Integrating computational and mixture-based screening of combinatorial libraries

Activity landscape characterization has been demonstrated to be a valuable tool in lead optimization, virtual screening, and computational modeling of active compounds. In this work, we present a general protocol to explore systematically the activity landscape of a lead series using 11 2D and 3D structural representations. As a test case we employed a set of 48 bicyclic guanidines (BCGs) with kappa-opioid receptor binding affinity, identified in our group. MACCS keys, graph-based three point pharmacophores, circular fingerprints, ROCS shape descriptors, and the TARIS approach, that compares structures based on molecular electrostatic potentials, were employed as orthogonal descriptors. Based on 'activity cliffs' common to a series of descriptors, we introduce the concept of consensus activity cliffs. Results for the current test case suggest that the presence or absence of a methoxybenzyl group may lead to different modes of binding for the active BCGs with the kappa-opioid receptor. The most active compound (IC50 = 37 nM) is involved in a number of consensus cliffs making it a more challenge query for future virtual screening than would be expected from affinity alone. Results also reveal the importance of screening high density combinatorial libraries, especially in the "cliff-rich" regions of activity landscapes. The protocol presented here can be applied to other data sets to develop a consensus model of the activity landscape.

BackgroundActivity cliffs, which refer to pairs of molecules that are structurally similar but show significant differences in their potency, can lead to model representation collapse and make the model challenging to distinguish them.ResultsOur research indicates that as molecular similarity increases, graph-based methods struggle to capture these nuances, whereas image-based approaches effectively retain the distinctions. Thus, we developed MaskMol, a knowledge-guided molecular image self-supervised learning framework. MaskMol accurately learns the representation of molecular images by considering multiple levels of molecular knowledge, such as atoms, bonds, and substructures. By utilizing pixel masking tasks, MaskMol extracts fine-grained information from molecular images, overcoming the limitations of existing deep learning models in identifying subtle structural changes. Experimental results demonstrate MaskMol’s high accuracy and transferability in activity cliff estimation and compound potency prediction across 20 different macromolecular targets, outperforming 25 state-of-the-art deep learning and machine learning approaches. Visualization analyses reveal MaskMol’s high biological interpretability in identifying activity cliff-relevant molecular substructures. Notably, through MaskMol, we identified candidate EP4 inhibitors that could be used to treat tumors.ConclusionsThis study raises awareness about activity cliffs and introduces a novel method for molecular image representation learning and virtual screening, advancing drug discovery and providing new insights into structure-activity relationships (SAR).Supplementary InformationThe online version contains supplementary material available at 10.1186/s12915-025-02389-3.

Activity cliffs are formed by structurally similar compounds with significant differences in potency and represent an extreme form of structure-activity relationships discontinuity. By contrast, regions of structure-activity relationships continuity in compound data sets result from the presence of structurally increasingly diverse compounds retaining similar activity. Previous studies have revealed that structure-activity relationships information extracted from large compound data sets is often heterogeneous in nature containing both continuous and discontinuous structure-activity relationships components. Structure-activity relationships discontinuity and continuity are often represented by different compound series, independent of each other. Here, we have searched different compound data sets for the presence of structure-activity relationships continuity within the vicinity of prominent activity cliffs. For this purpose, we have designed and implemented a computational approach utilizing particle swarm optimization to examine the structural neighborhood of activity cliffs for continuous structure-activity relationships components. Structure-activity relationships continuity in the structural neighborhood of activity cliffs was relatively rarely observed. However, in a number of cases, notable structure-activity relationships continuity was detected in the vicinity of prominent activity cliffs. Exemplary local structure-activity relationships environments displaying these characteristics were analyzed in detail. Thus, the structure-activity relationships environment of activity cliffs must not necessarily be discontinuous in nature, and local structure-activity relationships continuity and discontinuity can occur in a concerted manner in series of structurally related compounds.

