KRAS mutations are among the most prevalent oncogenic drivers implicated across a broad spectrum of human cancers. The protein regulates critical cellular processes by orchestrating proliferation and survival pathways, making it a prime therapeutic target. However, the development of inhibitors capable of robustly modulating pan-KRAS activity has been hindered for decades by the exceptionally high affinity of KRAS for guanine nucleotides (GDP/GTP) and the scarcity of well-defined, druggable pockets on its surface. These properties have rendered conventional small-molecule approaches largely ineffective, spurring demand for novel strategies that can circumvent these biological constraints.

Recognizing this challenge, the team at Insilico Medicine applied their AI-driven platform, Chemistry42, to design a diverse set of compound libraries featuring unique central cores or chemotypes. The generative chemistry models embedded within Chemistry42 span over 40 distinct generative frameworks, integrating diverse AI architectures to propose molecular structures with tailored physicochemical and biological properties. This exhaustive exploration enabled scaffold hopping—a process of identifying and substituting core scaffolds of molecules—allowing the team to traverse chemical space beyond what traditional medicinal chemistry methods might achieve.

The initial phase of the project began with a precursor molecule exhibiting selectivity toward a specific KRAS mutant variant. From this starting point, the design process expanded swiftly, leveraging virtual screening modules of Chemistry42 that integrated rigorous structure-activity relationship (SAR) evaluations. In silico modeling refined potential candidates by simulating molecular interactions within the KRAS binding landscape, and iterative optimization was applied meticulously to the amino side chains to optimize binding affinity and selectivity profiles.

Following computational refinement, synthetic chemistry and biochemical assays validated the AI-generated candidates. The selected compounds demonstrated pan-KRAS inhibition with potencies measured in the upper nanomolar range, a noteworthy achievement given the high bar for KRAS-targeting agents. Importantly, these inhibitors exhibited mild selectivity favoring KRAS mutants over wild-type variants—up to a 4-fold difference in potency—signifying potential therapeutic indices that could mitigate off-target effects on normal cells. Furthermore, cell-based assays confirmed robust inhibitory activity in KRAS mutant tumor cell lines, emphasizing translational promise.

Metabolic profiling at this stage revealed the inhibitors possessed acceptable cytochrome P450 (CYP) inhibition profiles, thereby reducing concerns of adverse drug-drug interactions in vivo. This pharmacokinetic property is crucial for clinical development, where metabolic stability and interaction potential directly influence dosing, safety, and efficacy. The integration of AI-enabled predictive models in Chemistry42 likely expedited this aspect by foreseeing and flagging metabolic liabilities prior to synthesis.

Alex Zhavoronkov, PhD, Founder and CEO of Insilico Medicine, highlighted the transformative role of AI throughout this research. He emphasized how the convergence of generative chemistry and sophisticated molecular modeling has enabled the navigation of complex target landscapes, especially for proteins like KRAS that were once deemed refractory to drug discovery. This synergy underscores a paradigm shift where artificial intelligence is not merely a supplementary tool but a driving force accelerating the transition from computational concepts to tangible clinical candidates.

The research draws its lineage from Insilico’s foundational work dating back to 2016, when the company first articulated the concept of using generative AI to design novel therapeutic molecules in a peer-reviewed article. This pioneering vision laid the groundwork for the development of the Pharma.AI platform, which has since evolved into a comprehensive generative AI-powered ecosystem spanning biology, chemistry, medicine development, and research science. To date, Insilico Medicine has contributed over 200 scientific publications and holds more than 600 patents and patent applications, emphasizing its leadership at the intersection of AI and biotechnology.

By applying this sophisticated platform, Insilico Medicine is pushing the boundaries of what is achievable in medicinal chemistry, setting a new benchmark for computationally driven drug design. The successful development of pan-KRAS inhibitors of novel chemotypes signals a turning point in the fight against cancers driven by KRAS mutations. It also showcases the vital role AI can play in unlocking targets traditionally considered “undruggable,” potentially transforming treatment modalities across oncology and beyond.

As these novel inhibitors advance through preclinical validation, they represent promising candidates not only for therapeutic intervention but also as a testament to the growing impact of AI-driven approaches in modern drug discovery. The ability to integrate deep learning, molecular simulations, and automated synthesis in a tightly coupled pipeline may shorten drug development timelines, reduce costs, and increase the success rates of novel therapeutics.

Moreover, Insilico’s work hints at future applications where generative AI platforms might tackle an even wider array of targets with similar complexities and challenges. The collaboration between human expertise and AI augments medicinal chemistry with unparalleled speed and precision, potentially ushering in a new era of personalized and precision medicine.

For the broader scientific community, this success serves as a clarion call to embrace AI methodologies as not only complementary but essential in reimagining drug discovery workflows. Insilico Medicine’s Chemistry42 platform continues to evolve, integrating the latest technological innovations and expanding its reach across diverse therapeutic areas such as fibrosis, immunology, pain, obesity, metabolic disorders, and even novel fields like advanced materials and agriculture.

In essence, this breakthrough marks a remarkable convergence of computational chemistry, molecular biology, and artificial intelligence, delivering tangible progress against a historically elusive target and illuminating the pathway for future innovations in cancer therapeutics.

Subject of Research: Development of novel pan-KRAS inhibitors using generative AI-driven structure-based drug design and scaffold hopping.

Article Title: Identification of novel pan-KRAS inhibitors via Structure-Based drug design, scaffold hopping, and biological evaluation.

