From Molecular Lego to High-Quality Data: Insilico Medicine and Saudi Aramco Advance AI-Driven MOF Discovery with sorbaMOF DB

August 11, 2026 — August 10, 2026 — Insilico Medicine (“Insilico”; HKEX: 3696), a clinical-stage generative artificial intelligence (AI)-driven drug discovery company, today introduces sorbaMOF DB, a validated and extensible materials database, together with a benchmarked computational protocol designed to ensure reliable CO₂ adsorption predictions. The research has been submitted as a preprint paper on ChemRxiv under the title “sorbaMOF DB: An Extensible MOF Adsorption Database Platform with a Case Study on Direct Air Capture of CO₂.”

This development is driven by the growing global need for direct air capture (DAC) technologies capable of removing CO₂ directly from the atmosphere. The database focuses on MOFs, a cutting-edge class of highly tunable porous materials regarded as among the most promising candidates for DAC However, reliable experimental and simulation data for MOFs under these exact conditions have historically been extremely scarce, creating a fundamental bottleneck for data-driven materials discovery.

To bridge this critical data gap, the research introduces sorbaMOF DB, a curated database comprising approximately 240,000 MOF structures, placing a strong emphasis on data quality. Each structure undergoes rigorous validation to ensure suitability for high-throughput simulation workflows. The significance of this work is based on three key components: (1) a high-quality dataset of materials with reliable adsorption data for CO₂ capture; (2) a transferable computational protocol that improves the reliability of adsorption simulations; and (3) a machine learning–based analysis that enables the identification of structural features associated with high-performance materials.

By addressing key data bottlenecks in AI-driven materials discovery for carbon capture, sorbaMOF DB provides reliable CO₂ adsorption predictions under ambient DAC conditions. By combining a benchmarked and transferable simulation protocol with interpretable machine learning and by accounting for real-world factors such as framework flexibility and competitive water adsorption, the platform establishes a generalizable, physically grounded infrastructure for accelerating the development of next-generation gas-separation and carbon-management technologies.

This study represents a new step in Insilico Medicine’s application of AI to materials discovery and the development of advanced carbon-capture technologies. It is the first joint research effort involving Insilico, Saudi Aramco, the University of Montpellier, and King Abdullah University of Science and Technology (KAUST), bringing together Insilico’s expertise in computational materials discovery, the two academic institutions’ leading capabilities in MOFs research, and Saudi Aramco’s strategic focus on carbon management. The collaboration lays the foundation for a broader partnership supporting Saudi Aramco’s global carbon-management roadmap.

The study also marks the latest milestone in the ongoing collaboration between Insilico and Saudi Aramco. Since signing a Memorandum of Understanding (MOU) at the LEAP Technology Conference, the two companies have worked together to apply generative AI to advanced materials science and sustainable technology development.

MOFs often described as “molecular Lego” because of their highly tunable structures—offer a vast design space for carbon-capture applications. However, this complexity also creates significant data challenges: generative AI models and high-throughput computational screening depend on the accuracy and structural integrity of the underlying datasets.

To address widespread geometric inconsistencies and structural errors in public MOF databases, Insilico and Saudi Aramco previously co-developed and released the Sanity Pipeline, a multilevel structural-validation framework that integrates ultrafast MOF decomposition through LibCIF with oxidation-state defect detection using OxiChecker. In the present study, the consortium applied these rigorous validation principles at scale, helping to close a critical data gap in carbon-capture research and enabling more reliable, physically grounded materials discovery.

The significance of sorbaMOF DB extends far beyond a static repository. By addressing longstanding computational challenges, including structural inaccuracies in public repositories and the limited physical reliability of adsorption simulations, the platform establishes an automated, generalizable infrastructure for building trustworthy adsorption databases.

Crucially, this work bridges the traditional gap between computational screening and experimental reality. By incorporating real-world operational parameters, such as host framework flexibility and competitive water adsorption, the workflow delivers highly realistic performance bounds that can directly guide physical synthesis and laboratory validation.

Designed as an evolving, extensible platform, the benchmarked workflow and interpretable machine learning protocols are readily transferable to a broad range of target gases (such as hydrogen, methane, or toxic pollutants) and diverse porous material classes. By combining rigorous structural sanitization, physically grounded interaction modeling, and interpretable AI, the consortium establishes high-quality data foundations for generative AI and foundation models, paving the way for a new generation of scalable, AI-driven materials discovery aligned with global net-zero goals.