Discontinuity in structure-activity relationships (SARs) is caused by so-called activity cliffs and represents one of the major caveats in SAR modeling and lead optimization. At activity cliffs, small structural modifications of compounds lead to substantial differences in potency that are essentially unpredictable using quantitative structure-activity relationship (QSAR) methods. In order to better understand SAR discontinuity at the molecular level of detail, we have analyzed different compound series in combinatorial analog graphs and determined substitution patterns that introduce activity cliffs of varying magnitude. So identified SAR determinants were then analyzed on the basis of complex crystal structures to enable a structural interpretation of SAR discontinuity and underlying activity cliffs. In some instances, SAR discontinuity detected within analog series could be well rationalized on the basis of structural data, whereas in others a structural explanation was not possible. This reflects the intrinsic complexity of small molecule SARs and suggests that the analysis of short-range receptor-ligand interactions seen in X-ray structures is insufficient to comprehensively account for SAR discontinuity. However, in other cases, SAR information extracted from ligands was incomplete but could be deduced taking X-ray data into account. Thus, taken together, these findings illustrate the complementarity of ligand-based SAR analysis and structural information.

Activity cliffs (ACs) are formed by structurally similar compounds with large differences in activity. Accordingly, ACs are of high interest for the exploration of structure-activity relationships (SARs). ACs reveal small chemical modifications that result in profound biological effects. The ability to foresee such small chemical changes with significant biological consequences would represent a major advance for drug design. Nevertheless, only few attempts have been made so far to predict whether a pair of analogues is likely to represent an AC-and even fewer went further to quantitatively predict how "deep" a cliff might be. This might be due to the fact that such predictions must focus on compound pairs. Matched molecular pairs (MMPs), defined as pairs of structural analogs that are only distinguished by a chemical modification at a single site, are a preferred representation of ACs. Herein, we report new strategies for AC prediction that are based upon two different approaches: (i) condensed graphs of reactions, which were originally introduced for modeling of chemical reactions and were here adapted to encode MMPs, and, (ii) plain descriptor recombination-a strategy used for quantitative structure-property relationship (QSPR) modeling of nonadditive mixtures (MQSPR). By applying these concepts, ACs were encoded as single descriptor vectors used as input for support vector machine (SVM) classification and support vector regression (SVR), yielding accurate predictions of AC status (i.e., cliff vs noncliff) and potency differences, respectively. The latter were predicted in a compound order-sensitive manner returning the signed value of expected potency differences between AC compounds.

Activity cliffs are generally defined as pairs of structurally similar compounds having large differences in potency. The analysis of activity cliffs is of general interest because structure-activity relationship (SAR) determinants can often be deduced from them. Critical questions for the study of activity cliffs include how similar compounds should be to qualify as cliff partners, how similarity should be assessed, and how large potency differences between participating compounds should be. Thus far, activity cliffs have mostly been defined on the basis of calculated Tanimoto similarity values using structural descriptors, especially 2D fingerprints. As any theoretical assessment of molecular similarity, this approach has its limitations. For example, calculated Tanimoto similarities might often be difficult to reconcile and interpret from a chemical perspective, a point of critique frequently raised in medicinal chemistry. Herein, we have explored activity cliffs by considering well-defined substructure replacements instead of calculated similarity values. For this purpose, the matched molecular pair (MMP) formalism has been applied. MMPs were systematically derived from public domain compounds, and activity cliffs were extracted from them, termed MMP-cliffs. The frequency of cliff formation was determined for compounds active against different targets, MMP-cliffs were analyzed in detail, and re-evaluated on the basis of Tanimoto similarity. In many instances, chemically intuitive activity cliffs were only detected on the basis of MMPs, but not Tanimoto similarity.

