Gatekeeper mutations activate FGF receptor tyrosine kinases by destabilizing the autoinhibited state.

Many types of human cancers are being treated with small molecule ATP-competitive inhibitors targeting the kinase domain of receptor tyrosine kinases. Despite initial successful remission, long-term treatment almost inevitably leads to the emergence of drug resistance mutations at the gatekeeper residue hindering the access of the inhibitor to a hydrophobic pocket at the back of the ATP-binding cleft. In addition to reducing drug efficacy, gatekeeper mutations elevate the intrinsic activity of the tyrosine kinase domain leading to more aggressive types of cancer. However, the mechanism of gain-of-function by gatekeeper mutations is poorly understood. Here, we characterized fibroblast growth factor receptor (FGFR) tyrosine kinases harboring two distinct gatekeeper mutations using kinase activity assays, NMR spectroscopy, bioinformatic analyses, and MD simulations. Our data show that gatekeeper mutations destabilize the autoinhibitory conformation of the DFG motif locally and of the kinase globally, suggesting they impart gain-of-function by facilitating the kinase's ability to populate the active state.

Deciphering the Allosteric Activation Mechanism of SIRT6 Using Molecular Dynamics Simulations.

As a member of the histone deacetylase protein family, the NAD+-dependent SIRT6 plays an important role in maintaining genomic stability and regulating cell metabolism. Interestingly, SIRT6 has been found to have a preference for hydrolyzing long-chain fatty acyls relative to deacetylation, and it can be activated by fatty acids. However, the mechanisms by which SIRT6 recognizes different substrates and can be activated by small molecular activators are still not well understood. In this study, we carried out extensive molecular dynamic simulations to shed light on these mechanisms. Our results revealed that the binding of the myristoylated substrate stabilizes the catalytically favorable conformation of NAD+, while the binding of the acetyl-lysine substrate leads to a loose binding of NAD+ in SIRT6. Based on these observations, we proposed a reasonable allosteric binding mode for myristic acid, which can enhance the catalytic activity of SIRT6 by stabilizing the binding of NAD+ with His131 as well as the acetylated substrate. Furthermore, our molecular dynamics simulations demonstrated that synthetic SIRT6 activators, such as UBCS039, MDL-801, and 12q, block the flipping of ribose in NAD+ and therefore can stabilize substrate-NAD+-His131 interactions in a manner similar to fatty acids. In summary, our newly proposed activation mechanism of SIRT6 highlights the importance of protein-substrate interactions, which would facilitate the rational design of new SIRT6 activators.

Peptide Tethering: Pocket-Directed Fragment Screening for Peptidomimetic Inhibitor Discovery.

Constrained peptides have proven to be a rich source of ligands for protein surfaces, but are often limited in their binding potency. Deployment of nonnatural side chains that access unoccupied crevices on the receptor surface offers a potential avenue to enhance binding affinity. We recently described a computational approach to create topographic maps of protein surfaces to guide the design of nonnatural side chains [J. Am. Chem. Soc. 2017, 139, 15560]. The computational method, AlphaSpace, was used to predict peptide ligands for the KIX domain of the p300/CBP coactivator. KIX has been the subject of numerous ligand discovery strategies, but potent inhibitors of its interaction with transcription factors remain difficult to access. Although the computational approach provided a significant enhancement in the binding affinity of the peptide, fine-tuning of nonnatural side chains required an experimental screening method. Here we implement a peptide-tethering strategy to screen fragments as nonnatural side chains on conformationally defined peptides. The combined computational-experimental approach offers a general framework for optimizing peptidomimetics as inhibitors of protein-protein interactions.

Molecular basis for receptor tyrosine kinase A-loop tyrosine transphosphorylation.

A long-standing mystery shrouds the mechanism by which catalytically repressed receptor tyrosine kinase domains accomplish transphosphorylation of activation loop (A-loop) tyrosines. Here we show that this reaction proceeds via an asymmetric complex that is thermodynamically disadvantaged because of an electrostatic repulsion between enzyme and substrate kinases. Under physiological conditions, the energetic gain resulting from ligand-induced dimerization of extracellular domains overcomes this opposing clash, stabilizing the A-loop-transphosphorylating dimer. A unique pathogenic fibroblast growth factor receptor gain-of-function mutation promotes formation of the complex responsible for phosphorylation of A-loop tyrosines by eliminating this repulsive force. We show that asymmetric complex formation induces a more phosphorylatable A-loop conformation in the substrate kinase, which in turn promotes the active state of the enzyme kinase. This explains how quantitative differences in the stability of ligand-induced extracellular dimerization promotes formation of the intracellular A-loop-transphosphorylating asymmetric complex to varying extents, thereby modulating intracellular kinase activity and signaling intensity.

