Daniel Saliba,Eiman A Osman,Abdelrahman Elmanzalawy,Christopher Saab,Son Bui,Serhii Hirka,Shaun Anderson,Violeta Toader,Michael D Dore,Felix J Rizzuto,Donatien de Rochambeau,Maureen McKeague,Hanadi F Sleiman

Nature Chemistry

DOI: 10.1038/s41557-026-02099-5

Abstract

Aptamers are a versatile alternative to antibodies as they are smaller, easier to synthesize and less immunogenic. However, while antibodies are composed of 20 chemically diverse amino acids and are established therapeutics, aptamers are composed of only 4 similar nucleobases, thereby limiting their therapeutic potential. Aptamer chemical modifications are restricted to maintain compatibility with enzymatic selection. Here we introduce aptamer-like encoded oligomers (alenomers), highly chemically modified aptamers that are read and sequenced using a DNA code branching from and corresponding to the target-binding oligomer. We build ~300,000-member DNA-encoded libraries using an automated DNA synthesizer and split-and-pool methods, and screen them for protein binding via next-generation sequencing. In contrast to aptamers, alenomers are not restricted by the need for conservative enzyme-compatible modifications. They can thus explore an almost limitless chemical space, enabling the discovery of highly stable, high-affinity protein-binding aptamers, while offering structural insights into their interactions with target molecules.

Tonglin Yu , Jian Ma , Xiaodi Su , Jianhong Tang , Yujian He , Xiangyu Chen , Li Wu

Advanced Synthesis & Catalysis

DOI: 10.1002/adsc.70499

Abstract

DNA‐encoded libraries (DELs) are a powerful technology increasingly used in drug discovery for screening lead molecules. The key to success is dependent on the chemical space covered by DELs. Enzymes can catalyze complex chemical reactions under mild conditions, making them highly attractive for constructing structurally diverse DELs. However, traditional enzymatic transformations have been considered unsuitable for DEL construction due to their narrow substrate scope, leading to slow progress in this field. Here, we challenge this conventional perception by introducing the catalytic promiscuity of hydrolases into three‐component reactions on DNA. We successfully performed three enzyme‐promoted coupling reactions directly on DNA: the Aza‐Diels–Alder reaction, the Biginelli reaction, and the Mannich reaction.These reactions bring N‐heterocyclic bridged rings, pyrimidines, and α‐branched amine‐based nitrogen‐containing pharmacophores to DELs. Importantly, the entire coupling process on DNA tags does not require the use of harmful metals or stoichiometric organic catalysts, nor does it involve additional immobilization of the DNA strand. Using green and inexpensive hydrolases, the reactions can proceed directly in aqueous mixed solutions. Due to the mildness of enzyme‐catalyzed reaction conditions, all three reactions are highly compatible with DNA tags.

Miklos Feher , Rebecca J. Swett , Ryan T. Walsh , Erin Davis , Christopher I. Williams

ACS Medicinal Chemistry Letters

DOI: 10.1021/acsmedchemlett.6c00141

Abstract

DNA-encoded library (DEL) screening enables identification of small-molecule binders from libraries containing billions of compounds, yet much of the resulting structure–activity relationship (SAR) information remains underutilized. Here, we describe DEL2PH4, an automated ligand-based workflow that converts DEL screening data into three-dimensional pharmacophore models by integrating statistically enriched compounds with structurally related unenriched analogs, which serve as negative examples during model construction. The resulting pharmacophores capture consensus interaction features across DEL families and enable the extraction of actionable 3D SAR information from primary DEL screening data, independent of resynthesis or activity measurements. Application to a MerTK kinase DEL screen demonstrates strong enrichment of positives over decoy molecules in retrospective benchmarking, recovery of known experimentally validated actives from external data sets, and consistency with experimentally determined X-ray binding modes. DEL2PH4 provides a general strategy for translating DEL screening outputs into interpretable 3D models that support virtual screening, scaffold hopping, and medicinal chemistry optimization.