Synthetic combinatorial methods now make it practical to readily produce hundreds of thousands of individual compounds, but it is clearly impractical to screen each separately in vivo. We theorized that the direct in vivo testing of mixture-based combinatorial libraries during the discovery phase would enable the identification of novel individual compounds with desirable antinociceptive profiles while simultaneously eliminating many compounds with poor absorption, distribution, metabolism, or pharmacokinetic properties. The TPI 1346 small-molecule combinatorial library is grouped in 120 mixtures derived from 26 functionalities at the first three positions and 42 functionalities at the fourth position of a pyrrolidine bis-cyclic guanidine core scaffold, totaling 738,192 compounds. These 120 mixtures were screened in vivo using the mouse 55 degrees C warm water tail-withdrawal assay to identify mixtures producing antinociception. From these data, two fully defined individual compounds (TPI 1818-101 and TPI 1818-109) were synthesized. These were examined for antinociceptive, respiratory, locomotor, and conditioned place preference effects. The tail-withdrawal assay consistently demonstrated distinctly active mixtures with analgesic activity that was blocked by pretreatment with the non-selective opioid antagonist, naloxone. Based on these results, synthesis and testing of TPI 1818-101 and 1818-109 demonstrated a dose-dependent antinociceptive effect three to five times greater than morphine that was antagonized by mu- or mu- and kappa-opioid receptor selective antagonists, respectively. Neither 1818-101 nor 1818-109 produced significant respiratory depression, hyperlocomotion, or conditioned place preference. Large, highly diverse mixture-based libraries can be screened directly in vivo to identify individual compounds, potentially accelerating the development of promising therapeutics.

Activity cliffs (ACs) are formed by structurally similar compounds with large potency differences. Accordingly, ACs reveal determinants of structure-activity relationships. This makes ACs highly interesting and relevant for medicinal chemistry and chemoinformatics. So far, ACs have been defined on the basis of generally applied molecular similarity and potency difference criteria. We present the first assessment of ACs taking target set-dependent compound potency distributions into account, leading to a new target set-dependent definition of ACs. The formation of these ACs is analyzed in detail. Second-generation ACs are obtained on the basis of target set-dependent potency difference thresholds. Compared with generally defined ACs, target set-dependent ACs have further increased medicinal chemistry relevance.

Very small chemical changes in active compounds causing large potency effects are of particular interest in medicinal chemistry and drug design. We have systematically searched active compounds with available high-confidence activity data for pairs of structural analogs with dual-atom replacements and additional analogs with corresponding single-atom replacements. From ~287,000 unique qualifying compounds with activity against nearly 1900 unique targets, ~3500 target-based analog pairs with dual-atom replacements were identified. These included 852 pairs with significant differences in compound potency, representing a set of previously unobserved activity cliffs. Comparing these pairs with corresponding single-atom replacement analogs, which were frequently identified, made it possible to systematically analyze how potency changes propagated from single- to dual-atom replacements. The analysis uncovered different potency effects and revealed that individual atom replacements were often decisive for activity cliff formation. For a limited number of activity cliffs, X-ray structures of targets in complex with cliff compounds were available, which aided in rationalizing potency alterations among analogs with single- or dual-atom replacements. The analog pairs identified herein provide a rich resource of structure-activity relationship information and attractive test cases for calibrating computational methods.

Tyrosine1 and phenylalanine4 in dynorphin A (Dyn A) have been reported to be important residues for opioid agonist activity and for potency at kappa receptors. The glycine residues in the 2 and 3 positions of dynorphin A may affect the relative orientation of the aromatic rings in positions 1 and 4, but their flexibility precludes careful analysis. To examine these effects on dynorphin A, we previously have synthesized the linear analogues [D-Ala3]Dyn A(1-11)-NH2 (2) and [Ala3]Dyn A(1-11)-NH2 (3) and reported their biological activities. Analogues 2 and 3 displayed affinities for the central kappa opioid receptor (IC50 = 0.76 and 1.1 nM, respectively) similar to that of Dyn A(1-11)-NH2 (1) (IC50 = 0.58 nM) and greatly enhanced selectivities for kappa vs mu and kappa vs delta receptors (IC50 ratios of 350 and 1300 for 2, and 190 and 660 for 3, respectively). These results suggest that the structure and lipophilicity of the amino acid present in position 3 of Dyn A(1-11)-NH2 as well as the conformational changes they induce in the message sequence of dynorphin have important effects on potency and selectivity for kappa opioid receptors. To further investigate structure-activity relationships involving the residue at the 3 position of Dyn A(1-11)-NH2, a series of Dyn A analogues with aromatic, charged, and aliphatic side chain substitutions at the 3 position was designed, synthesized, and evaluated for their affinities for kappa, mu, and delta opioid receptors. It was found that analogues with lipophilic amino acids at the 3 position of Dyn A(1-11)-NH2 generally displayed higher affinity but similar selectivities for the kappa receptor than analogues with charged residues at the same position. It is suggested that the structural, configurational, and steric/lipophilic effects of amino acids at position 3 of Dyn A(1-11)-NH2 may play an important role in potency and selectivity for the kappa receptor.

