This workshop focuses on structure-based methods for rational RNA aptamer optimization. Topics such as structure preparation, protein-RNA aptamer interaction analysis and in silico RNA aptamer engineering will be covered. Participants will explore an RNA aptamer bound to thrombin, examine the interface between the aptamer and thrombin for steric fit and key interactions, generate molecular surfaces to look at surface properties such as electrostatics, identify opportunities for nucleotide optimization, make chemical modifications and assess changes to affinity and stability. In addition, RNA-protein docking will be demonstrated, along with tools for analyzing interactions and contact points in bound poses. The siRNA builder will also be demonstrated for constructing siRNA duplexes from sequence data with different modification patterns.

This workshop focuses on methods for analyzing and optimizing peptide-protein interactions in the binding site. Participants will learn peptide-protein structure preparation, peptide sequence optimization using both natural and non-natural amino acids, and perform conformational analysis. Peptide-protein docking will be demonstrated, along with tools for analyzing protein-ligand interactions to assess contact points. Advanced conformational searching techniques using distance restraints will also be described.

This workshop explores some essential ligand-based methods for guiding drug discovery projects. Participants will learn how to efficiently use MOEsaic to perform SAR analyses of compound datasets, using approaches such as R-group profiling and matched molecular pairs (MMP) analysis, to identify relationships within chemical series. The session highlights computational strategies for extracting meaningful insights from structure-activity data to support informed decision-making in drug design. The workshop also covers molecular alignments and methods for generating pharmacophore queries in the absence of a protein-ligand complex, to help understand the requirements for activity and search for novel compounds.

This workshop focuses on structure-based antibody design. Topics such as protein-protein interaction analysis, in silico protein engineering, affinity modeling, and antibody homology modeling will be covered. Participants will explore an antibody-antigen co-crystallized complex by generating and examining molecular surfaces and visualizing protein-protein contacts in 3D. Antibody properties will be evaluated using specialized protein property descriptors and protein patch analysis. The session will also cover the application of protein engineering tools for affinity and property optimization in the context of antibody developability. Additionally, participants will learn to optimize antibody homology models, including identifying glycosylation sites and their selective modification using a specialized MOE Project antibody database. The full workflow for high-throughput antibody homology modeling, from sequence to structure to property calculations for developability analysis, will be demonstrated.

This workshop describes cheminformatics methods for managing and analyzing datasets of small molecules, as well as generating quantitative structure-activity relationship (QSAR) models. Participants will learn how to import data, prepare MOE databases, calculate molecular descriptors, and analyze trends using sorting, coloring, and plotting tools. Advanced database filtering, data clustering, and diverse subset selection techniques will also be explored. The workshop will include sessions on developing both linear regression and binary QSAR models. Finally, data mining of large compound libraries will be demonstrated using substructure and molecular fingerprint similarity searching.

This workshop provides an overview of SVL capabilities and its application within MOE. Participants will be introduced to the essentials of the SVL scripting language, including its use in automating tasks, customizing workflows, and enhancing MOE functionalities for specific research needs.

This workshop covers essential methods for aligning protein sequences, superposing structures, loop modeling, building fusion protein models, and conducting protein-protein docking. Participants will learn techniques for grafting and refining antibody CDR loops, as well as using a knowledge-based approach to scFv fusion protein modeling with the Linker Modeler application. The session will also cover protein-protein docking of an antibody to an antigen and epitope mapping. Finally, the workshop will guide participants through a complete workflow for generating a QSAR model to predict and analyze protein/biologics solubility.

This workshop presents common molecular modeling and design techniques for identifying, optimizing, and prioritizing small-molecule drug candidates. The workshop begins by describing structure preparation and pocket analysis to understand key protein-ligand interactions. Participants will explore template-based docking of congeneric series and pharmacophore-guide docking (to preserve key interactions) to predict how small molecules bind to a protein target. This workshop also presents a series of de novo fragment-based drug design applications, focusing on scaffold hopping and fragment growing methods to generate novel compound ideas, optimize structures in the pocket, and score them. It also discusses the use of molecular descriptors, protein-ligand interaction fingerprints (PLIFs), and pharmacophore models to guide the drug design process. Finally, a method for generating closely related analogs through bioisosteric replacements is presented.