Trevor A. Zandi, Michael J. Romanowski, Jessica S. Viscomi, Karl Gunderson, Zher Yin Tan, Bingqi Tong, Simone Bonazzi, Frédéric J. Zécri, Stuart L. Schreiber, Gregory A. Michaud

Nature Communications

DOI: 10.1038/s41467-026-71512-x

Abstract

Molecular glues are small molecules that engage their target and presenter proteins cooperatively. FKBP12 molecular glues (FK506 and rapamycin) were discovered several decades ago and have been used clinically, but our understanding of the breadth of FKBP12 molecular glues and targets has yet to be fully revealed. To expand the target classes of FKBP12 molecular glues, we construct and screen a multi-million-member non-macrocyclic FKBP12-ligand DNA-encoded library using 25 structurally distinct proteins. Synthesis and validation of select hits in biophysical and cell-based assays confirm FKBP12-dependent molecular-glue recruitment to bromodomain-containing protein 9 (BRD9) and quinoid dihydropteridine reductase (QDPR). One glue shows no measurable binding to QDPR alone but has appreciable binding in the presence of FKBP12 using either purified proteins or intact cells. The sites of recruitment are characterized with mutational analysis, competition-based methods and X-ray crystallography. The results of this study confirm that FKBP12-binding DELs can yield molecular glues generating highly selective FKBP12-target protein interactions.

Overview

Cyclic peptidomimetics (CPM) have attracted growing attention in drug discovery because they combine the developability of small molecules with the target-recognition capability of larger biomolecules. Yet their complex macrocyclic topologies, noncanonical amino acids, and diverse cross-linking chemistries continue to challenge conventional AI-based molecular modeling methods.

Recently, the Computational Chemistry team at HitGen introduced CycWeave, a token-free dual-view coarse-grained graph neural framework designed for complex modular molecular systems. By adopting a representation strategy that better matches the modular nature of CPMs and DNA-encoded library (DEL) compounds, CycWeave demonstrated robust and competitive performance in both CPM membrane permeability prediction and large-scale DEL enrichment modeling. (Preprint available on ChemRxiv, https://chemrxiv.org/doi/full/10.26434/chemrxiv.15001512/v1)

01 Challenges in Current Computational Modeling

In AI-driven drug discovery, computational modeling of CPMs and structurally complex DEL compounds faces two major limitations:

1. Atom-level graphs often fail to capture global topology

Conventional graph neural networks (GNNs) primarily focus on local atoms and bonds, but often struggle to effectively represent the higher-order topological organization characteristic of cyclic peptide-like systems, such as scaffold architecture, branch placement, and connection patterns.

2. Vocabulary-dependent token models have limited generalization

Many existing peptide or fragment-based modeling methods rely on predefined vocabularies or tokenization schemes. In realistic CPM-oriented DEL settings, however, noncanonical monomers and open-ended chemical modifications are common. As a result, such methods can suffer from out-of-vocabulary limitations and reduced generalizability in open chemical space.

Figure 1. Summary of existing molecular modeling strategies for CPM

02 Core Design Logic of CycWeave

To address these challenges, CycWeave introduces a new representation framework specifically designed for structurally complex and modular molecules.

1. Dual-view graph architecture

CycWeave represents each molecule simultaneously as an atom-level graph and a fragment-level coarse-grained graph. The atom-level view captures local chemical environments, while the coarse-grained view explicitly preserves modular structure by decomposing molecules into scaffold, branch, and connection-level components, including key chemical relations such as amide linkages, ring connection sites, and disulfide bonds. The two views are coupled and fused within a unified neural architecture, enabling coordinated modeling of both local detail and global topology.

2. Token-free continuous fragment embeddings

A central innovation of CycWeave is its token-free design. Instead of mapping fragments into discrete symbolic tokens, the framework uses continuous ECFP-based fragment embeddings to initialize coarse-grained nodes. This avoids dependence on a fixed vocabulary and enables the model to generalize more naturally to novel noncanonical monomers and open-ended chemical modifications.

