Identication of Isosteric Replacements of Glycosyl Domain of Ligands by Data Mining

Biologically equivalent replacements of key moieties in molecule rationalizes scaffold hopping, patent busting or R-group enumeration, yet heavily depending upon the expert-dened space therefore is subjective and might be biased to the chemistries they get used to. Most importantly, these explorations are often informatively incomplete since it is often conned within try-and-error cycle, only meaning what kind of substructures are suitable for the replacement occur, but fail to disclose the driving forces to support such interchanges. The Protein Data Bank (PDB) repository involving receptor-ligand interactional information reminds poorly exploited. However, manual screening the PDB become almost impossible to excavate the bioisosteric know-how with the exponentially increase of data. Therefore, a textual content parsing workow is developed to automatedly mine local structural replacement (LSR) of specic structure. Taking the glycosyl domain for instance, a total of 41652 replacements that overlap on nucleotide ribose were identied and categorized based on their SMILE codes. Predominately ring system, such as aliphatic aromatic ring, yet amide and sulfonamide replacement also occurred. We believe these ndings may enlighten medicinal chemists to design and optimize ligand structure using bioisosteric replacement strategy.


Introduction
Medicinal chemists are always keen to improve the potency of small-molecule toward their biological targets using variety of computational approaches, exempli ed by high-throughput screening, quantitative structureactivity relationship and fragment-based drug design, but the successful rate of drug discovery project remain low [1] and identi cation of potent compounds is expensive. However, it is often observed that some biomacromolecules accommodate endogenous ligands (EL) with high a nity. Meanwhile minor modi cation of EL may trigger massive activity cliff and can be readily accepted by the live organism given their structural intimacy. Therefore, reassemble of EL skeleton using partial structural replacement/exchange strategy is often under consideration to optimise the pharmacodynamics properties of the leads or coin new drug candidates. Although handful of means to obtain these chemical building blocks are available, bioisosteric transformation is among prioritised hierarchy thanks to its faithful identity in terms of molecular recognition [2]. Bioisosteric information could be attained either by applying medicinal chemistry knowledge or by mining databases.
as computer readable les, therefore facilitate the digitalized compilation of bioisosteric replacements. ChEMBL [4] is freely accessible database of more than 1.9 million small molecules with bioassay data curated from literature. BIOSTER, for example, contains bioisosteric transformations collected from medicinal chemistry literature published during past 4 decades. Based on these data and the Matched Molecular Pair (MMP) approach, molecules in ChEMBL that display bioisosteric features are identi ed, which also allows for the leverage of potential changes in biological properties with bioisosteric transformation; the above-mentioned information were currently presented as SwissBioisostere database for both non-commercial and commercial users [5]. Spark™(Cresset, UK) [6] screens a database of more than 600,000 fragments to pinpoint bioisosteres that display similar steric and electronic features as the interest domain of molecules. Recently, bioisosteric analogs are identi ed by a deep neural network trained on a large corpus of experimentally validated analogs extracted from medicinal chemistry knowledge accumulated in nearly fty years efforts [7].
Rapidly growing openly accessible structural databases, including The Cambridge Structural Database (CSD) [8,9]and Protein Data Bank (PDB) [10] enable another opportunity to obtain new bioisosteric replacements in more automatic and robust manner. Through data mining and informatic curation [11], massive valuable bioisosteric know-how related with drug discovery and development can be disclosed. Frank et al [12] used crystal structure information from the CSD to inclusively study the geometrical and energetic aspects of the tetrazole-carboxylic acid bioisosterism by comparison of the conformational preferences and intermolecular interactions. After superimposition of holo proteins in the PDB with a reference protein, fragments reside in the same binding site are considered as potential bioisosteric candidates [13]. Following the similar idea, the query and the reference ligands complexed with the same protein were fragmented into a set of fragments and compared each other by their volume overlap, the pair with score higher than a given threshold forms a bioisosteric peer [14]. Further on, the similarity of binding site subpocket is quanti ed based on pharmacophore ngerprints, hence enables the both intra-and inter-family comparisons of proteins for bioisosteric replacements for ligand substructures [15]. FragVLib [16], a virtual library of fragments, allows for bioisosteric replacements identi cation based on a subgraph matching tool that nds similar binding pockets according to their 3D structures and chemical similarity of the atoms. Desaphy et al developed sc-PDB-Frag [17] that implements bioisosteric searches by converting protein-ligand interaction patterns to graphs; bioisosteres are de ned as any pair of ligands that share similar interaction patterns with their reference protein.