Unified Deep Learning Model for Multitask Reaction Predictions with Explanation.

There is significant interest and importance to develop robust machine learning models to assist organic chemistry synthesis. Typically, task-specific machine learning models for distinct reaction prediction tasks have been developed. In this work, we develop a unified deep learning model, T5Chem, for a variety of chemical reaction predictions tasks by adapting the "Text-to-Text Transfer Transformer" (T5) framework in natural language processing (NLP). On the basis of self-supervised pretraining with PubChem molecules, the T5Chem model can achieve state-of-the-art performances for four distinct types of task-specific reaction prediction tasks using four different open-source data sets, including reaction type classification on USPTO_TPL, forward reaction prediction on USPTO_MIT, single-step retrosynthesis on USPTO_50k, and reaction yield prediction on high-throughput C-N coupling reactions. Meanwhile, we introduced a new unified multitask reaction prediction data set USPTO_500_MT, which can be used to train and test five different types of reaction tasks, including the above four as well as a new reagent suggestion task. Our results showed that models trained with multiple tasks are more robust and can benefit from mutual learning on related tasks. Furthermore, we demonstrated the use of SHAP (SHapley Additive exPlanations) to explain T5Chem predictions at the functional group level, which provides a way to demystify sequence-based deep learning models in chemistry. T5Chem is accessible through https://yzhang.hpc.nyu.edu/T5Chem.

Delta Machine Learning to Improve Scoring-Ranking-Screening Performances of Protein-Ligand Scoring Functions.

Protein-ligand scoring functions are widely used in structure-based drug design for fast evaluation of protein-ligand interactions, and it is of strong interest to develop scoring functions with machine-learning approaches. In this work, by expanding the training set, developing physically meaningful features, employing our recently developed linear empirical scoring function Lin_F9 (Yang, C. J. Chem. Inf. Model. 2021, 61, 4630-4644) as the baseline, and applying extreme gradient boosting (XGBoost) with Δ-machine learning, we have further improved the robustness and applicability of machine-learning scoring functions. Besides the top performances for scoring-ranking-screening power tests of the CASF-2016 benchmark, the new scoring function ΔLin_F9XGB also achieves superior scoring and ranking performances in different structure types that mimic real docking applications. The scoring powers of ΔLin_F9XGB for locally optimized poses, flexible redocked poses, and ensemble docked poses of the CASF-2016 core set achieve Pearson's correlation coefficient (R) values of 0.853, 0.839, and 0.813, respectively. In addition, the large-scale docking-based virtual screening test on the LIT-PCBA data set demonstrates the reliability and robustness of ΔLin_F9XGB in virtual screening application. The ΔLin_F9XGB scoring function and its code are freely available on the web at (https://yzhang.hpc.nyu.edu/Delta_LinF9_XGB).

CovBinderInPDB: A Structure-Based Covalent Binder Database.

Covalent inhibition has emerged as a promising orthogonal approach for drug discovery, despite the significant challenge in achieving target specificity. To facilitate the structure-based rational design of target-specific covalent modulators, we developed an integrated computational protocol to curate covalent binders from the RCSB Protein Data Bank (PDB). Starting from the macromolecular crystallographic information files (mmCIF) in the PDB archive, covalent bond records, which indicate the side chain modification of amino acid residue by a covalent binder, were collected and cleaned. Then, residue-binder adducts, which are products of chemical reactions between targeted residues and covalent binders, were recovered with the help of the Chemical Component Dictionary in PDB. Finally, several strategies were employed to curate the pre-reaction forms of covalent binders from the adducts. Our curated CovBinderInPDB database contains 7375 covalent modifications in which 2189 unique covalent binders target nine types of amino acid residues (Cys, Lys, Ser, Asp, Glu, His, Met, Thr, and Tyr) from 3555 complex structures of 1170 unique protein chains. This database would set a solid foundation for developing and benchmarking computational strategies for covalent modulator design and is freely accessible at https://yzhang.hpc.nyu.edu/CovBinderInPDB.

