Insitro

Insitro TherML™ Platform Architectural Assessment

The Insitro architecture is defined by its end-to-end TherML™ platform, which industrializes the transition from biological insight to clinic-ready therapeutic leads. As of January 2026, the system has achieved modality-agnostic status, integrating physics-informed AI from the CombinAbleAI acquisition to design complex biologics alongside established small molecule and oligonucleotide pipelines 📑. The architecture is built on the principle of Causal Biology, using machine learning to identify genetic perturbations that drive disease-relevant phenotypes in massive, proprietary datasets 📑.

Core Computational & Design Engine

The platform centers on Biological Foundation Models (BFMs) that interpret high-content imaging and multi-omics data to reveal disease drivers 🧠.

• TherML™ Design Layer: Simultaneously optimizes for both potency and developability (ADMET/PK) using physics-informed molecular dynamics surrogates and proprietary Quantitative Adaptive Libraries (QALs) 📑.
• Causal Discovery Engine: Interrogates human clinical data and cellular models to identify genetic intervention points with the highest probability of clinical success 📑.
• HPC Infrastructure: Leverages a massive compute cluster equipped with NVIDIA H100 GPUs to execute high-fidelity physics simulations and train foundational models 📑.

⠠⠉⠗⠑⠁⠞⠑⠙⠀⠃⠽⠀⠠⠁⠊⠞⠕⠉⠕⠗⠑⠲⠉⠕⠍

Operational Scenarios

• Antibody Optimization: Input: Target protein structure + baseline antibody sequence → Process: Physics-informed simulation via 100k+ MD surrogates to predict flexibility and binding affinity → Output: Optimized biologic lead with validated manufacturability scores 📑.
• Closed-loop Target Validation: Input: Causal model hypothesis for ALS → Process: Autonomous liquid-handling robotics execute pooled optical screening (POSH) in human iPSC-derived neurons → Output: Confirmed genetic modifiers of the disease phenotype ready for ChemML™ optimization 📑.

Evaluation Guidance

Technical evaluators should verify the following architectural characteristics:

• Modality-Agnostic Throughput: Benchmark the latency involved in switching the design engine's focus between small molecules and complex biologics within a single therapeutic program 🧠.
• ADMET Predictive Accuracy: Organizations should validate the performance of TherML™'s ADMET models—developed through internal data and the 2025 Lilly partnership—against class-specific constraints 📑.
• Causal Validation: Verify the degree of agreement between ML-derived causal targets and historical clinical trial successes for similar metabolic or neurodegenerative pathways 🌑.