On the other hand, it is often di cult for non-computer science background ones to obtain such information, so as to refer to drug design and lead optimization. Therefore, the effort toward user-friendly web servers that does not require computational or programming skills but is favourable to medicinal chemists to quickly search and get new ideas about possible bioisosteric replacements have been released. For instance, BoBER [18] identi ed bioisosteric replacements using local binding site; ProBiS-ligands [19] sought for similar local spatial arrangements of physiochemically similar surface functional groups in binding sites; taking liganded structure as an input and choosing speci c substructures want to replace, FragRep try to nd suitable bioisosteric fragments that where the structure is different, while interaction patterns with the protein pocket are similar [20].
Ribose (a naturally occurring pentose sugar), commonly seen as D-ribose, is indispensable components of nucleotides and primarily for the assembly of RNA in all living organisms [21]. It is also part of Ribo avin (vitamin B2) [22], adenosine triphosphate (ATP). Closely related sugar 2-deoxyribose is the building block of DNA [23]. ADP and AMP, the metabolites of ATP, are more stable and are common represented in macromolecule complex as endogenous ligands. The ribose containing molecules play critical pharmacology roles, for example, ADP ribose is a speci c agonist of the purinergic P2Y1 receptor [24], leading to Ca 2+ mobilization in rat pulmonary arterial smooth muscle cell [25]. Cyclic ADP-ribose and NAADP function as Ca 2+ messengers and Ca 2+ stores [26] in cells. Chemically, ribose is a ve-member ring fragment composed of four carbon and one oxygen atoms. Three of the carbons attached to hydroxyl groups (1'-, 2'-and 3'-). The fourth carbon attached to the fth carbon atom connecting a hydroxyl group (5'-). Intuitively, hydroxyl-rich structure makes ribose itself hydrophilic. In practice, 1'-and 5'-hydroxyl of ribose are normally substituted or replaced by hydrophobic moieties.
Isosteric replacement of ribose groups is a classic practice in medicinal chemistry. For example, the modi cation of ribose of nucleosides, has led to the discovery of several drugs ( Figure 1): such as cancer therapeutics Floxuridine (FUdR, 5-uoro-2-deoxyuridine) [27] where the 2'-hydroxyl of 5-uoro-uridine is absent. Cytarabine (Ara-C) is a stereoisomer of cytidine with the D-ribose [28], replaced with D-arabinose. Antivirals such as Vidarabine (Ara-A) [29] combined an adenine base and a D-arabinose sugar and acyclovir host a truncated ribose structure comparing to guanosine. Zidovudine, or azidothymidine [30] has the 3'-hydroxyl replaced by azido moiety. Other nucleoside analogues involved in inhibition of HIV-1 reverse transcriptase, including Didanosine [31] and Zalcitabine[32] whose 2'-and 3'-hydroxyl of ribose are totally deleted. Blocking the puckering Zalcitabine deoxyribose ring by introducing a double bond between 2'-and 3'-carbon gave rise to another HIV-1 reverse transcriptase inhibitor Stavudine [33]. The systematic structure-activity relationship investigation of nucleoside glycosyl domain also suggested the modi cation of ribose is a promising method toward successful drug development [34]. The above-mentioned attempts can be thought as nucleosides in which the nitrogenous bases are attached to unnaturally occurred ribose via a beta-N(1)-glycosidic bond and ribose fragment undergoes either stereoisomerism, hydroxyl truncation and chemical modi cation, but nearly experience the scaffold change. Hence, brand-new ribose isosteres/bioisosteres are highly desired to generate new analogues with improved properties.