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

Graph neural network (GNN)-based deep learning (DL) models have been widely implemented to predict the experimental aqueous solvation free energy, while its prediction accuracy has reached a plateau partly due to the scarcity of available experimental data. In order to tackle this challenge, we first build a large and diverse calculated data set Frag20-Aqsol-100K of aqueous solvation free energy with reasonable computational cost and accuracy via electronic structure calculations with continuum solvent models. Then, we develop a novel 3D atomic feature-based GNN model with the principal neighborhood aggregation (PNAConv) and demonstrate that 3D atomic features obtained from molecular mechanics-optimized geometries can significantly improve the learning power of GNN models in predicting calculated solvation free energies. Finally, we employ a transfer learning strategy by pre-training our DL model on Frag20-Aqsol-100K and fine-tuning it on the small experimental data set, and the fine-tuned model A3D-PNAConv-FT achieves the state-of-the-art prediction on the FreeSolv data set with a root-mean-squared error of 0.719 kcal/mol and a mean-absolute error of 0.417 kcal/mol using random data splits. These results indicate that integrating molecular modeling and DL would be a promising strategy to develop robust prediction models in molecular science. The source code and data are accessible at: https://yzhang.hpc.nyu.edu/IMA.

Two symmetric arginine residues play distinct roles in Thermus thermophilus Argonaute DNA guide strand-mediated DNA target cleavage.

Bacterium Thermus thermophilus Argonaute (Ago; TtAgo) is a prokaryotic Ago (pAgo) that acts as the host defense against the uptake and propagation of foreign DNA by catalyzing the DNA cleavage reaction. The TtAgo active site consists of a plugged-in glutamate finger with two arginine residues (R545 and R486) located symmetrically around it. An interesting challenge is to understand how they can collaboratively facilitate enzymatic catalysis. In Kluyveromyces polysporus Ago, a eukaryotic Ago, the evolutionarily symmetrical residues are arginine and histidine, both of which function to stabilize the plugged-in catalytic tetrad conformation. Surprisingly, our simulation results indicated that, in TtAgo, only R545 is involved in the cleavage reaction by serving as a critical structural anchor to stabilize the catalytic tetrad Asp-Glu-Asp-Asp that is completed by the insertion of the glutamate finger, whereas R486 is not involved in target cleavage. The TtAgo-mediated target DNA cleavage occurs in a substrate-assisted mechanism, in which the pro-Rp (Rp, a tetrahedral phosphorus center with "R-type" chirality) oxygen of scissile phosphate acts as a general base to activate the nucleophilic water. Our unexpected theoretical findings on distinct roles played by R545 and R486 in TtAgo catalysis have been validated by single-point site-mutagenesis experiments, wherein the target cleavage is abolished for all mutants of R545. In sharp contrast, the cleavage activity is maintained for all mutants of R486. Our work provides mechanistic insights on the catalytic specificity of Ago proteins and could facilitate the design of new gene-editing tools in the long term.

Clinical Application of a Modified Local Transposition Flap (Parallelogram Flap) Surgery in Repairing Fingertip Defects.

OBJECTIVE: Parallelogram flap was performed for transverse finger amputation with the loss of distal pulp, nails, and bone. This study aimed to compare the clinical effects of parallelogram flap, antegrade homodigital island flaps, and reverse digital artery island flaps in fingertip reconstruction. PATIENTS AND METHODS: From January 2017 to January 2021, clinical patient data with parallelogram flaps (78 cases), antegrade homodigital island flaps (78 cases), and reverse digital artery island flaps (78 cases) to repair fingertip defects were collected and analyzed. Two hundred thirty-four cases (234 fingers) were included in our study. All operations were performed by one surgical team. The operation time, 2-point discrimination, total active movement, and the Michigan Hand Questionnaire (MHQ) of the injured fingers were recorded to evaluate the therapeutic effect. RESULTS: Parallelogram flaps (group A), antegrade homodigital island flaps (group B), and reverse digital artery island flaps (group C) had survived postoperatively. The operative duration of group A is the shortest (A < B < C, P < 0.05). At the last 6-month follow-up, there was no difference with the 2-point discrimination of the palmar part of the flaps in group A and group B but better than group C (P < 0.05). There was no difference with the total active movement of injured figures in 3 groups (P > 0.05). The MHQ summary scores in group A were much higher than those in group B and group C (P < 0.05). Evaluation of the MHQ subscale performance showed that the overall hand function, activities of daily living, work performance, and pain score had no differences (P > 0.05), but aesthetics and satisfaction score was the highest in group A (A > B > C, P < 0.05). CONCLUSIONS: The reconstruction of transverse finger amputation using parallelogram flaps can achieve a shorter operation time, a more satisfying appearance. Parallelogram flaps and antegrade homodigital island flaps can both achieve a better sensory recovery. Parallelogram flaps is a better choice for reconstruction of transverse finger amputation with the loss of distal pulp, nails, and bone.

