Maxwell Kleinsasser ,  Brayden J. Halverson ,  Edward Kraft ,  Sean Francis-Lyon ,  Sarah E. Hugo ,  Mackenzie R. Roman ,  Ben Miller ,  Andrew D. Blevins ,  Ian K. Quigley

arXiv - QuanBio - Biomolecules 

Abstract

The quality and consistency of training data remain critical bottlenecks for protein-ligand binding prediction. Public affinity datasets, aggregated from thousands of labs and assay formats, introduce biases that limit model generalization and complicate evaluation. DNA-encoded chemical libraries (DELs) offer a potential solution: unified experimental protocols generating massive binding datasets across diverse chemical and protein target space. We present Hermes, a lightweight transformer trained exclusively on DEL data from screens against hundreds of protein targets, representing one of the largest and most protein-diverse DEL training sets applied to protein-ligand interaction (PLI) modeling to date. Despite never seeing traditional affinity measurements during training, Hermes generalizes to held-out targets, novel chemical scaffolds, and external benchmarks derived from public binding data and high-throughput screens. Our results demonstrate that DEL data alone captures transferable protein-ligand interaction representations, while Hermes' minimal architecture enables inference speeds suitable for large-scale virtual screening.

Summary

The paper introduces Hermes, a lightweight transformer-based model trained exclusively on DNA-encoded library (DEL) screening data across 239 protein targets. Despite never using traditional affinity measurements (e.g., IC50, Kd), Hermes generalizes to unseen protein targets, novel chemical scaffolds, and external benchmarks derived from public binding data. The model demonstrates that DEL data alone captures transferable protein-ligand interaction representations, with inference speeds 500–700× faster than state-of-the-art structure-based models like Boltz-2, making it highly suitable for large-scale virtual screening.

Highlights

• Strong generalization: Achieves mean AUROC of 0.68 on the DEL Protein Split (unseen proteins) and 0.60 on Public Binders/Decoys (external benchmarks), with significantly better performance for kinase targets due to kinase-enriched training data.
• Speed advantage: Processes 28.2 samples/second/GPU on H200 hardware, far outpacing Boltz-2 (0.04 samples/second on H100), critical for cost-effective virtual screening.
• Limitations: Performance drops on the DEL Chemical Library Split (AUROC ~0.56), suggesting challenges in generalizing to entirely new chemical libraries. Data binarization (binary binding labels) and noise in DEL screening results constrain model expressivity.
• Practical impact: Highlights DEL datasets as a scalable, unified alternative to fragmented public affinity data (e.g., ChEMBL), with potential to accelerate drug discovery pipelines.

Conclusion

Hermes demonstrates that DEL-derived data alone can train generalizable protein-ligand binding prediction models without reliance on traditional affinity measurements. Its success underscores the value of large-scale, consistent DEL screening data for capturing transferable biological interactions. As DEL datasets continue to grow beyond public affinity resources, DEL-trained models like Hermes are poised to drive the next generation of computational drug discovery, particularly for targets underrepresented in existing public data. Future improvements could incorporate structural augmentation and continuous binding strength modeling to address current limitations.

