insitro expands AI drug discovery with CombinAbleAI acquisition

insitro, the AI therapeutics company built on causal biology, has announced the acquisition of CombinAbleAI alongside the launch of its new TherML™ (Therapeutic Machine Learning) platform. The deal, expected to close in late January 2026, marks a significant expansion of insitro’s end-to-end drug discovery capabilities.

The acquisition is designed to complete insitro’s full-stack, modality-agnostic platform, spanning small molecules, oligonucleotides, antibodies and other complex biologics. By integrating biology-driven target discovery with molecular design and developability assessment, the company aims to generate clinic-ready assets more rapidly while reducing the risk of late-stage failure.

Designing potency and manufacturability together

Traditionally, drug discovery has prioritised molecular potency before addressing whether a candidate can be manufactured and developed at scale. insitro believes this sequential approach is a major contributor to attrition.

“Drug discovery has traditionally optimised molecules for potency before assessing developability – often discovering that highly potent candidates face manufacturing constraints,” said Philip Tagari, chief scientific officer, insitro. “By integrating CombinAbleAI’s physics-informed, AI-driven design for complex biologic therapeutics with our causal biology platform, we treat potency and manufacturability as interdependent design criteria from the outset. This fundamentally changes the translation from biological insight to viable therapeutic – instead of optimising sequentially and hoping for serendipitous alignment, we’re designing for both simultaneously.”

Launching the TherML platform

The acquisition supports the launch of TherML, insitro’s integrated platform for AI-driven therapeutic design across multiple modalities. TherML brings together predictive machine learning, large-scale automation and experimental feedback loops to optimise efficacy and safety together.

“We are delighted to welcome the CombinAbleAI team in Israel as fellow insitrocytes, alongside our colleagues in the US, Poland and Malaysia,” said Dr Daphne Koller, founder and CEO of insitro. “Their addition helps launch our integrated TherML platform, which is a critical part of our end-to-end industrialised AI-enabled system built for repeatable, scalable, predictable drug discovery.”

A modality-agnostic AI engine

For complex biologics such as multi-specific antibodies and T-cell engagers, TherML incorporates CombinAbleAI’s physics-informed optimisation engine, pre-trained on more than 100,000 molecular dynamics surrogates to accurately predict protein structure and flexibility. This allows for the design of large molecules that balance binding performance with stability and manufacturability.

In small-molecule discovery, the platform leverages insitro’s proprietary Quantitative Adaptive Libraries to densely map chemical space and generate high-resolution training data. For oligonucleotides, TherML applies AI and large-scale automation to industrialise siRNA design, optimising sequence selection and target knockdown across diverse targets. These capabilities are further enhanced by advanced models for predicting ADMET properties, developed using insitro’s internal datasets and industry collaborations.

Reducing attrition through earlier optimisation

By moving optimisation for drug-like properties earlier in the discovery workflow, TherML aims to reduce late-stage attrition by balancing efficacy with developability from the outset. The platform is directly connected to insitro’s automated laboratories, allowing rapid experimental validation and continuous improvement of predictive models tailored to each target and modality.

“Effective therapeutic design requires optimisation across affinity, selectivity, stability and manufacturability,” said Dr Noam Katz, co-founder of CombinAbleAI. “We’re very excited to join insitro and integrate CombinAbleAI’s physics-informed AI modelling into insitro’s end-to-end discovery platform to reliably and rapidly translate high-value targets into molecules that perform as intended in the real world.”