For drug developers, the most expensive failures are the ones that happen late. A candidate can look perfect in a static 3D model, only to falter in a living system because of a hidden, transient pocket on a protein. Varosync, a New York-based AI-native biotechnology research lab founded in 2025, is building its computational systems to find those pockets first. Its core bet is that a deeper, dynamic understanding of protein shapes can reveal ‘cryptic’ allosteric sites, potentially leading to more selective drugs with fewer side effects [Perplexity Sonar Pro Brief, retrieved 2024].
The Wedge of Dynamic Simulation
Varosync’s technical approach hinges on moving beyond static snapshots. The company’s AI engine is designed to transform traditional 3D protein models into dynamic simulations, aiming to predict how a protein moves and where fleeting, functional pockets appear [Perplexity Sonar Pro Brief, retrieved 2024]. This focus on conformational dynamics is what the company terms ‘physics-informed machine learning.’ The goal is to enable isoform-aware molecular design, where subtle structural differences between related proteins can be exploited for therapeutic gain while avoiding off-target effects. This is a data-intensive problem, and the company’s stack reportedly includes a knowledge graph over millions of molecules and large-scale molecular embedding models trained on corpora of hundreds of millions of compounds [Perplexity Sonar Pro Brief, retrieved 2024].
The architecture is built to address what Varosync identifies as a core inefficiency: disconnected data modalities in drug discovery. By attempting to unify unstructured historical research with high-performance molecular simulation, the platform aims to surface non-obvious connections within a company’s own data that might otherwise be missed [Perplexity Sonar Pro Brief, retrieved 2024]. For biopharma teams, the promised output is clearer decision-making before committing significant capital to a preclinical or clinical program.
A Founder’s Path from Academic Research
The company’s technical direction is closely tied to co-founder and CEO Harry Kabodha. Prior to Varosync, Kabodha conducted computational research at Columbia University’s Irving Institute for Cancer Dynamics, applying deep learning to problems in protein dynamics [Perplexity Sonar Pro Brief, retrieved 2024]. This academic grounding in a specific, hard problem in biophysics informs the company’s narrow technical wedge. His co-founder is Ayman Khaleq, a partner at law firm Morgan, Lewis & Bockius LLP who co-leads its Middle East practice and specializes in cross-border and Sharia-compliant transactions [Ayman A. Khaleq | Investment Funds Lawyer, retrieved 2026]. This pairing suggests a balance of deep technical focus and strategic business development, particularly for future fundraising and international partnerships.
The team is rounded out by strategic advisor Dr. Michel Azoulay, whose background as a chief medical officer with a focus on precision medicine and multi-omics platforms adds clinical and translational expertise [LinkedIn, retrieved 2026]. Early engineering work, including building major parts of the ML infrastructure, was led by AI/ML engineer Aniket Ghosh [Aniket Ghosh | AI/ML Engineer, retrieved 2026]. The company’s estimated size ranges from 2-10 to 11-50 employees across different sources, placing it firmly in the early-stage build phase [LinkedIn, retrieved 2024] [F6S, retrieved 2024].
| Role | Name | Key Background |
|---|---|---|
| Co-founder, CEO | Harry Kabodha | Computational research in protein dynamics at Columbia University [Perplexity Sonar Pro Brief, retrieved 2024] |
| Co-founder | Ayman Khaleq | Partner at Morgan Lewis, cross-border fund formation and transactions [Ayman A. Khaleq |
| Strategic Advisor | Dr. Michel Azoulay | Chief Medical Officer, precision medicine and AI [LinkedIn, retrieved 2026] |
| AI/ML Engineer | Aniket Ghosh | Built Varosync's drug-development ML stack [Aniket Ghosh |
Navigating a Crowded and Capital-Intensive Field
Varosync enters a space dense with well-funded competitors applying AI to drug discovery. The competitive set includes public companies like Recursion and Exscientia, as well as heavily backed private players such as Insilico Medicine, Xaira Therapeutics, and Generate: Biomedicines. Differentiation in this field often comes down to the specificity of the scientific approach, the quality of proprietary data, and the ability to shepherd a program toward clinical validation. Varosync’s early-stage status and undisclosed funding level mean it is operating with less publicly visible capital than many of its rivals, which have raised hundreds of millions to billions of dollars.
The company’s participation in accelerator programs like Larta Heal.LA and Columbia Engineering’s Start Me Up Bootcamp provides early ecosystem support and validation [Columbia Engineering Start Me Up Bootcamp]. However, the path to traction is long. The primary risks for a company like Varosync are not just technical but commercial and biological.
- The Validation Gap. A compelling simulation must eventually be proven in a wet lab and, ultimately, in a patient. The leap from in silico prediction to in vivo efficacy remains the grand challenge for the entire AI-driven discovery field.
- The Data Moats. Larger competitors have often spent years building proprietary experimental datasets to train their models. Varosync’s ability to access or generate high-quality, proprietary biological data will be critical to its models’ performance.
- The Partner Puzzle. As a young research lab, Varosync’s likely initial route to market is through partnerships with biopharma companies. Securing a flagship collaboration with a credible player would be a key traction signal, demonstrating that its ‘failure-aware intelligence’ translates into tangible value for an R&D team.
The Next Twelve Months
For a pre-seed company in this domain, the immediate horizon is about proving the core technology and translating it into a viable service or partnership model. Key milestones to watch will include the publication of any peer-reviewed validation of its methods, the announcement of a first pharmaceutical or biotech partnership, and a subsequent funding round to scale its computational and experimental capabilities. The involvement of a strategic advisor like Azoulay, with ties to clinical development, suggests an intent to build with an eye toward translational relevance from the start.
The ambition is humane, even if the tools are computational. For patients waiting on new treatments for cancers, autoimmune disorders, or rare diseases, the standard of care today is often a limited menu of options with significant side-effect profiles. Drug discovery remains a process of immense attrition, where promising early candidates frequently fail due to unforeseen toxicity or lack of efficacy in later-stage trials. Varosync’s thesis is that better intelligence on protein dynamics and failure modes upstream can make that process less wasteful and more reliable. The next phase for the company is to show that its models can not only find a cryptic site but help design a molecule that safely and effectively targets it, moving from an intriguing simulation to a tangible candidate for the clinic.
Sources
- [Columbia Engineering Start Me Up Bootcamp] Program participant listing for Varosync
- [LinkedIn, retrieved 2024] Harry Kabodha profile and company size estimate | https://linkedin.com/in/harrykabodha
- [F6S, retrieved 2024] Varosync company profile and size estimate
- [Ayman A. Khaleq | Investment Funds Lawyer, retrieved 2026] Professional background of co-founder Ayman Khaleq
- [LinkedIn, retrieved 2026] Profile of strategic advisor Dr. Michel Azoulay | https://www.linkedin.com/in/azoulay
- [Aniket Ghosh | AI/ML Engineer, retrieved 2026] Background on AI/ML engineer Aniket Ghosh