Amanda W. Dombrowski，Florent Samain

ACS Medicinal Chemistry Letters

DOI: 10.1021/acsmedchemlett.6c00047

Abstract

Over the past 30 years, the field of DNA-encoded libraries (DELs) has become a mature and robust technology platform for the identification of ligands against relevant biological protein targets. Most of the major innovative pharmaceutical companies have integrated DEL platforms into their drug discovery workflows. Indeed, DELs have significantly impacted drug discovery efforts in the last 10 years with the identification of ligands that have progressed into clinical trials for various disease indications. One could assume that there are likely even more DEL-derived ligands that have reached the clinic, but candidate stories do not necessarily mention the hit generation methods utilized. (1−3) The fundamental concept of DELs emerged in a theoretical paper from Lerner and Brenner in 1992. (4) DELs are defined as collections of small molecules that are covalently attached to unique DNA tags, serving as amplifiable identification barcodes. Encoding procedures allow the generation and screening of combinatorial libraries of high diversity and unprecedented size. Preferential binding molecules identified by high throughput DNA sequencing of libraries after affinity capture procedures are typically resynthesized and tested to characterize their binding properties. (5) Success of DEL-based drug discovery screening campaigns is affected by various factors, including quality and diversity of the libraries, screening protocols, protein selection conditions, downstream validation and data analysis methods. Advances in the field have expanded DEL applications beyond traditional synthesis and screening procedures. Recently, the incorporation of artificial intelligence (AI) and machine learning (ML) techniques in DEL workflows has been accelerating, driving significant advancements and extending the potential of technology. (6,7) This ACS Medicinal Chemistry Letters Collection features recent success stories that highlight the value and impact of DEL technology in drug discovery. The following report, A Novel Small Molecule Allosteric Inhibitor of IL-17A from a DNA-Encoded Library, demonstrates the ability of DEL to identify small molecules that directly modulate the IL-17 pathway by inhibiting IL-17A with their cognate receptors, proving that small molecules can mimic the action of macromolecular biologics to disrupt high affinity protein–protein interactions. DEL macrocycle-like libraries have also proven effective for the recognition of larger protein surfaces (DNA-Encoded Macrocyclic Peptide Libraries Enable the Discovery of a Neutral MDM2–p53 Inhibitor). Beyond traditional inhibitor identification, DEL technology has become a pivotal approach for the development of targeted protein degradation (TPD) therapeutics. The report, Discovery of Small-Molecule Ligands for the E3 Ligase STUB1/CHIP from a DNA-Encoded Library Screen, from AstraZeneca scientists, shows that DEL can be effectively used to identify small molecule ligands for STIP1 homology and U-box containing protein 1 (STUB1), an E3 ligase that contains protein–protein interaction (PPI) sites, where previously only peptide binding ligands have been discovered. DEL hits can also serve as tools to provide structural basis for further hit-to-lead progression, thus accelerating medicinal chemistry lead optimization activities (Structural and Molecular Insight into the PWWP1 Domain of NSD2 from the Discovery of Novel Binders Via DNA-Encoded Library Screening; Optimization of a Novel DEL Hit That Binds in the Cbl-b SH2 Domain and Blocks Substrate Binding; Chemical Space Profiling of SARS-CoV-2 PLpro Using DNA-Encoded Focused Libraries). By leveraging the broad chemical diversity offered by DEL techniques, researchers have been able to identify robust covalent inhibitors that specifically interact with cysteine residues. This targeted approach helps improve both the selectivity and efficacy of potential therapeutic agents (Identification and Evaluation of Reversible Covalent Binders to Cys55 of Bfl-1 from a DNA-Encoded Chemical Library Screen). Recent literature indicates that DEL designs have become more strategic, influenced by progress in synthetic methodologies, expanded chemical diversity, and enhanced access to structural biology data. The perspective, Design of DNA Encoded Libraries for Medicinal Chemistry, provides a comprehensive analysis of DEL derived hits that enabled medicinal chemistry programs from publications between 2020 and 2025. The development of computational tools has offered new opportunities to rationalize DEL designs and DEL data set analysis. The integration of DEL technology with computational approaches, such as machine learning, continues to unleash the potential of the technology (Highly Selective Novel Heme Oxygenase-1 Hits Found by DNA-Encoded Library Machine Learning beyond the DEL Chemical Space; Evaluating the Diversity and Target Addressability of DELs using Scaffold Analysis and Machine Learning). This Collection features 10 publications, which highlight the rapid growth in many areas of the DEL field. We thank the authors for their work, which we hope will inform and inspire our readers to expand upon that work. This article references 7 other publications. This article has not yet been cited by other publications.
 