Efficacy and safety of anterior cervical discectomy and fusion (ACDF) through mini-incision and posterior laminoplasty (LAMP) for treatment of long-level cervical spondylosis: a retrospective cohort study.

BACKGROUND: The efficacy and safety of anterior cervical discectomy and fusion (ACDF) through mini-incision and posterior laminoplasty for long-level cervical spondylosis were investigated. METHOD: From January 2018 to September 2019, clinical patients data with 3-4 segments (C3-7) cervical spondylotic radiculopathy, cervical spondylotic myelopathy, or mixed cervical spondylosis who received ACDF (42 cases) throughwith mini-incision or LAMP (36 cases) treatment were retrospectively collected and analyzed. The operative time, bleeding volume, incisive length, and hospital stay were recorded. Moreover, the intervertebral height, functional segment height, cervical lordosis, cervical hyperextension and hyperflexion range-of-motion (ROM) and ROM in all directions of the cervical spine before and after the operation were measured. Additionally, all relevant postoperative complications were also recorded. Then, the therapeutic effects of both surgical methods were investigated. RESULTS: Patients in the ACDF group had less bleeding, shorter incision, and fewer hospitalization days than the LAMP group. There was no significant difference in JOA, VAS score of the upper limb, NDI score after surgery between two groups. Postoperative intervertebral height and functional segment height in the ACDF group were significantly higher than those before the operation, and postoperative functional segment height of the ACDF group was significantly higher than that of the LAMP group. Moreover, the postoperative cervical lordosis angle in the ACDF group was significantly larger than the LAMP group. There was no significant difference between preoperative and postoperative ROM in all directions of the cervical spine for the two groups. CONCLUSIONS: Both ACDF through mini-incision and LAMP are effective treatments for long-level cervical spondylosis. However, ACDF through mini-incision shows minor trauma, less bleeding, fast recovery, and it is beneficial for cervical lordosis reconstruction.

Protein-Ligand Docking in the Machine-Learning Era.

Molecular docking plays a significant role in early-stage drug discovery, from structure-based virtual screening (VS) to hit-to-lead optimization, and its capability and predictive power is critically dependent on the protein-ligand scoring function. In this review, we give a broad overview of recent scoring function development, as well as the docking-based applications in drug discovery. We outline the strategies and resources available for structure-based VS and discuss the assessment and development of classical and machine learning protein-ligand scoring functions. In particular, we highlight the recent progress of machine learning scoring function ranging from descriptor-based models to deep learning approaches. We also discuss the general workflow and docking protocols of structure-based VS, such as structure preparation, binding site detection, docking strategies, and post-docking filter/re-scoring, as well as a case study on the large-scale docking-based VS test on the LIT-PCBA data set.

Lin_F9: A Linear Empirical Scoring Function for Protein-Ligand Docking.

Molecular docking is one of the most widely used computational tools in structure-based drug design and is critically dependent on accuracy and robustness of the scoring function. In this work, we introduce a new scoring function Lin_F9, which is a linear combination of nine empirical terms, including a unified metal bond term to specifically describe metal-ligand interactions. Parameters in Lin_F9 are obtained with a multistage fitting protocol using explicit water-included structures. For the CASF-2016 benchmark test set, Lin_F9 achieves the top scoring power among all 34 classical scoring functions for both original crystal poses and locally optimized poses with Pearson correlation coefficients (R) of 0.680 and 0.687, respectively. Meanwhile, in comparison with Vina, Lin_F9 achieves consistently better scoring power and ranking power with various types of protein-ligand complex structures that mimic real docking applications, including end-to-end flexible docking for the CASF-2016 benchmark test set using a single or an ensemble of protein receptor structures, as well as for D3R Grand Challenge (GC4) test sets. Lin_F9 has been implemented in a fork of Smina as an optional built-in scoring function that can be used for docking applications as well as for further improvement of scoring functions and docking protocols. Lin_F9 is accessible through https://yzhang.hpc.nyu.edu/Lin_F9/.

