Scalable Agentic Reasoning for Designing Biologics Targeting Intrinsically Disordered Proteins DOI: XXXXXXX.XXXXXXX Conference: Platform for Advanced Scientific Computing; June 29– July 1, 2026; Bern, Switzerland PASC ’26: Platform for Advanced Scientific Computing, June 29– July 1, 2026, Bern, Swit
Matthew Sinclair 1,† , Moeen Meigooni 1,† , Archit Vasan 1,† , Ozan Gokdemir 1,2 , Xinran Lian 1 , Heng Ma 1 , Yadu Babuji 2 , Alexander Brace 1,2 , Khalid Hossain 1 , Carlo Siebenschuh 1,2 , Thomas Brettin 1 , Kyle Chard 2 , Christopher Henry 1 , Venkatram Vishwanath 1 , Rick L. Stevens 1,2 , Ian T · 2025
Read Scalable Agentic Reasoning for Designing Biologics Targeting Intrinsically Disordered Proteins DOI: XXXXXXX.XXXXXXX Conference: Platform for Advanced Scientific Computing; June 29– July 1, 2026; Bern, Switzerland PASC ’26: Platform for Advanced Scientific Computing, June 29– July 1, 2026, Bern, Swit by Matthew Sinclair 1,† , Moeen Meigooni 1,† , Archit Vasan 1,† , Ozan Gokdemir 1,2 , Xinran Lian 1 , Heng Ma 1 , Yadu Babuji 2 , Alexander Brace 1,2 , Khalid Hossain 1 , Carlo Siebenschuh 1,2 , Thomas Brettin 1 , Kyle Chard 2 , Christopher Henry 1 , Venkatram Vishwanath 1 , Rick L. Stevens 1,2 , Ian T (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-15930.txt.
Introduction
Matthew Sinclair and co authors present StructBioReasoner, a multi agent system for designing biologics against disordered proteins, in a 2025 paper. About 80 percent of cancer related proteins contain long disordered regions, and their lack of stable structure makes them hard drug targets. Diffusion models can design binders, but real discovery needs reasoning across many shapes and tools.
It runs a tournament in which expert agents compete to propose and refine treatment ideas, using prior papers, structure prediction, modeling, and stability checks on fast computers. For the Der f 21 target, over half of 787 validated candidates beat human designed reference binders in binding free energy. For NMNAT-2, three binding modes were found.
Catalog record
| CREATOR | Matthew Sinclair 1,† , Moeen Meigooni 1,† , Archit Vasan 1,† , Ozan Gokdemir 1,2 , Xinran Lian 1 , Heng Ma 1 , Yadu Babuji 2 , Alexander Brace 1,2 , Khalid Hossain 1 , Carlo Siebenschuh 1,2 , Thomas Brettin 1 , Kyle Chard 2 , Christopher Henry 1 , Venkatram Vishwanath 1 , Rick L. Stevens 1,2 , Ian T |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-15930
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Scalable Agentic Reasoning for Designing Biologics Targeting Intrinsically Disordered Proteins DOI: XXXXXXX.XXXXXXX Conference: Platform for Advanced Scientific Computing; June 29– July 1, 2026; Bern, Switzerland PASC ’26: Platform for Advanced Scientific Computing, June 29– July 1, 2026, Bern, Switzerland ISBN: 978-1-4503-XXXX-X/18/06 Matthew Sinclair 1,† , Moeen Meigooni 1,† , Archit Vasan 1,† , Ozan Gokdemir 1,2 , Xinran Lian 1 , Heng Ma 1 , Yadu Babuji 2 , Alexander Brace 1,2 , Khalid Hossain 1 , Carlo Siebenschuh 1,2 , Thomas Brettin 1 , Kyle Chard 2 , Christopher Henry 1 , Venkatram Vishwanath 1 , Rick L. Stevens 1,2 , Ian T. Foster 1,2 , Arvind Ramanathan 1,2∗ Abstract . Intrinsically disordered proteins (IDPs) represent crucial therapeutic targets due to their significant role in disease–approximately 80% of cancer-related proteins contain long disordered regions – but their lack of stable secondary/tertiary structures makes them “undruggable”. While recent computational advances, such as diffusion models, can design high-affinity IDP binders, translating these to practical drug discovery requires autonomous systems capable of reasoning across complex conformational ensembles and orchestrating diverse computational tools at scale. To address this challenge, we designed and implemented StructBioReasoner, a scalable multi-agent system for designing biologics that can be used to target IDPs. StructBioReasoner employs a novel tournament-based reasoning framework where specialized agents compete to generate and refine therapeutic hypotheses, naturally distributing computational load for efficient exploration of the vast design space. Agents integrate domain knowledge with access to literature synthesis, AI-structure prediction, molecular simulations, and stability analysis, coordinating their execution on HPC infrastructure via an extensible federated agentic middleware, Academy. We benchmark StructBioReasoner across Der f 21 and NMNAT-2 and demonstrate that over 50% of 787 designed and validated candidates for Der f 21 outperformed the human-designed reference binders from literature, in terms of improved binding free energy. For the more challenging NMNAT-2 protein, we identified three binding modes from 97,066 binders, including the well-studied NMNAT2:p53 interface. Thus, StructBioReasoner lays the groundwork for agentic reasoning systems for IDP therapeutic discovery on Exascale platforms. 