Summary

This editorial introduces a special collection of 10 publications highlighting the rapid evolution and maturation of DNA-encoded library (DEL) technology over the past 30 years. DELs, which consist of small molecules covalently attached to unique DNA barcodes for identification, have transitioned from a theoretical concept (first proposed by Lerner and Brenner in 1992) to a robust, industry-standard platform for drug discovery. Major pharmaceutical companies have now integrated DEL platforms into their workflows, with several DEL-derived ligands progressing into clinical trials over the last decade. The editorial emphasizes that recent advances—including AI/ML integration, novel library designs, and expanded applications beyond traditional inhibitor discovery—are significantly extending the potential of this technology.

Highlights

1. Diverse Therapeutic Applications

• IL-17A Inhibition: DEL technology successfully identified small molecule allosteric inhibitors that disrupt high-affinity protein-protein interactions (PPIs), demonstrating that small molecules can mimic macromolecular biologics.
• Macrocyclic Libraries: DEL macrocycle-like libraries enabled the discovery of neutral MDM2-p53 inhibitors, effective for recognizing larger protein surfaces.
• Targeted Protein Degradation (TPD): DEL screening identified small molecule ligands for STUB1/CHIP (an E3 ligase containing PPI sites), where previously only peptide binders were known.

2. Structural and Mechanistic Insights

• DEL hits provide structural foundations for hit-to-lead optimization, accelerating medicinal chemistry efforts. Examples include:
• Novel binders to the PWWP1 domain of NSD2
• Inhibitors targeting the Cbl-b SH2 domain
• Chemical space profiling of SARS-CoV-2 PLpro

3. Covalent Inhibitor Discovery

• Strategic DEL designs enable the identification of reversible covalent binders targeting specific cysteine residues (e.g., Cys55 of Bfl-1), improving selectivity and efficacy.

4. Integration of Artificial Intelligence and Machine Learning

• AI/ML techniques are increasingly incorporated into DEL workflows for:
• Rationalizing library designs
• Analyzing DEL datasets
• Discovering hits beyond traditional DEL chemical space (e.g., highly selective heme oxygenase-1 inhibitors)
• Scaffold analysis and target addressability prediction

5. Strategic Library Design Evolution

• Recent DEL designs have become more sophisticated, influenced by:
• Progress in synthetic methodologies
• Expanded chemical diversity through validated on-DNA chemical reactions
• Enhanced access to structural biology data

Conclusion

The editorial concludes that DEL technology has become a mature, robust, and indispensable platform in modern drug discovery. The featured collection demonstrates the technology's versatility across multiple applications—from traditional inhibitor identification to targeted protein degradation and covalent drug discovery. The integration of computational approaches, particularly machine learning, continues to unlock new potential and extend DEL capabilities beyond its original scope. The editors (Amanda W. Dombrowski and Florent Samain) express gratitude to the contributing authors and anticipate that these works will inform and inspire readers to further expand the boundaries of DEL technology in pharmaceutical research.

Vasanthanathan Poongavanam ,  S. Pauliina Turunen ,  Kristian Sandberg ,  Ulrika Yngve ,  Johan Wannberg

Drug Discovery Today 

DOI: 10.1016/j.drudis.2026.104629

Abstract

DNA-encoded libraries (DELs) combined with machine learning (ML) offer a powerful paradigm for hit identification. However, sequencing-derived enrichment data are inherently noisy and biased, often resulting in models that overfit to specific chemical libraries. In this review, we critically evaluate the capabilities and limitations of DEL-ML, illustrating key challenges using Aurora Kinase A (AURKA) DEL affinity selection data. We demonstrate that standard ML models often struggle to generalize to unseen chemical space because of the specific structural constraints of combinatorial libraries. Furthermore, we discuss the necessity of rigorous denoising strategies and evaluate approaches, such as domain adaptation, to mitigate these limitations, offering a roadmap for building robust models capable of exploring diverse chemical space.