Another endogenous ligand that underwent considerable modi cation of ribose [35] is cyclic ADP-ribose (cADPR), a signalling molecule that has been shown to regulate Ca 2+ mobilization intracellularly. cADPR consist of two ribose, namely "northern" and "southern" one [36]. The modi cation of cADPR ribose has led to a few chemical entities with useful pharmacology properties. For example, cyclic aristeromycin diphosphoribose(cArisDPR), featured the furanose oxygen in "southern" ribose replaced by methylene, was a hydrolysis resistant agonist; with the half-lives improved from 15 minutes of cADPR to 170 minutes of cArisDPR when it is incubated in sea urchin egg homogenates [37]. However, 3'-O-methyl-cADPR based on "southern" ribose substation became an antagonist of cADPR induced Ca 2+ release [38]. "Southern" ribose appeared nonessential for the binding of cADPR to the human ADP-ribosyl cyclase CD38 catalytic domain since the replacement of the N9-ribose with a butyl chain generates an analogs which inhibited the hydrolysis of cADPR [39]. Similarly, the substitution of "northern" ribose by different alkene chain led to several cADPR analogues who were able to permeate in intact human Jurkat T-lymphocytes and act as agonists [40]. 2'-NH 2 -cADPR, with an amino group replacing the 2'-hydroxyl group of cADPR in the "northern" ribose, was an agonist in the T-lymphocyte system with the EC 50 of 7 μM as compared to 13 μM of cADPR [41] and hydrolyzed nearly 100-fold slower than cADPR [42]. Other analogues of cADPR, cyclic ADP-4-thioribose, in which the "northern" ribose of cADPR was replaced by a 4-thioribose was completely resistant in rat brain microsomal extract and induced the release of Ca 2+ ions in a concentration-dependent manner with an EC 50 value of 36 nM in sea urchin egg homogenate testing, while cADPR and cADPcR gave the EC 50 value of 214 and 54 nM, respectively [43].
The substitution of the ribose often occurred in 2'-and 3'-position where the number of both hydrogen bond donors and acceptors are reduced, which is also important for cell permeability, especially for ligands targeting to central nervous system (CNS). For example, the polarity of 4-nitrobenzylthioinosine reduced by replacing ribose moiety with substituted (aryl)benzyl group. These chemical properties very different replacement led to two equally active analogues with Ki value of 39 nM but 101 and 85 Å 2 of polar surface area (PSA) individually [44], a signi cant decrease comparing to PSA of 154 Å 2 of parent compounds. In some cases, the ribose moiety is not important for the binding, replacement the ribosyl group with a hydrophobic group might be rational because the polar hydroxyl groups of the ribose moiety are entropically unfavourable (require higher desolvation energy) but have no contribution to binding a nity. For instance, the change of ribose moiety in compound 1.6 (IC 50 , 100 mM) with benzyl compound (IC 50 , 91 mM) showed better potency [45]. If structurebased drug design or other computer tools suggest that the hydrogen bonds between ribosyl hydroxyl and protein residues contributed little to the overall binding a nity [46], while hydrophobicity dominated binding pocket prefer the hydrophobic moiety, then the replacement of ribose with a alkyl group should be placed. For example, cyclohexylethyl group replacement (IC 50 , 97 mM) of ribose (IC 50 , 118 mM) lead to comparable potency compound that minimize desolvation costs and gain van der Waals interactions with fructose 1,6bisphosphatase [47]. In the process of investigation adenosine 5′-diphosphoribose (ADPR) to modulate transient receptor potential melastatin 2 channel, it was found that replacing the terminal ribose of ADPR with a cyclopentyl group resulted in a weak antagonistis yet but remain attractive because the anomeric center stereochemistrys was removed, which negates the possibility of intramolecular interaction between the pyrophosphate and a ribose hydroxyl group, and therefore may lead to instability [48].
Although considerable ribose modi cation/replacement protocols have been elaborated for variety of applications, few of them speci cally address the necessity ribose replacement from molecular interactions perspective, meaning the identi cation of proper ribose bioisosteres remain the trial-and-error circle. Given the importance the bioisosterism in contemporary medicinal chemistry practice [49] and wide application [50,51], for the rst time, we have developed a cheminformatics work ow to mine the PDB for isosteric replacements of biologically important phosphate [52]. Besides the classical isosteres of phosphate, such as carboxylate, sulfone or sulfonamide [53], unexpected replacements that do not conserve charge or polarity, such as aryl, aliphatic, or positively charged groups were found. The disclosed results were timely applied for online computational tools development [54], cheminformatics re nement [55], probe ligands [56] and Plasmodium falciparum pyrophosphatase inhibitors [57] synthesis, demonstrated the necessity and demand of phosphate replacement in drug development streamline. The data mining work ow we previously presented could be generalized to exploit structural isosteres of any chemical fragments interested. Ideally, the query group could be de ned by the user and the work ow could automatically select the proper reference compounds for data extraction. In this study, we aim not only to identify the possible structural replacements of ribose moiety, but understand the mechanism which drive the occurrence of replacements speci cally from molecular recognition perspective.

