Marissa Dolorfino, Daniel Santos Perez, Yao Fu, Shu-Hang Lin, Sean McCarty, Matthew J. O’Meara, Terra Sztain

Journal of Chemical Information and Modeling

DOI: 10.1021/acs.jcim.6c01611

Abstract

Predicting protein–ligand binding is a central challenge in computational drug discovery, and while machine learning (ML) and cofolding methods have advanced rapidly, their ability to generalize beyond training or parametrization regimes remains insufficiently understood. DNA-encoded libraries (DELs) enable ultralarge screening of billions of molecules simultaneously, providing a useful testbed for evaluating these approaches at scale. 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 cofolding methods, including Schrödinger Glide, Rosetta GALigandDock, and Boltz-2 with three biologically diverse protein targets screened against libraries containing multiple DEL synthesis formats. While ML excels in-distribution, OOD hit discrimination is dependent on both the target and ligand context, with no single method consistently dominating. These findings demonstrate that benchmark performance alone is insufficient to predict OOD performance, highlighting the need for system-dependent evaluation of binding prediction methods. We provide an open-source package for assessing protein–ligand prediction methods and analyzing high-throughput screening data: DEL-iver.

Artur Menzeleev , Sathya Chitturi , Geraint Davies , Tony Schroeder , Alpha Lee

Digital Discovery

DOI: 10.1039/d6dd00162a

Abstract

DNA-encoded library (DEL) screening enables the rapid generation of billion-scale structure-activity datasets and has become a powerful approach for identifying chemical starting points against challenging biological targets. Recent work has leveraged DEL data to train machine learning (ML) models that screen external databases of purchasable compounds, expanding accessible chemical space and reducing the need for off-DNA synthesis. However, DEL datasets contain substantial noise and sampling biases arising in part from synthesis inefficiencies, uneven sampling, and sequencing variability. These factors can obscure true structure-activity relationships and complicate the evaluation and benchmarking of ML models trained on these data. Here, we present DEL Simulator, an open-source framework for systematically studying how DEL experimental parameters influence data quality and downstream ML model performance. The simulator models the key stages of DEL workflows, including library synthesis, selection, and sequencing, and provides a controlled environment for analyzing how factors such as library composition, reaction efficiency, and sequencing depth affect enrichment measurements and predictive modeling. Using this framework, we evaluate preprocessing strategies such as disynthon aggregation and identify regimes in which they improve predictive accuracy. We also observe counterintuitive trends, including reduced model performance with increasing library size under certain conditions. By connecting experimental design choices with data fidelity and model behavior, DEL Simulator provides a principled framework for analyzing DEL datasets and improving ML-driven virtual screening workflows.

Haipeng Yu, Xiaoran Wang, Xinyu Wang, Yizhen Yin

Chemistry - A European Journal

DOI:10.1002/chem.71680

Abstract

Covalent drugs have attracted increasing attention in modern drug discovery because of their prolonged target engagement, enhanced pharmacological efficacy, and potential to address traditionally challenging protein targets. The rapid development of electrophilic warhead chemistry, chemical biology methods, and high-throughput screening technologies has shifted covalent drug discovery from serendipitous identification to target-oriented and rational design. In recent years, phage display, mRNA display, and DNA-encoded library technologies have enabled the efficient de novo discovery of covalent peptides and small molecules, expanding the accessible chemical space for targeting cysteine, lysine, tyrosine, serine, and other reactive residues. Meanwhile, artificial intelligence has further accelerated covalent drug research by supporting covalent database construction, binding-site prediction, virtual screening, molecular generation, and structure-based ligand optimization. This article reviews recent advances in high-throughput strategies for targeted covalent drug discovery, with a focus on phage display, mRNA display, DNA-encoded libraries, and artificial intelligence (AI)-assisted approaches. We further discuss current challenges and future opportunities in the field and provide our perspectives on emerging directions for next-generation covalent ligand discovery.