Summary

This review critically examines the integration of machine learning (ML) with DNA-encoded library (DEL) technology for drug discovery. While DEL-ML offers a powerful paradigm for hit identification by generating massive binding datasets (10⁶–10¹² data points), the authors identify a critical "generalizability gap" that limits the practical utility of current models. Using Aurora Kinase A (AURKA) as a case study with OpenDEL 4.0 screening data (~1.5 million data points), the authors demonstrate that standard ML models achieve high accuracy on internal validation but frequently fail to generalize to structurally novel scaffolds due to domain shift—the substantial difference between DEL chemical space and known pharmacological compounds. The review provides methodological best practices for data preprocessing, denoising, and validation, while evaluating advanced strategies such as domain adaptation to improve model robustness. The authors argue that future DEL-ML development must move beyond simple accuracy maximization toward explicit handling of distribution shifts to transform DEL-ML from a retrospective analysis tool into a reliable engine for novel chemical discovery.

Highlights

1. The Generalizability Challenge in DEL-ML

• Models trained on DEL data often memorize library-specific building blocks rather than learning transferable structure-activity relationships
• The BELKA competition revealed that models perform well on test sets within the same chemical space but fail on structurally novel scaffolds
• Domain shift between DEL training data and external compound collections represents a fundamental barrier to practical application

2. Data Quality and Preprocessing Considerations

• DEL sequencing data contains unique noise profiles including matrix binding, DNA-tag interference, unequal synthesis yields, and "jackpot" effects
• Multiple denoising strategies are evaluated: fold-enrichment, Z-scores for ultra-large libraries, disynthon aggregation, and uncertainty-aware probabilistic loss functions
• Critical importance of subtracting background noise from control experiments (matrix/bead-only) to prevent false positives

3. Class Imbalance and Data Splitting Strategies

• DEL selections produce highly imbalanced datasets (10¹–10⁴ binders vs. up to ~10¹² nonbinders)
• Random splitting leads to overoptimistic metrics due to high structural similarity within DEL congeneric series
• Scaffold-based or library-based splitting provides more rigorous assessment of generalizability to novel chemotypes
• Undersampling nonbinders (e.g., 1:1 ratio) can boost external sensitivity from ~1% to 20–30%, though this may reflect bias exploitation rather than true generalization

4. Molecular Representation and Model Architectures

• Traditional fingerprints and physicochemical descriptors often fail to capture subtle variations in DEL compounds
• Graph neural networks (GNNs) and variational autoencoders (VAEs) show promise but require careful handling of linker/DNA-tag artifacts
• Compositional (disynthon) approaches reduce sparsity but risk losing "whole-molecule" structural fidelity
• Conformal prediction frameworks provide calibrated confidence intervals essential for prioritizing predictions in noisy DEL environments

5. Domain Adaptation as a Solution Strategy

• Covariate shift correction reduces divergence between source (DEL) and target (known binder) domains
• Using high-confidence predictions from diverse compound collections (e.g., Enamine REAL Diversity Set) as an intermediate domain improves generalization
• Domain adaptation reduced PCA centroid distance from 0.77 to 0.32 between DEL training data and known AURKA space
• Retraining with both predicted binders and nonbinders improved Matthews Correlation Coefficient (MCC) from 0.2 to 0.4 on external datasets while maintaining 20–39% sensitivity

6. AURKA Case Study Findings

• OpenDEL 4.0-derived binders tended to be larger, more lipophilic, and less polar compared to known AURKA inhibitors
• Despite overall domain shift, highly enriched DEL hits from sublibrary 27 shared conserved hinge-binding motifs with established inhibitors (e.g., VX-680)
• Mechanistic alignment between DEL hits and known binders confirms that domain shift, rather than fundamental binding mode differences, drives prediction failures

Conclusion

The integration of DELs with ML presents transformative opportunities for early drug discovery, but realizing this potential requires overcoming the critical generalizability gap. The primary challenge is not data volume but data nature: intrinsic structural biases and systematic false negatives (often linker-induced) cause models to memorize library-specific artifacts rather than learn transferable pharmacophore principles. High internal validation metrics frequently mask failures to extrapolate to novel, pharmacologically relevant scaffolds.