Dataset Construction to Explore Chemical Space with 3D Geometry and Deep Learning.

A dataset is the basis of deep learning model development, and the success of deep learning models heavily relies on the quality and size of the dataset. In this work, we present a new data preparation protocol and build a large fragment-based dataset Frag20, which consists of optimized 3D geometries and calculated molecular properties from Merck molecular force field (MMFF) and DFT at the B3LYP/6-31G* level of theory for more than half a million molecules composed of H, B, C, O, N, F, P, S, Cl, and Br with no larger than 20 heavy atoms. Based on the new dataset, we develop robust molecular energy prediction models using a simplified PhysNet architecture for both DFT-optimized and MMFF-optimized geometries, which achieve better than or close to chemical accuracy (1 kcal/mol) on multiple test sets, including CSD20 and Plati20 based on experimental crystal structures.

Identification and structure-function analyses of an allosteric inhibitor of the tyrosine phosphatase PTPN22.

Protein-tyrosine phosphatase nonreceptor type 22 (PTPN22) is a lymphoid-specific tyrosine phosphatase (LYP), and mutations in the PTPN22 gene are highly correlated with a spectrum of autoimmune diseases. However, compounds and mechanisms that specifically inhibit LYP enzymes to address therapeutic needs to manage these diseases remain to be discovered. Here, we conducted a similarity search of a commercial database for PTPN22 inhibitors and identified several LYP inhibitor scaffolds, which helped identify one highly active inhibitor, NC1. Using noncompetitive inhibition curve and phosphatase assays, we determined NC1's inhibition mode toward PTPN22 and its selectivity toward a panel of phosphatases. We found that NC1 is a noncompetitive LYP inhibitor and observed that it exhibits selectivity against other protein phosphatases and effectively inhibits LYP activity in lymphoid T cells and modulates T-cell receptor signaling. Results from site-directed mutagenesis, fragment-centric topographic mapping, and molecular dynamics simulation experiments suggested that NC1, unlike other known LYP inhibitors, concurrently binds to a "WPD" pocket and a second pocket surrounded by an LYP-specific insert, which contributes to its selectivity against other phosphatases. Moreover, using a newly developed method to incorporate the unnatural amino acid 2-fluorine-tyrosine and 19F NMR spectroscopy, we provide direct evidence that NC1 allosterically regulates LYP activity by restricting WPD-loop movement. In conclusion, our approach has identified a new allosteric binding site in LYP useful for selective LYP inhibitor development; we propose that the 19F NMR probe developed here may also be useful for characterizing allosteric inhibitors of other tyrosine phosphatases.

AlphaSpace 2.0: Representing Concave Biomolecular Surfaces Using ß-Clusters.

Modern rational modulator design and structure-function characterization often concentrate on concave regions of biomolecular surfaces, ranging from well-defined small-molecule binding sites to large protein-protein interaction interfaces. Here, we introduce a ß-cluster as a pseudomolecular representation of fragment-centric pockets detected by AlphaSpace [J. Chem. Inf. Model. 2015, 55, 1585], a recently developed computational analysis tool for topographical mapping of biomolecular concavities. By mimicking the shape as well as atomic details of potential molecular binders, this new ß-cluster representation allows direct pocket-to-ligand shape comparison and can be used to guide ligand optimization. Furthermore, we defined the ß-score, the optimal Vina score of the ß-cluster, as an indicator of pocket ligandability and developed an ensemble ß-cluster approach, which allows one-to-one pocket mapping and comparison among aligned protein structures. We demonstrated the utility of ß-cluster representation by applying the approach to a wide variety of problems including binding site detection and comparison, characterization of protein-protein interactions, and fragment-based ligand optimization. These new ß-cluster functionalities have been implemented in AlphaSpace 2.0, which is freely available on the web at http://www.nyu.edu/projects/yzhang/AlphaSpace2.

An in silico mechanistic insight into HDAC8 activation facilitates the discovery of new small-molecule activators.