3. Support for self-supervised pretraining

CycWeave also supports a self-supervised pretraining–fine-tuning paradigm. Through a masked fragment recovery task, the model learns to reconstruct original continuous fragment fingerprints from surrounding structural context. This allows CycWeave to learn transferable structural priors from large unlabeled DEL-related CPM chemical spaces and improves its applicability to downstream tasks with limited labeled data.

Figure 2. Schematic overview of the token-free coarse-grained dual-view framework of CycWeave.

03 Application Validation: Developability Assessment and DEL Screening Modeling

The research team systematically evaluated CycWeave in two practically important application scenarios.

1. CPM membrane permeability prediction

Membrane permeability is jointly influenced by local physicochemical features and higher-order structural organization. On public benchmark datasets including PAMPA, Caco-2, MDCK, and RRCK, CycWeave achieved the strongest overall performance on the major benchmarks after pretraining and fine-tuning. Notably, it reached an R² of 0.728 in Caco-2 and 0.701 on the aggregated dataset, outperforming representative intermediate-granularity baselines such as PepLand and PeptideCLM. These results support the value of token-free dual-view representation for developability-related property prediction.

2. DEL enrichment modeling against TfR1

The team further applied CycWeave to DEL enrichment modeling against transferrin receptor 1 (TfR1), a biologically and translationally relevant target in drug delivery research. Because DEL enrichment signals are count-derived and typically overdispersed, the model used a negative binomial negative log-likelihood loss rather than a simple mean squared error objective. Under 10-fold scaffold-split evaluation, CycWeave outperformed both the general-purpose graph learning baseline Chemprop and the classical ECFP-MLP baseline. It achieved R² = 0.596, AUC-ROC = 0.962, and AP = 0.764, demonstrating strong regression fit as well as effective prioritization of enriched compounds under class imbalance.

In addition, latent-space visualization using t-SNE showed that enriched DEL compounds were organized into multiple separated yet internally compact clusters, suggesting that CycWeave not only improves predictive performance but may also help reveal distinct latent chemotypes or scaffold series for downstream hit triaging and series analysis.

04 Summary and Outlook

The results of CycWeave suggest that, for complex modular molecular systems such as cyclic peptidomimetics and DEL compounds, chemically meaningful coarse-grained decomposition combined with a token-free open representation can substantially improve computational modeling performance.

As a unified molecular representation backbone, CycWeave is expected to support not only CPM property prediction, but also a broader range of AI-for-chemistry applications, including DEL activity modeling, selectivity analysis, pharmacokinetic property prediction, and multi-objective molecular optimization.

Marissa D Dolorfino, Daniel Santos Perez, Yao Fu, Shu-Hang Lin, Sean McCarty, Matthew James O'Meara, Terra Sztain

bioRxiv - Biophysics

DOI: 10.64898/2026.04.18.719394

Abstract

DNA-encoded libraries (DELs) enable ultra-large screening of billions of molecules simultaneously. However, various limitations of DELs have prompted interest in training machine learning (ML) models on these large datasets to extrapolate predictions to non-DEL compounds. A recent NeurIPS competition revealed that even top performing ML models trained on DEL data failed at generalizing to out-of-distribution (OOD) chemical space. We investigated whether integrating structural modeling could bridge this generalization gap. We systematically assessed state-of-the-art ML, docking, and co-folding methods with three biologically diverse protein targets screened against libraries containing multiple DEL synthesis formats, and show that while ML excels in-distribution, the optimal approach for OOD hit discrimination performance is both target and ligand dependent. We conclude that, regardless of performance reported in aggregated benchmarks, rigorous, system-dependent pilot testing is critical for reliable virtual screening predictions. We provide these workflows and analysis tools in an open-source package: DEL-iver.