Methods
We relied on ve external programs to implement the work ow ( Figure 2A): Blastp compares the query protein sequences to sequence of PDB and outputs protein homologues with prede ned the statistical signi cance; TM-align generates optimized residue-to-residue alignment of two protein structures and supply optimal superposition of the them; ShaEP calculates the tness of overlaid molecular fragments; Babel translate the extracted molecular fragment into SMILES string; and EMBOSS Needle implements global pairwise sequence alignment. Having these tools complied and integrated with Python scripts, the work ow rstly retrieves from PDB the three-dimensional structures of protein that homologize with protein bound with a nucleotide designated as "reference ligands". For the protein with multiple identical chains, only single representative one is kept and append PDB code_chain identi er as output name. Secondly, the work ow acquires and aligns homologues of each reference protein and preserves target proteins with a non-nucleotide ligand bound at identical site of the reference ligand. Those with no ligand atom at the structural isostere site are removed. Thirdly, local structural replacements are extracted based on sphere whose centre de ned by O2', O3' and O4' atoms of nucleotide, setting radius of 2.5 Å. Lastly, the fragments are categorized according to equivalent SMILES codes into 16 exclusive groups; cycle group is complex and tedious, therefore further decomposed into 15 subgroups for clarity.

Results And Discussion
Data Curation The prototype of data organization, SMILES-based folders and contents of the running presented are almost identical to already published by us [52], but several features are mended: The scripts, previously written in  (Table 1). In total 2734 reference proteins and 40929 local structural replacements are identi ed. The bias of these data concerning different nucleotides, for instance 759 cases of ADP binding protein versus only 24 cases of CDP, re ects the massive crystallographic and drug development project to purine over pyrimidine derivatives (Table 1). The substructures extracted are hierarchically organized ( Figure 2B) according to a decomposition SMILES code and the hierarchy is given as a text le (Supporting Information 1). The results are also archived into folder and downloadable, composed of structure les in pdb format and organized hierarchically into folders (Supporting Information 2). These data can be visualized by using computational tools, such as PyMol [58].  Table 1).
To focus on the ribose replacements of most interest while put aside many uninteresting or nearly identical ones, and ignore the very small replacements (Supporting Information 5) containing fewer than three atoms, a classi cation SMILES code-oriented paradigm is proposed for the extracted fragments with a structural t on ribose groups. The same folder naming rules is followed as previously. The largest and complex folder is cycle one, it is reasonable since ribose itself is ring-based structure. This folder, therefore is further divided into subfolders, and nominated as cycle.*, herein * stands for the speci c atom included in the extracted ring fragments. For instance, cycle.S means the extracted substructures in this folder contain sulphur atom. Noticeably, the cycle.P folder are most dominant (Figure 3) except cycle.other folder, in particular when Uracil, Cytosine and Guanine-binding proteins are references.

Enumeration of Ribose Structural Isosteres
Most of ligands are found to be anchored at the adenine-ribose sites, as can be inferred from the higher number of replacements of adenylic ribose ( Figure 3A-C) compared to guanylic, cytidylic and thymidinic ones ( Figure  3D-L). Among all possible LSR of ribose for different nucleotide except referencing with ADP and ATP, cyclic moieties containing phosphorus (designated as cycle.P folder, the grey pie in Figure 3A, D-L) remarkably outnumber other fragment replacements.
Classi cation of the structural replacements is important while complicated task. Except SMILES codes-based data sorting method described above, we furthermore tried to cluster the fragments according to the shape and electrostatic potential scores calculated by ShaEP, which are integrated into the computational work ow. For each cluster of fragments, neither electrostatic potential nor shape favour the classi cation of the fragments, implying the di culty to address this task (Figure 4). We therefore maintain the SMILES based segregation, due to the advantage of helping the analysis and interpretation.
Apolar aliphatic ring. The data relative to the examples presented in this section, replacements of ribose is given in Table 2. The examples of SARs discussed are presented in Table 3.