This workshop covers searching and analysis of structural data in PSILO®, and organization of structural families in MOE-Project. Participants will learn to consolidate macromolecular and protein-ligand structural data, generate and navigate advanced queries, and streamline analysis with 3D contact and protein pocket similarity searches. The session also covers organizing structure-based drug design projects, aligning protein families, configuring MOE-Project databases, and using Protein-Ligand Interaction Fingerprints (PLIF) analysis to compare ligand binding modes. Attendees will gain practical skills for efficient structural data mining and project management.

Finding selective small molecule modulators against members of closely related enzyme families can be challenging, especially if high degrees of selectivity are required. Molecular glues can uniquely address this challenge by leveraging the binding surface of a partner protein to effectively discriminate between enzymatic family members. Using structure-guided tools, we designed a novel series of molecular glues that stabilize a weak, endogenous interaction of a protein of interest with an E3 ligase complex component, thereby inducing its rapid degradation. The unique interface between the protein pair enables designs that are active only in the presence of both partners, with limited interactions in the absence of ternary complex formation, thereby avoiding liabilities associated with targeting the partner protein.

Structural information provides critical insights to accelerate antibody discovery and development. Structure analysis helps us select better antibody clones by evaluating binder diversity while predicting functional and developability profiles. I will also highlight structure-guided engineering tools using a combination of with physics-based and machine learning methods for affinity maturation, cross-reactivity optimization, liability mitigation, and humanization. This should provide a comprehensive overview for leveraging structural biology to reduce costs, compress timelines, and ultimately improve success rates in biologics discovery and optimization.

For the past two decades, virtual screening (VS) has been an efficient hit finding approach for drug discovery. Today, billions of commercially accessible compounds in ultra large libraries (ULLs) are routinely screened, and many successful examples of VS have been reported. VS methods continue to evolve, including machine learning and physics-based methods. In this talk recent examples of VS in drug discovery will be presented and prospective hit finding results from the critical assessment of computational hit-finding experiments (CACHE) challenge will be discussed. Cost considerations for conducting VS will be examined and chemical space coverage and library selections for VS will be analyzed. Though an impressive number of prospective ULL VS campaigns have generated potent and structurally novel hits across many target classes, many successful contemporary VS approaches still use considerably smaller focused libraries. At the end a decision tree of what VS approach is best suited for a specific condition is proposed.

Over the last few years, numerous computational tools have appeared in the literature to model bifunctional degraders. Generally, these studies focus on predicting the structures of degrader-containing ternary complexes, with only occasional investigations into predicting the efficacy of potential degrader designs. In this talk, we will discuss recent results with our well-established ternary complex modeling tool, Method 4B, focused entirely on its ability to score potential degrader designs. This talk will establish best-practices for using Method 4B across diverse systems, focusing on the level of precision that can be expected for prospective degrader screening campaigns.

Imagine a computational chemistry work experience where you tell the computer what computational experiments you would like to run in English, and the computer runs them and reports back.

AI agents represent an emerging paradigm for orchestrating computational chemistry workflows through natural language interfaces. Unlike traditional computational chemistry workflows that require extensive documentation lookups, writing glue code line-by-line, and manual configuration, agentic systems perceive user intent, reason about appropriate computational tools, autonomously execute multi-step calculations, and generate comprehensive reports—all from conversational prompts. I will explore the architectural principles underlying effective computational chemistry AI agents, share my experience and best practices I’ve learned over the past year and a half, including spec-driven and test-driven development guardrails, provenance, red-team audits, and multi-coding-agent development workflows. I’ll then show some illustrative examples of (1) computational experiments orchestrated by AI agents, and (2) new computational chemistry workflows developed with AI. I will highlight a hybrid molecular mechanics and QM neural network potential approach I developed that can, for example, quickly calculate approximate ligand dihedral scans for pose analysis or design efforts to reduce ligand strain, and provide perspective on where I see the frontier headed, including autonomous, self-improving AI Agent systems that can help computational chemists spend more time asking the right scientific questions and less time debugging scripts.