Research interest in the development of histone deacetylase 8 (HDAC8) activators has substantially increased since loss-of-function HDAC8 mutations were found in patients with Cornelia de Lange syndrome (CdLS). A series of N-acetylthioureas (e.g., TM-2-51) have been identified as HDAC8-selective activators, among others; however, their activation mechanisms remain elusive. Herein, we performed molecular dynamics (MD) simulations and fragment-centric topographical mapping (FCTM) to investigate the mechanism of HDAC8 activation. Our results revealed that improper binding of the coumarin group of fluorescent substrates leads to the "flipping out" of catalytic residue Y306, which reduces the enzymatic activity of HDAC8 towards fluorescent substrates. A pocket between the coumarin group of the substrate and thed catalytic residue Y306 was filled with the activator TM-2-51, which not only enhanced binding between HDAC8 and the fluorescent substrate complex but also stabilized Y306 in a catalytically active conformation. Based on this newly proposed substrate-dependent activation mechanism, we performed structure-based virtual screening and successfully identified low-molecular-weight scaffolds as new HDAC8 activators.

Incorporating Explicit Water Molecules and Ligand Conformation Stability in Machine-Learning Scoring Functions.

Structure-based drug design is critically dependent on accuracy of molecular docking scoring functions, and there is of significant interest to advance scoring functions with machine learning approaches. In this work, by judiciously expanding the training set, exploring new features related to explicit mediating water molecules as well as ligand conformation stability, and applying extreme gradient boosting (XGBoost) with Δ-Vina parametrization, we have improved robustness and applicability of machine-learning scoring functions. The new scoring function ΔvinaXGB can not only perform consistently among the top compared to classical scoring functions for the CASF-2016 benchmark but also achieves significantly better prediction accuracy in different types of structures that mimic real docking applications.

Exploring fragment-based target-specific ranking protocol with machine learning on cathepsin S.

Cathepsin S (CatS), a member of cysteine cathepsin proteases, has been well studied due to its significant role in many pathological processes, including arthritis, cancer and cardiovascular diseases. CatS inhibitors have been included in D3R-GC3 for both docking pose prediction and affinity ranking, and in D3R-GC4 for binding affinity ranking. The difficulties posed by CatS inhibitors in D3R mainly come from three aspects: large size, high flexibility and similar chemical structures. We have participated in GC4; our best submitted model, which employs a similarity-based alignment docking and Vina scoring protocol, yielded Kendall's τ of 0.23 for 459 binders in GC4. In our further explorations with machine learning, by curating a CatS specific training set, adopting a similarity-based constrained docking method as well as an arm-based fragmentation strategy which can describe large inhibitors in a locality-sensitive fashion, our best structure-based ranking protocol can achieve Kendall's τ of 0.52 for all binders in GC4. In this exploration process, we have demonstrated the importance of training data, docking approaches and fragmentation strategies in inhibitor-ranking protocol development with machine learning.

Lesion Sensing during Initial Binding by Yeast XPC/Rad4: Toward Predicting Resistance to Nucleotide Excision Repair.

Nucleotide excision repair (NER) excises a variety of environmentally derived DNA lesions. However, NER efficiencies for structurally different DNA lesions can vary by orders of magnitude; yet the origin of this variance is poorly understood. Our goal is to develop computational strategies that predict and identify the most hazardous, repair-resistant lesions from the plethora of such adducts. In the present work, we are focusing on lesion recognition by the xeroderma pigmentosum C protein complex (XPC), the first and required step for the subsequent assembly of factors needed to produce successful NER. We have performed molecular dynamics simulations to characterize the initial binding of Rad4, the yeast orthologue of human XPC, to a library of 10 different lesion-containing DNA duplexes derived from environmental carcinogens. These vary in lesion chemical structures and conformations in duplex DNA and exhibit a wide range of relative NER efficiencies from repair resistant to highly susceptible. We have determined a promising set of structural descriptors that characterize initial binding of Rad4 to lesions that are resistant to NER. Key initial binding requirements for successful recognition are absent in the repair-resistant cases: There is little or no duplex unwinding, very limited interaction between the ß-hairpin domain 2 of Rad4 and the minor groove of the lesion-containing duplex, and no conformational capture of a base on the lesion partner strand. By contrast, these key binding features are present to different degrees in NER susceptible lesions and correlate to their relative NER efficiencies. Furthermore, we have gained molecular understanding of Rad4 initial binding as determined by the lesion structures in duplex DNA and how the initial binding relates to the repair efficiencies. The development of a computational strategy for identifying NER-resistant lesions is grounded in this molecular understanding of the lesion recognition mechanism.