaptive chemical libraries feeds predictive ADMET models.

The platform’s ability to industrialize and loop data from real-world experiments back into its design models adds to its credibility. Few companies can execute on both computational and wet-lab fronts at this scale.

Where this leaves other AI-native players focused on single modalities

insitro’s TherML rollout implicitly raises the bar for AI-native biotech firms that are still focused on single-modality playbooks. Companies specializing in small molecule generation via generative models, or in RNA therapeutics using sequence optimization, may find themselves under pressure to expand either their modality coverage or deepen their design-to-validation loops.

In contrast to players like Recursion Pharmaceuticals or Exscientia, whose focus has largely remained on small molecules and phenotypic screening, insitro is now claiming the higher ground of modality-independence. Meanwhile, large pharma incumbents are beginning to internalize AI-native capabilities rather than rely solely on platform deals. This makes full-stack, multi-modality AI players more attractive as long-term partners—or acquisition targets.

By integrating early biological insight with downstream therapeutic constraints into a single design loop, insitro is positioning itself not as a toolmaker, but as a vertically integrated engine for drug creation. That distinction could determine its strategic value as the field consolidates.

Regulatory and commercial challenges still loom over AI-generated biologics

Despite the technical sophistication of TherML, insitro’s ambitions must still contend with the realities of regulatory oversight and clinical validation. Designing AI-native biologics does not eliminate the need for robust preclinical and clinical trials. Regulatory bodies such as the U.S. Food and Drug Administration and the European Medicines Agency continue to evaluate AI-generated candidates with traditional lenses—focusing on pharmacokinetics, toxicity, immunogenicity, and clinical efficacy.

Moreover, while insitro’s internal programs have largely focused on metabolic and neurodegenerative diseases, biologics often dominate in oncology and immunology—fields with complex trial designs, high comparator efficacy, and evolving biomarker requirements. If insitro aims to compete in these spaces, it will need to prove that TherML-designed biologics can outperform both legacy antibodies and next-generation cell-based therapies on clinically meaningful endpoints.

Adoption risks also remain. Clinicians and payers are still wary of unproven AI-derived drugs, especially in therapeutic areas with established standard-of-care treatments. Commercialization, manufacturing scale-up, and payer negotiation remain hurdles that even the most elegantly designed molecule cannot circumvent.

Why CombinAbleAI’s origin in AION Labs adds institutional weight

CombinAbleAI’s incubation within AION Labs adds institutional legitimacy to the acquisition. AION is not a typical accelerator. It was founded by a coalition of global pharma and tech companies—Pfizer, Merck KGaA, AstraZeneca, Teva, Amazon Web Services—along with the Israel Biotech Fund and the Israel Innovation Authority. The venture studio’s model emphasizes early technical validation, industry-relevant problem statements, and streamlined commercialization roadmaps.

That backing differentiates CombinAbleAI from many academic spinouts or early-stage AI-biotech ventures with limited validation. It also suggests that insitro acquired more than just a promising algorithm. The CombinAbleAI team brings technical depth, enterprise readiness, and credibility within pharma partnership networks, making it more likely that the TherML platform will attract licensing or co-development interest from major drug developers.

What comes next: Metrics to watch as TherML scales

In the short term, stakeholders will be watching for signs of throughput and pipeline maturity. How many clinic-ready assets does TherML generate in the next 12 to 24 months? How fast can it move from target nomination to IND-enabling studies? And does it enable first-in-class programs where biology and modality alignment was previously a barrier?

TherML’s integration with insitro’s automated labs creates the potential for real-time model refinement, but it also demands capital-intensive infrastructure. With over $800 million in funding, insitro has the runway to absorb early development costs, but investors will expect progress on lead optimization timelines, early clinical validations, and partnership revenue.

Ultimately, TherML will be judged on whether it can materially reduce time to clinic and de-risk development for complex modalities. If it can consistently turn high-confidence biological hypotheses into scalable, manufacturable therapeutics, insitro will have earned its place among the few AI-native drug creators rewriting the rules of modality design.

  AI drug discovery, AION Labs, antibody optimization, biologics, CombinAbleAI, insitro, siRNA therapeutics, small molecule design, TherML