The developability evaluation for antibody-related modalities includes a wide range of factors, such as non-specific binding, self-interaction, hydrophobicity, electrostatic properties, and structural stability. This evaluation is crucial in biologics drug discovery research for the early identification of candidates with high manufacturability as well as favorable behavior in human, and it is providing numerous benefits like improvements in drug discovery speed and reductions in duration and costs associated with bioprocess development.

In particular, the in-silico approach to predict developability from sequence information has grown exponentially alongside advancements in antibody-related technology platforms and is actively researched in both academia and the pharmaceutical industry. Recently, we established a wet evaluation system for high-throughput analysis of multiple developability parameters and created in-silico workflow for predicting developability by leveraging accumulated wet data and machine learning techniques. This machine learning utilizes protein feature quantities generated by MOE. The details and accuracy of this in-silico developability workflow will be presented.

Authors: Lindsay Hammack, Arlinda Rezhdo, Miracle Ozonma, Ridley Hutton, Ariel Tocher, Nestor Alonso Gutierrez, Morgan Walker, Katie Wright, Neelam Chavan, Neelakantan Thiyyadi Vasudenvan, Shihao Xu, Charles Rosadini, Hye-lin Ha, Valeria Busygina
Affiliation: Merck & Co., Inc., Rahway, NJ, USA

Therapeutic antibody discovery has traditionally relied on mouse immunization, an approach that has delivered numerous clinical successes. However, for certain targets, mouse campaigns may yield low-affinity binders with limited epitope diversity. To address these limitations, we implemented rabbit immunization as a complementary strategy to expand both affinity and epitope coverage for a high-priority pipeline program.

We compared outcomes from parallel mouse and rabbit immunization campaigns against the same target. The mouse campaign generated primarily low-affinity monoclonal antibodies (mAbs), whereas the rabbit campaign produced 63 high-affinity mAbs. Although rabbit-derived mAbs are widely used as research tools, their translation into therapeutics has remained limited, with only a small number of FDA-approved rabbit antibody-based molecules reported to date. A key challenge is that rabbit antibodies may contain developability liabilities, including paired and unpaired cysteines in the CDRs, which are less frequently observed in murine-derived antibodies. In addition, structural differences in rabbit mAbs require careful humanization to preserve potency and epitope engagement. We describe engineering and display workflows to optimize rabbit hits into program leads, including preemptive developability optimization, rabbit-specific humanization strategies, and rationally designed display libraries to enhance affinity.

These findings suggest that rabbit immunization can improve discovery outcomes for difficult targets by delivering high-affinity mAbs with broad epitope coverage. Combined with an integrated engineering and humanization toolkit, rabbit-derived binders can be advanced into therapeutic leads, supporting the use of rabbit immunization as a complementary strategy to strengthen therapeutic antibody discovery.

MGAT1 is an N-glycosyltransferase essential for the synthesis of N-glycans. Cell surface glycans serve as immune checkpoints, playing a key role in cancer immune evasion. Knockout (KO) of MGAT1 enhances immune recognition and promotes T-cell–mediated killing, with enzymatic activity being necessary for this phenotype. These findings position MGAT1’s catalytic function as an attractive target for cancer therapy. In this study, we report the discovery of novel MGAT1 binders and inhibitors with sub micromolar potency. Compounds were identified through two independent screening approaches: a UDP-GloTM-based high-throughput screen (HTS) measuring inhibition of MGAT1 enzymatic activity, and a DNA-encoded library (DEL) screen selecting for molecules that reproducibly bind to MGAT1. High-resolution crystal structures reveal detailed interactions between MGAT1 and the compounds, clearly identifying a binding site distinct from the active site. Surface plasmon resonance (SPR) competition assays further demonstrate that these inhibitors bind noncompetitively with respect to the endogenous product, UDP. Together, these results validate allosteric inhibition of MGAT1 as a novel and tractable strategy for impeding MGAT1 activity. Additionally, our findings lay the foundation for future structure-based optimization of MGAT1 inhibitors with potential application in cancer immunotherapy.

