Insilico Medicine, a pioneering company in generative artificial intelligence (AI) for drug discovery, has launched a suite of advanced AI models designed for both chemistry and biology. Developed through its innovative MMAI Gym for Science framework, these models exhibit state-of-the-art performance across over 50 benchmark tasks in drug discovery.
Historically, language models have primarily been viewed as tools for conversational applications. However, in the complex realm of drug discovery, traditional computational methods have often outperformed them. The launch of MMAI Gym marks a pivotal shift, illustrating that language-model architectures can be effectively adapted to tackle specialized, data-scarce challenges in scientific research. According to Alex Zhavoronkov, Ph.D., the founder and CEO of Insilico Medicine, this advancement signifies a transformative step toward achieving predictive AI capabilities that can significantly benefit human health.
Evolving Drug Discovery Approaches
In the past, drug discovery tasks that required minimal data, such as ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) and target potency prediction, have largely been the domain of specialized computational models. Language models have generally been sidelined in this competitive field. However, recent findings from the MMAI Gym indicate that this landscape is changing, as language models begin to match or exceed the performance of established methods in specific tasks.
The newly released models from MMAI Gym are compact language models tailored for scientific applications. Each model is meticulously fine-tuned for a specific category of tasks and is trained across multiple related challenges within that category. To ensure robust comparisons, these models were benchmarked against traditional computational methods using identical datasets and evaluation setups, allowing for direct performance assessments.
Specialized Models in Chemistry
The current portfolio of chemistry models includes specialists focused on chemical synthesis, ADMET prediction, and potency prediction for GPCR (G protein-coupled receptors) and kinase panels.
For ADMET prediction, the dedicated model was assessed across 28 tasks, achieving state-of-the-art performance compared to established computational methods. Notably, it demonstrated exceptional results in predicting drug-drug interaction risks, cytotoxicity, and pharmacokinetic properties. By accurately forecasting these properties, the model can help identify potential liabilities early in the drug development process and streamline compound prioritization before resource-intensive experimental studies. The model’s category-focused design enables it to assess multiple ADMET-related endpoints simultaneously, enhancing efficiency in drug development.
In terms of target activity prediction, Insilico Medicine has developed two focused models: one for GPCRs, encompassing 44 receptors, and another for kinases, which includes 67 enzymes. Both models achieved state-of-the-art IC50 prediction performance, facilitating a variety of drug discovery applications, from virtual screening to the design of multi-target compounds.
Innovations in Chemical Synthesis
The chemical synthesis models specialize in single-step retrosynthesis and are built on Liquid AI’s compact 2.6B-parameter architecture. These retrosynthesis specialists have demonstrated superior performance compared to leading dedicated methods on standard benchmarks as well as more complex out-of-distribution evaluations. An earlier version of these models is available on the Microsoft Marketplace, with an upgraded version expected soon. Beyond their individual capabilities, these models can also serve as integral components in broader computational drug discovery and synthesis planning workflows.
Advances in Biological Applications
The MMAI Gym approach extends beyond chemistry, incorporating specialist models trained on a variety of biological data types and prediction tasks. This extension maintains the category-focused approach and has yielded models that, on select benchmarks, outperform larger general-purpose models. Together, these models exhibit state-of-the-art performance across more than 70 benchmark tasks in both chemistry and biology.
The evaluation of these models is conducted through DDD Bench, a standardized benchmarking framework developed by Insilico Medicine to systematically assess AI models across critical drug discovery and development tasks.
A New Era for Language Models
The significance of these advancements lies not merely in the ability of language models to engage in discussions about chemistry or biology but in their newfound capacity to compete with traditional scientific methods. The MMAI Gym framework is centered around transforming language models into specialized scientific experts through targeted training and rigorous benchmarking against established techniques.
This transformation was first highlighted at NeurIPS 2025, where Insilico Medicine showcased how focused scientific training can enhance the performance of general-purpose language models. Since then, the MMAI Gym initiative has evolved through extensive research aimed at improving multi-task chemistry modeling and domain adaptation.
Financial Milestones and Future Prospects
In the financial realm, Insilico reported approximately $106 million in revenue during the first half of 2026, marking a remarkable 287% increase year-over-year. This period also marked the company’s first profitable half-year since going public, with an adjusted net profit exceeding $51 million. This success is attributed to various out-licensing, co-development, and research collaborations with prominent global partners such as Eli Lilly, Servier, and Takeda. The total contract value of transactions for Insilico in 2026 has reached around $7.3 billion, contributing to a cumulative contract value of approximately $11 billion since 2021.
On the R&D front, Insilico has nominated nine development candidates within nine months of 2026, achieving a record in pipeline productivity and reaching eight clinical milestones across its proprietary and co-developed programs. Notably, Rentosertib (ISM001-055), recognized as the world’s first drug candidate discovered and developed using generative AI, is currently undergoing a Phase III trial for idiopathic pulmonary fibrosis (IPF).
Conclusion
Insilico Medicine’s release of advanced AI models signifies a groundbreaking advancement in drug discovery, providing new tools for researchers to harness the power of AI in addressing complex scientific challenges. As the landscape of drug development continues to evolve, these innovations promise to enhance efficiency and effectiveness in bringing new therapies to market, paving the way for a healthier future.
- Key Takeaways:
- Insilico Medicine has launched specialized AI models for drug discovery.
- The MMAI Gym framework enables language models to compete with traditional methods.
- Significant financial growth and successful drug development milestones highlight the company’s impact.
- Advanced models extend beyond chemistry to include biological applications.
- The integration of AI in drug discovery is poised to revolutionize the industry.
Read more → www.news-medical.net