Web References:
https://pubs.acs.org/doi/10.1021/acsmedchemlett.5c00080
https://www.insilico.com

References:
Aladinskiy, V. et al. (2025) ‘Identification of novel pan-KRAS inhibitors via Structure-Based drug design, scaffold hopping, and biological evaluation,’ ACS Medicinal Chemistry Letters [Preprint].

Image Credits: Insilico Medicine

Keywords:
Generative AI, pan-KRAS inhibitors, medicinal chemistry, artificial intelligence, molecular modeling, scaffold hopping, drug discovery, oncology, druggable targets, structure-based design

The transition from a nonmetallic to a metallic state typically involves changes in electron mobility and the formation of a conduction band, a process that in solids is often associated with defining crystal structure and temperature-dependent properties. However, liquids defy some of these constraints due to their inherent molecular disorder and dynamism. Professor Jungwirth and colleagues have long focused on the peculiar case of alkali metals dissolved in liquid ammonia, a classical system where metallic behavior emerges abruptly as the solution changes color from blue to a lustrous golden hue. This visual transformation heralds the formation of a conduction network from free electrons donated by alkali metal atoms, yet until now, the intermediate stages within this transition have remained elusive.

Central to this new investigation is the application of state-of-the-art molecular dynamics simulations that can faithfully capture the behavior of electrons and ions on ultrafast timescales and at atomic resolution. The simulations reveal a previously overlooked regime where the system does not settle into either the metallic or nonmetallic state but instead exhibits a rapid “flipping” or oscillation between these electronic configurations. This "third phase" defies traditional equilibrium descriptions and compels a reconsideration of the criteria defining phase transitions in conductive liquids. The switching occurs on the scale of tens of femtoseconds (one femtosecond being one quadrillionth of a second), orders of magnitude faster than conventional experimental techniques have been able to probe.

Professor Jungwirth underscores the novelty of this phenomenon: “No one had previously realized that such a system might oscillate so rapidly between these two fundamentally different electronic states. These dynamics had simply not been accounted for in theoretical or experimental frameworks before.” This insight opens new vistas for understanding how conduction emerges and dissolves in disordered environments, potentially impacting fields as diverse as materials science, condensed matter physics, and electrochemistry.

Capturing such rapid transitions experimentally presents a formidable challenge. The temporal resolution required transcends the capabilities of conventional spectroscopy methods. To address this, the IOCB Prague team intends to harness ultrafast laser technology, which can deliver pulses short enough to act as snapshots of these fleeting electronic states. PhD student Marco Vítek, first author of the study, explains, “Our aim is to develop an experimental setup using ultrafast laser pulses that can effectively ‘freeze’ the system’s state at different moments, allowing us to observe these flips directly. Some of these laser sources are already available at our institute, bringing this ambitious goal within reach.”

Collaboration with internationally renowned institutions, including the University of Oxford, the Faculty of Mathematics and Physics at Charles University, and the J. Heyrovský Institute of Physical Chemistry, has been instrumental in pushing the boundaries of computational and experimental prowess required for this research. Together, the teams have employed a combination of high-performance computing and sophisticated quantum mechanical models to achieve unprecedented simulation accuracy.

The implications of this discovery extend beyond merely redefining a classic physical transition. The intermittent, ultrafast switching points to a novel quantum-mechanical process intrinsic to liquids containing dissolved alkali metals. It suggests that conductivity in such systems does not arise from a smooth crossover but rather from a dynamic equilibrium with rapid fluctuations. This challenges the universality of static phase diagrams and compels theorists to incorporate time-dependent variables into models of liquid metals.

If experimentally verified, this phenomenon could have broad technological ramifications. Understanding ultrafast electronic state fluctuations could inform the design of novel materials with tailored conduction properties, potentially impacting energy storage, catalysis, and sensor technologies. Moreover, it may necessitate new interpretations of data gathered from systems where fast electronic processes play a critical role.

The study, published in the prestigious journal Nature Communications, represents a milestone in physical chemistry and condensed matter physics. The research leverages computational simulation/modeling methodologies that capture phenomena inaccessible through classical experimental setups alone. This computational insight, complemented by planned ultrafast spectroscopic experiments, defines a new milestone in the exploration of liquid state physics.

Of particular interest is the methodological framework developed by the team, which integrates electronic structure calculations with molecular dynamics to simulate real-time electron behavior in complex liquid environments. This multiscale modeling offers a blueprint for future studies investigating transient phenomena in other chemical and physical systems.

In summary, the discovery of a rapid flipping behavior between electrolyte and metallic states in ammonia solutions of alkali metals opens a fresh theoretical and experimental frontier. It challenges prevailing dogma, invites a reexamination of the metal-nonmetal transition, and bridges the gulf between chemical physics and ultrafast spectroscopy. This study signals a profound advance in our grasp of the liquid metal state and sets the stage for experimental verification that could rewrite textbooks on physical phase transitions.

Subject of Research: Physical sciences – Chemistry – Chemical physics / Molecular dynamics

Article Title: Rapid flipping between electrolyte and metallic states in ammonia solutions of alkali metals

News Publication Date: 8-May-2025

Web References:
http://dx.doi.org/10.1038/s41467-025-59071-z

References:
Vitek, M.; Igor Rončević; Marsalek, O.; Schewe, H. C.; Jungwirth, P. Rapid Flipping between Electrolyte and Metallic States in Ammonia Solutions of Alkali Metals. Nat. Commun. 2025, 16 (1).

Image Credits:
Photo: Tomáš Belloň/IOCB Prague

Keywords:
Chemical physics, Molecular dynamics