1. Introduction Intrinsically disordered proteins (IDPs) comprise between 30% and 40% of the human proteome and play critical roles in cellular regulation, signaling, and disease pathogenesis ( 27 ) . Approximately 80% of cancer-associated proteins contain long intrinsically disordered regions, making these proteins essential therapeutic targets ( 57 ) . Yet IDPs have historically been classified as “undruggable” because they lack the stable three-dimensional structures and well-defined binding pockets that conventional drug discovery methods require. The conformational heterogeneity of IDPs—their existence as dynamic ensembles rather than fixed structures—poses fundamental challenges for rational biologics design ( 62 ) . Recent computational breakthroughs have begun to address this challenge. Methods based on diffusion models, including modified RFdiffusion and the Logos framework have demonstrated the ability to generate high-affinity protein binders (K d = 3–100 nM) for previously intractable IDP targets ( 37 ; 65 ) . These approaches achieve success by freely sampling conformations of both target and binder during the design process, effectively navigating the conformational landscape that makes IDPs difficult to target. However, translating these individual computational capabilities into practical drug discovery requires autonomous systems that can: (1) reason about which computational strategies to apply for a given IDP target, (2) orchestrate diverse tools including structure prediction, molecular dynamics, and thermodynamic analysis, and (3) scale efficiently across high-performance computing resources to explore vast design spaces. This challenge sits squarely at the intersection of artificial intelligence (AI), structural biology, and high-performance computing (HPC). While AI agents have shown promise for autonomous scientific discovery in biology, chemistry and materials science ( 7 ; 40 ) , existing agentic systems for protein engineering have focused primarily on ordered proteins or small molecule design. No current framework specifically addresses the unique requirements of IDP-targeting biologics: reasoning about conformational ensembles, selecting appropriate computational tools from the expanding arsenal of IDP-specific methods, and executing compute-intensive workflows at HPC scale. Furthermore, the computational demands of IDP biologics design—involving ensemble generation, exhaustive conformational sampling, and iterative refinement—necessitate scalable architectures that can leverage distributed GPU resources and minimize tool-calling overhead. Figure 1. StructBioReasoner design/architecture. A user provides a high-level design goal, and the agent system dynamically selects from specialized agents to execute the task iteratively. Each specialized agent is orchestrated via the Academy agentic framework and has access to a variety of tools, datasets including literature and other related data, as well as history of actions taken by the agents to run the tools. The planner and reasoner agents act in tandem to present results to the user, which can then be refined in subsequent interactions. We present StructBioReasoner, a scalable multi-agent system for autonomous IDP-targeting biologics design that addresses these challenges. This system ( Figure 1 ) employs a tournament-based reasoning framework where specialized agents-focused on structural stability, evolutionary conservation, energetic optimization, and rational design principles—compete to generate and refine hypotheses for engineering peptides/biologics targeting IDPs. The tournament architecture enables parallel hypothesis evaluation and ranking, naturally distributing computational work across available resources. Each agent integrates domain knowledge with access to computational tools including various AI-structure prediction approaches, molecular dynamics (MD) simulation, and physics-based stability prediction. The system reasons about which tools to invoke for a given design challenge and coordinates their execution on HPC infrastructure through workflow orchestration frameworks. Our contributions are threefold. First, we demonstrate a multi-agent reasoning architecture designed to address the conformational complexity of IDP targets, where agents must reason about ensemble properties rather than a single structure. Second, we present a scalable tournament framework that enables efficient parallel hypothesis generation and evaluation, addressing the computational intensity of IDP biologics design. Third, we validate our approach through case studies on two model IDP systems, including (1) house dust mite allergen Der f 21 and (2) Nicotinamide mononucleotide adenylyltransferase 2 (NMNAT-2) showing that our agentic reasoning framework can identify stabilizing mutations matching literature-validated strategies while effectively utilizing distributed supercomputing resources. This work establishes a foundation for autonomous discovery of IDP-targeting therapeutics on emerging exascale platforms, advancing both the scientific understanding of these challenging targets and the computational methods required to drug them effectively. 2. Related Work We review related work in agentic systems for biologics design, computational methods for IDPs, and HPC for biological AI. 2.1. Agentic Systems for Biologics Design The integration of large language models with specialized tools has enabled autonomous scientific discovery across domains. In chemistry and biology, platforms such as ChemCrow, Kosmos, and CoScientist demonstrated that LLMs could plan syntheses, control laboratory robotics, and iteratively optimize reactions ( 40 ; 44 ; 7 ) . Together with the rapid maturation of open-source frameworks ( 34 ; 35 ; 13 ) for building agentic workflows—enabling explicit control flow, tool use, and multi-agent coordination—these successes demonstrate the feasibility of agentic scientific workflows for orchestrating complex, multi-stage experimental campaigns. Protein engineering further leverages these capabilities. Virtual Lab coordinates multiple specialized agents under the guidance of an LLM Principal Investigator to execute therapeutic protein design, producing 92 nanobody designs with over 90% expression success ( 56 ) . ProtAgents and SAMPLE extended this approach by integrating LLM reasoning with physics-based simulations, structural analysis, and cloud-executed closed-loop optimization ( 20 ; 52 ) . More comprehensive platforms like DrugAgent and PharmAgents