Apolar aliphatic ring commonly appeared in drug development project [60], but hardly addressed as an isostere of ribose. Our survey disclosed several cases of its structural replacement as such. In bovine ribosomal S6 kinases RPS6KB1 (S6K1) (example 1, Figure 5A-C), ligand 15e in Table 3 offers the highest IC 50 of 0.0198 μM among N-1 substituted benzimidazole oxadiazole analogs [61]. The compound with ethyl replacement (15c in Table 3, IC 50 of 0.366 μM) presents 18-fold lower IC 50 toward S6K1, the compound 15d with benzyl replacement completely loses activity, suggesting that a pocket small and con ned exists, only cyclopropylmethyl group can tightly ts the hydrophobic pocket formed by Gly50, Tyr54, Val57, and Phe327 ( Figure 5B). Besides, compounds containing this ring system had been also been reported as inhibitors of the AGC kinases mitogen and stress-activated protein kinase (MSK1) [62], Rho kinase, and ATPbinding site of protein kinase [63]. Although apolar aliphatic ring cannot mimic the hydrogen-bonding interactions of the ribose 2'-and 3'-hydroxyls toward biological target, it is often observed that they occupy space close to where the ribose ring of nucleoside binds, exempli ed by KS4's [64] cyclobutyl moiety ( Figure  5F) in c-Src kinase, cyclopentane group( Figure 5I) of ML8 [65] in phosphatidylinositol 3-kinase and cyclohexylmethyl group of NW1 in cyclin-dependent kinase ( Figure 5L) [66]. A closer look underline that the driven force anchoring cycloaliphatic ring in proximity to the ribose is hydrophobic interactions established by a hydrophobic patch on the glycine-rich loop of the receptor, assisted by the bulk of residues of valine for instance. The fused bicyclohexyl group of ligand BYB (VX-787N in reference [67]) packed with Phe325 residue of in uenza B cap-binding domain [67], where ribose moiety of GTP occupied the space of biclohexyl ( Figure 5O).  Table 4.  Six-member heterocyclic ring can also be the structural replacement of ribose. In Homo sapiens PDE4B tetrahydropyrimidin-2(1H)-one is close to ribose ring of AMP, the carbonyl of 0CP ( Figure 6B) align with 3'-hydroxyl of AMP ribose; a water is conserved but shift a bit between structures complexes ( Figure 6C) as such that bridge two histidine resides and ligands' oxygen atoms, implying 3'-hydroxyl of AMP is hydrogen bond acceptor. The 4-amino-cyclohexyl substituent at the C2 position of the purine ring in 24A is oriented toward the ribose-binding portion of the ATP site. Herein, the amino group of 24A is 2'-hydroxyl the counterpart of ( Figure   6E) in term of molecule recognition pattern as such a conserved water molecule bifurcately hydrogen bonded to the main NH of Ser345 and side chain of Asp348 [68]. Besides, the amino of 24A makes extensive contacts with backbone NH of Gln275 through a bridge water molecule.
Aromatic ring. The data relative to the examples presented in this section, replacements of ribose is given in Table 5. The UMP binding pocket of in uenza A virus endonuclease is open and large, therefor can accept molecules with different scaffold. For instance, T-shaped ligand 0N8 access the pocket while has no fragment overlaid with U5P ribose [69]. Butter y-like ligand E4Z co-crystalize to in uenza A virus endonuclease, with one wing consisting of a metal chelating polar head-group and another wing, a lipophilic tail-group that makes van der Waals contacts with speci c residues of the active site pocket. One of the tail-group aromatic ring of E4Z space the position of U5P ribose ( Figure 7B), intramolecularly π-π stacked with dihydropyridine (centroid distance of benzene ··· pyridine 3.8 Å); the binding of E4Z to in uenza A virus endonuclease signi cantly stabilizes the structure, with more than +32 °C ΔT m from 46 °C to 78 °C [70]. Similar to E4Z, by replacement of morpholine with tri uoropropane ( Figure 7D) to obtain analogue compound R07 [71] which has identical binding mode as such. The chloro-benzyl group of ligand BYE contact via hydrophobic interaction with PDE9's subpocket which made of residues His252, Met365, Leu420, Tyr424, and Phe456 [72]; the subpocket is also occupied by ribose of 5GP who is endogenous ligand of PDEs [73]. Ligand BYE had an IC 50 of 88 nM for the wild type PDE9A, its enantiomer PDB is 4 times more potent (22 nM), attributed to different orientations of uoromethyl groups of BYE and PDB [72].  Table 6. The examples of SARs discussed are presented in Table 7. In Pim1-AMP co-crystal structure, the ribose group is found deep inside the pocket and the 3'-OH joins an extensive hydrogen bond network formed inside the pocket by Lys67, Glu89, the backbone NH of Phe187, and a water molecule ( Figure 8A). The phenol of LI7 [76] sits inside the same pocket ( Figure 8B), with its hydroxyl group participating the highly conserved water mediated hydrogen bond network near Lys67, Glu89 and Phe187 in a direction similar to the 3'-OH of AMP. The same pocket also accommodates the catechol of QUE, with its two hydroxyl groups being involved ( Figure 8D) hydrogen bond network and the water molecule experiencing negligible shift comparing to reference [74]. Ligand QUE in Table X [75]) inside the binding pocket whereas MYC and MYF ip the B ring out toward solvent.