Danial Yazdan, A. Michael Downey, Ana Gimeno, Caishun Li, Edward N. Schmidt, Jeffrey Y. K. Wong, Caleb Loo, Jaesoo Jung, Ryan Qiu, Ewa Lis, Lily Lindmeier, June Ereño-Orbea, Jesús Jiménez-Barbero, Matthew S. Macauley, Ratmir Derda

Journal of the American Chemical Society

DOI: 10.1021/jacs.6c14068

Abstract

Glycan-binding proteins (GBPs) are among the most difficult drug targets, limiting clinical progress against therapeutically important GBPs. We employed bicyclic genetically encoded libraries (BiGELs), produced by the chemical modification of phage-displayed peptide libraries with twofold symmetric linchpins, to discover inhibitors of therapeutically relevant Siglec-7:GD3 interactions. Next-generation sequencing (NGS) analysis of BiGEL panning against Siglec-7 yielded 815 candidates, of which 23 hits yielded K_D = 1–100 μM, as determined by surface plasmon resonance (SPR). Competitive enzyme-linked immunosorbent assays (ELISA) identified leads that disrupted the Siglec-7:GD3 interaction with IC₅₀ = 3-300 μM. Machine learning models trained on NGS datasets identified additional inhibitors with equivalent potencies. Alanine scans of 8c (SWCRPATVNC, IC₅₀ = 3.8 μM) and 12c (SFCHYPTHVC, IC₅₀ = 11 μM) identified residues crucial for activity. Ring reshaping studies of 8c highlighted the critical role of bicyclic topology, yielding analog 46e (SAAAAAWCRPATVNC, IC₅₀ = 9.5 μM), which was further evolved into 67e (STVTQHWCRPATVNC). The multivalent display of lead bicycles alongside ∼100 glycans in a Liquid Glycan Array (LiGA) enabled the comparison of binding to Siglec-7-expressing cells. LiGA assays confirmed the binding of the bicycles to Siglec-7 but revealed considerable nonspecific interactions with receptor-negative cells. Saturation transfer difference nuclear magnetic resonance (STD-NMR) spectroscopy revealed that 46e binds to Siglec-7 at a site distinct from the V-Ig domain, suggesting inhibition through an allosteric site. Together, these results demonstrate that BiGEL enables the discovery of bicyclic peptides for undruggable Siglec targets, but highlights future challenges in molecular discoveries that aim to identify small, noncarbohydrate inhibitors of GBPs.

Tianxiong Mi, Lijun Fan, Amber Hackler, Diego B. Diaz, Chang Qi, Errol L. G. Samuel, Qi Gao, Tao Meng, Xingjian Xu, Erwin G. Abucayon, Edward N. DiNunzio, Marcelo J. Murai, Natalya Pissarnitski, Jordan De Jesus Silva, Christopher Sondey, Jack D. Scott

Journal of the American Chemical Society

DOI: 10.1021/jacs.6c07049

Abstract

Rapid, aqueous macrocyclization strategies that proceed in high conversion are valuable for ultralarge macrocyclic peptide (MP) library synthesis via display and DNA-encoded library (DEL) technology. Here we report an on-DNA macrocyclization based on sulfur(VI) fluoride exchange (SuFEx) that unites above features. This approach embeds a phenol and an aryl sulfonyl fluoride within a DNA-tagged peptide to accelerate SuFEx and trigger intramolecular cyclization immediately upon dissolution in basic aqueous buffer. MPs ranging from 15 to 52 membered rings bearing diverse amino acids were synthesized efficiently. The on-DNA conditions readily translate off DNA to furnish sulfonate and sulfonamide linked MPs. NMR analyses showed SuFEx-derived biaryl linkers act as conformational tuner: sequential changes in aryl substitution and linker length shift backbone conformations from extended strands to rigid turns. Leveraging this chemistry, we designed and synthesized ultralarge MP libraries via DEL technology. DEL screening followed by off-DNA hit validation identified a potent, de novo MP inhibitor of receptor-interacting serine/threonine kinase 1 (RIPK1). Collectively, these findings establish SuFEx cyclization as a robust, DEL-compatible strategy for programmable macrocycle design and drug discovery.

Ana L. Aguirre, Amanda W. Dombrowski, Noah P. Tu, Shubhendu S. Karandikar, Ying Wang

The Journal of Organic Chemistry 

DOI:  10.1021/acs.joc.6c00134

Abstract

In this work, we present the creation and implementation of a ChemBead-enabled, on-DNA high-throughput experimentation (HTE) platform to identify more general and mild reaction conditions for challenging C–N couplings. The platform was developed by adapting our small-molecule system to address the inherent challenges of on-DNA chemistry. The results generated by the platform were compared to the standard operating procedure for DNA-encoded library (DEL) production to ensure the translation of any promising conditions. The platform was used to discover a condition that has not been previously reported for on-DNA C–N coupling. This condition is best suited for secondary amines and select primary amines on short DNA substrates; primary aliphatic amines on elongated DNA-tagged aryl halide conjugates remain a challenge. This newly identified condition can be applied to DNA-encoded library constructions