The discovery of the RAS Switch II pocket has enabled direct targeting of KRAS, a long standing oncogenic driver. Following the discovery of covalent inhibitors for KRASG12C driven cancers, the field has rapidly expanded to investigate the identification of effective inhibitors against multiple KRAS mutations. Biochemical screening of a 1.5K compound library yielded an initial hit compound which demonstrated pan-allele KRAS activity while maintaining selectivity over HRAS and NRAS. Structure-based optimization from the initial hit yielded a macrocyclic lead series with improved biochemical and cellular activity. Optimization of the lead series was guided by a computational approach, which combined machine learning and physics based methods for prediction of relevant ADME and activity endpoints. Ultimately, a lead compound was identified by which highly optimized interactions in the pocket yielded strong affinity against KRASG12D and KRASG12V, potent inhibition of phospho-ERK in various KRAS-mutant cell lines, and tumor regression in mouse xenograft models.

Published work from the Repare Therapeutics team will be presented, with an emphasis on the particular SBDD approaches with MOE that helped advance multiple projects into clinical trials for oncology. This will include the discovery of selective inhibitors of the kinases PkMYT1 (leading to RP-6306 / Lunresertib) and PLK4 (RP-1664). Projects against both the polymerase domain and the ATPase domain of Pol-theta discovered pM inhibitors of each, which produced the ATPase inhibitor RP-2119 and the highly potent clinical candidate RP-3467.

An update on the work from the Variational AI team's collaboration with the Agora Open Science Trust (a Canadian charity) and Enamine on PRMT6 will also be presented. PRMT6 is a target for the neurodegenerative disorder Spinal Bulbar Muscular Atrophy (SBMA). PRMT6 has a novel allosteric site that is unlike any homologs, and it represents a very low data target for which only one series of inhibitors was known. Generative AI outputs from Enki were docked against this allosteric site and analyzed with MOE, which led to the discovery of a new hit with an entirely novel scaffold in under three months.

Macrocyclic peptides (MCPs) are an attractive therapeutic modality because they can bind tightly and selectively against difficult-to-drug binding sites. While high-affinity MCPs are typically identified using phage and mRNA display, optimizing them for drug-like properties remains challenging. Part of this challenge is our lack of foundational understanding and de novo prediction of 3-dimensional MCP conformations, which heavily influence physicochemical properties. This talk outlines my current work addressing the MCP structure prediction problem, specifically engineering MCP-specific molecular descriptors to more accurately quantify conformations and, consequently, predict drug-like properties.

Peptide therapeutics encounter major developability challenges, including rapid proteolysis, limited oral bioavailability, aggregation/fibrillation, and immunogenicity, that impede translation to the clinic. Here we describe Novo Nordisk’s integrated strategy to accelerate peptide discovery and mitigate these liabilities by combining targeted, high-quality data generation, robust data and ML engineering, and peptide-specific foundation models. We summarize principles and practical lessons for building harmonized peptide datasets and the experimental controls required for reproducible model training. We detail modeling approaches that exploit sequence-based embeddings and transfer learning. We also describe metrics and guardrails for model adoption, including uncertainty quantification and out-of-distribution detection, to explicitly define applicability domains and increase user trust. Further gains are expected from integrating physics-informed descriptors, scaling peptide-focused pretraining, and coupling robust uncertainty estimates to enable more generalizable and reliable deployment within design workflows.

Subcutaneous (SC) delivery of therapeutic antibodies can offer multiple benefits to patients and healthcare providers, including convenience, time savings, and cost reduction. To improve the SC injection experience, drug developers may seek a low injection volume (1-2 mL), which for some antibody drugs necessitates a high concentration solution (≥100 mg/mL) to meet dosage requirements. Several molecular-level challenges hinder the development of high concentration antibody drug products, including high viscosity caused by reversible self-association (RSA). Here, we take an enhanced rational design approach to reduce RSA via protein engineering. Using hydrogen-deuterium exchange mass spectrometry (HDX-MS) and in silico modeling in Molecular Operating Environment (MOE), we identified potential self-interaction hotspots on the surface of an in-house IgG1 which has known viscosity issues at high concentration. With this information, several variants were rationally designed which vastly improved on the parent antibody’s viscosity issues and offer insight into the mechanism of self-association. This workflow and strategy will be discussed alongside results and implications for future molecules.