Substituted heteroaromatic rings. The data relative to the examples presented in this section, replacements of ribose is given in Table 8. The examples of SARs discussed are presented in Table 9.   Table 9 Speci c Examples of SARs that Illustrate Ribose Replacements Among homo sapiens phosphodiesterases 4B (PDE4B) dialkoxyphenyl inhibitors, the oxazolidinones moiety of mesopram (5RM) [79] insect with AMP ribose and protrude into M pocket, with its carbonyl participating the hydrogen bond network involving the backbone carboxyl oxygen of Leu510, Cme430 and a conserved water ( Figure Figure 9E). Noticeable, the continuous fragment-based optimization of 0FS has led to the most potent compound (6 in reference [80]) with IC 50 of 0.42 nM toward Pim1. In Pim-1, the pyrazolopyrimidone core of 1OA lays on the space AMP ribose, with hydroxyl group direct hydrogen bonding to Lys 67 residue, pyrazolopyrimidine interacting with Val 52 residue through hydrophobic interaction; [81] 3'-OH of AMP also makes the key hydrogen bond contact with Lys67 residue.
Heteroaromatic rings. The data relative to the examples presented in this section, replacements of ribose is given in Table 10.  Amide. The data relative to the examples presented in this section, replacements of ribose is given in Table 11. The examples of SARs discussed are presented in Table 12. The carboxyl oxygen of 0FK accept a hydrogen bond from the Lys67 residue directly, while NH of 0FK, backbone NH of Phe160 and residue of Glu62 form hydrogen bond forms hydrogen bond network [80] in AMP binding site of Pim-1 (Figure 11 B); together the terminal amide of 0FK is bifurcatedly replaced by 3'-OH of AMP (Figure 11 A). In the DAPK3 complex, 7CP's s γ-lactam carbonyl establishes a hydrogen bond with a water molecule ( Figure  11 E) in the active site which itself is involved in a hydrogen bond network with Ser21 backbone carbonyl [85], in which AMP ribose and pyridone ring interweaved. Ligand JUJ (G150 in reference [86]) localized within the GTP binding pocket of cyclic GMP-AMP synthase (Figure 11 H), with the part of the hydroxyl-ethanone side chain attached to the non-planar six-membered ring sharing the same space with UTP ribose moiety (( Figure 11 I), hydrogen-bonded with Ser434 (Figure 11 H). Pseudomonas aeruginosa RmlA was screened against a library and identi ed HNR (compound 1 in Table 12) give IC 50 values of 0.22 μM, 100% inhibition at 10 μM, both 942 and NIQ (compound 3, 4 in Table 12 conrrespondingly) show about 30% inhibition at 10 μM, while 942 is 36% more potent than NIQ at 60 μM concentration [84]. Ligand 942 is commercially available analogue of HNR while less potent, indicated that replacement of the sulfonamide in HNR by an amide in 942 or an alkyl substituent in NIQ was unfavorable. Herein, the carboxyl oxgen of amide in 942 hydrgen bonded to backbone ( Sulfonamide. The data relative to the examples presented in this section, replacements of ribose is given in Table 13. The sulfonamide moiety of LZ2 forms water mediated hydrogen bond to the backbone carbonyl of Gln131 and to the residue of Asp86 (Figure 12 B), partly occupy the ribose pocket [87], give IC 50 of 120 μM toward CDK2.
Isoquinolinesulfonamide protein kinase inhibitor IQS acts in competition to ATP toward cAPK, give IC 50 of 1.2 μM. It shows the superposition with ATP (Figure 12 F)

Availability of data and materials
Scripts enabling reproduction of all the results obtained in the study are available at https://github.com/Yuezhou-Project/IsoIdenti er.

Competing interests
The authors declare that they have no competing interests.
using fragment-based X-ray crystallography and structure based drug design, J Med Chem, 51 (2008) Figure 1 Drugs based on the modi cation of nucleosides ribose   ligands are named according to their PDB 3-letter codes, the proteins are named according to PDB 4-letter codes. Putative hydrogen bonds, pi-pi staking interaction are shown as yellow dotted lines and distance is labelled. The carbon atom of ribose structural replacements in target ligand are highlighted in green while others are showed in black.