The authors advocate for a paradigm shift in DEL-ML development emphasizing:

Rigorous validation standards:
Moving beyond random splits to scaffold-based and out-of-distribution evaluation

Domain alignment strategies:
Explicit handling of distribution shifts through domain adaptation and transfer learning

Data diversity expansion:
Open-source DEL datasets spanning broader drug-like chemical space to reduce single-library bias

Integration of physics-based priors:
Incorporating docking constraints to reduce overfitting to synthetic artifacts

Uncertainty quantification:
Systematic use of conformal prediction and applicability domain assessment

By pivoting from simple accuracy maximization to robust domain alignment, DEL-ML can evolve from a retrospective analysis tool into a reliable engine for identifying novel chemical starting points. The establishment of standardized benchmarks and community resources will be essential to accelerate the development of generalizable predictive models capable of exploring the vast chemical space beyond individual DEL compositions.

Saan Voss , Amin Sagar , Arnaud Tiberghien , Richard J. L. Hughes , Liuhong Chen , Inmaculada Rioja , Mark Frigerio , Michael J. Skynner , David R. Spring

Journal of Peptide Science

DOI: 10.1002/psc.70071

Abstract

Bicyclic peptides are emerging as next generation therapeutics by combining the affinity and specificity of antibodies with the synthetic convenience of small molecules. Phage-encoded libraries of bicyclic peptides enable the discovery of high-affinity molecules against virtually any protein target. The generation of bicyclic peptides that advanced into clinical development involves the reaction of three cysteines in a peptide to a C₃-symmetric alkylating agent. In phage display, this chemical modification transforms a pool of conformationally flexible peptides into a library of structurally unique protein mimetics that are able to bind traditionally challenging protein surfaces like those with limited structural definition. In recent years, a new class of bicyclic peptides has emerged using a single atom-bismuth-in place of C₃-symmetric organic scaffolds, thus expanding into an unexplored chemical space at the intersection of inorganic chemistry and biology. This mini-review aims to reflect on the discovery, evolution and potential future applications of bismuth bicycle molecules.

In innovative drug discovery, the identification of compounds for "difficult-to-drug" (D2D) or "difficult-to-ligand" (D2L) targets remains a significant challenge. A recent study by the Roche Group, published in ACS Medicinal Chemistry Letters, systematically analyzed 21 hit-finding campaigns conducted between 2020 and mid‑2024 across its three R&D organizations: Pharma Research and Early Development (pRED), Genentech Research and Early Development (gRED), and the China Innovation Center of Roche (CICoR). The study evaluated the effectiveness of various screening strategies—including DNA‑encoded library (DEL), high‑throughput screening (HTS), and fragment‑based screening—with particular emphasis on the dual role of DEL technology as both a core screening tool and a ligandability assessment method. The findings provide the industry with a data‑informed reference framework to guide hit‑discovery strategy for challenging targets.

Key Finding 1: Correlation Between Target Classification and Project Success Rates
The study categorizes difficult-to-drug targets into two classes:

• Difficult-to-Drug (D2D) Targets: Targets with known or identifiable binding pockets, where it remains unclear whether binding will lead to the desired functional modulation (e.g., allosteric modulation, inhibition of protein-protein interactions).
• Difficult-to-Ligand (D2L) Targets: Targets lacking reliable structural information or assessed computationally as having no clearly druggable pocket (D-score < 0.7).

Figure 1. A diverse set of 21 difficult-to-drug targets and modes of action were analyzed.

Data analysis indicates that D2D targets show higher project advancement success rates. Among 7 D2D projects, 86% successfully yielded advanceable lead series, compared to 50% among 14 D2L projects. This suggests that target classification at the project outset can help predict technical success and guide resource allocation.

Key Finding 2: Performance Comparison of Screening Technologies
The study evaluated a total of 70 screening experiments across six common technologies. Figure 2 illustrates the frequency of each method's use across actual projects.

Figure 2. Multiple diverse screening methods were employed for a total of 70 screens on 21 projects. The graphic depicts the number of individual screens, color-coded by method.

In terms of breadth of application (Figure 3), DNA-encoded library technology was used in all 21 projects, high-throughput screening in 76% of projects, fragment screening in approximately half, while other methods were used less frequently.

Figure 3. Different hit finding methods were used with different frequency across the research units. Depicted is the percentage of projects for which a specific screen was used, e.g. a fragment screen was used on almost 50% of all projects.

Significant differences were observed in the performance of each method across two critical stages: "producing validated hits" and "conversion to lead series" (Figure 4 and table below):

Figure 4. Screening methods varied in their ability to produce validated hits and lead series. Plotted is the percentage of screens for each method that delivered validated hits (left columns) and lead series (right columns).

Data analysis reveals:

• DEL is widely applicable and demonstrates predictive value. This technology was used in all projects, achieved the highest success rate in yielding validated hits, and showed high positive predictive value (92%). This indicates that DEL can not only be used for hit identification but also serves as a reference tool for assessing target ligandability and guiding subsequent experimental strategies.
• High-throughput, covalent, and peptide screening show consistent performance in applicable contexts. These three methods each achieved a 50% success rate in advancing to lead series in applicable projects, demonstrating good reproducibility. They respectively offer advantages in functional activity screening, high-affinity binding, and large interface coverage.
• Fragment and virtual screening face challenges in the optimization phase. Although both performed reasonably well in initial hit identification, their success rates in optimizing hits into drug-like, advanceable lead series were low (<10%), suggesting they are more suitable for early exploratory phases.

Key Finding 3: Strategic Choices for Integrated Screening
The study identifies two common strategic approaches:

• "Broad Platform" Approach: Utilizes multiple screening methodologies in parallel to maximize exploration of diverse chemical spaces.
• "Selective Focused" Approach: Concentrates resources on a limited, prioritized set of screening methods tailored to the target. (e.g., DEL+HTS).

Data suggest that using more screening methods is not necessarily better. If three or more consecutive different screens on the same target fail to yield advanceable compounds, the likelihood of subsequent success may decrease significantly. Therefore, dynamically adjusting strategy based on target characteristics and early screening results can enhance R&D efficiency.

Conclusion
Roche's study not only provides strong evidence for the effectiveness of DEL technology but, more importantly, establishes an actionable decision-making framework. In an environment of rising drug development costs and increasing target difficulty, "informed integrated hit discovery" represents a strategic shift from experience-driven to data-driven thinking.

Looking ahead, with advancements in DEL technology itself (e.g., integrating machine learning for hit expansion), wider adoption of AI-based structure prediction tools, and the emergence of novel screening paradigms, we can anticipate the development of more intelligent and efficient integrated discovery platforms. These advancements hold the potential to transform more previously "undruggable" targets into revolutionary therapies for patients.

1. Gampe, C. M.; Worsdorfer, B.; Zou, G.; Ricci, A. Analyses of Recent Hit-Finding Campaigns for Difficult Targets Provides Guidance for Informed Integrated Hit Discovery. ACS Med. Chem. Lett. 2026. https://doi.org/10.1021/acsmedchemlett.5c00676

Christian M. Gampe , Bigna Wörsdörfer , Ge Zou , Antonio Ricci

ACS Medicinal Chemistry Letters

DOI: 10.1021/acsmedchemlett.5c00676

Abstract

Despite advancements in hit-finding technologies, many drug targets are considered difficult-to-drug (D2D) or difficult-to-ligand (D2L). Here, we present an analysis of 21 hit-finding campaigns across three research organizations within the Roche group, focusing on D2D and D2L targets. DNA-encoded library technology (DELT) was the most successful method in providing validated hits and lead series. High-throughput, covalent, and peptide screens also yielded progressable chemical matter in a substantial number of cases. In contrast, fragment and virtual screens, while effective in generating validated hits, demonstrated lower success rates. Stratifying targets into D2D and D2L categories provided a useful framework for estimating the likelihood of project success and informing additional screening strategies, with D2D targets showing higher rates of chemical enablement. Our findings indicate DELT as a valuable experimental tool for assessing ligandability and highlight the importance of informed integrated hit discovery by tailoring hit-finding strategies to target characteristics.