Systematic description of structure-activity relationships (SARs) of data sets and structure-property relationships (SPRs) is of paramount importance in medicinal chemistry and other research fields. To this end, structure-activity similarity (SAS) maps are one of the first tools proposed to describe SARs using the concept of activity landscape modeling. One of the major goals of the SAS maps is to identify activity cliffs defined as chemical compounds with high similar structure but unexpectedly very different biological activity. Since the first publication of the SAS maps more than ten years ago, these tools have evolved and adapted over the years to analyze various types of compound collections, including structural diverse and combinatorial sets with activity for one or multiple biological end points. The development of SAS maps has led to general concepts that are applicable to other activity landscape methods such as "consensus activity cliffs" (activity cliffs common to a series of representations or descriptors) and "selectivity switches" (structural changes that completely invert the selectivity pattern of similar compounds against two biological end points). Herein, we review the development, practical applications, limitations, and perspectives of the SAS and related maps which are intuitive and powerful informatics tools to computationally analyze SPRs.

For Abstract see ChemInform Abstract in Full Text.

This unit presents a combinatorial library method that consists of the synthesis and screening of mixture-based synthetic combinatorial libraries (SCL) of peptide molecules. The protocols employ peptide SCL to identify peptides recognized by mAbs and T cells. The first protocol uses a positional scanning SCL (PS-SCL) made up of hexapeptides to identify antigenic determinants recognized by mAbs. The 120 mixtures in the hexapeptide PS-SCL are tested for their inhibitory activity in a competitive ELISA. The second protocol uses a decapeptide PS-SCL to identify T cell peptide ligands. The 200 mixtures of the decapeptide PS-SCL are tested for their ability to induce proliferation. Support protocols cover optimization of the assay conditions for each mAb or T cell, to achieve the best level of sensitivity and reproducibility, and preparation of a hexapeptide PS-SCL, along with deconvolution approaches.

To address the imbalanced data problem in molecular docking-based virtual screening methods, this paper proposes a virtual screening method for drug proteins based on imbalanced data mining, which introduces machine learning technology into the virtual screening technology for drug proteins to deal with the imbalanced data problem in the virtual screening process and improve the accuracy of the virtual screening. First, to address the data imbalance problem caused by the large difference between the number of active compounds and the number of inactive compounds in the docking conformation generated by the actual virtual screening process, this paper proposes a way to improve the data imbalance problem using SMOTE combined with genetic algorithm to synthesize new active compounds artificially by upsampling active compounds. Then, in order to improve the accuracy in the virtual screening process of drug proteins, the idea of integrated learning is introduced, and the random forest (RF) extended from Bagging integrated learning technique is combined with the support vector machine (SVM) technique, and the virtual screening of molecular docking conformations using RF-SVM technique is proposed to improve the prediction accuracy of active compounds in docking conformations. To verify the effectiveness of the proposed technique, first, HIV-1 protease and SRC kinase were used as test data for the experiments, and then, CA II was used to validate the model of the test data. The virtual screening of drug proteins using the proposed method in this paper showed an improvement in both enrichment factor (EF) and AUC compared with the use of the traditional virtual screening, for the test dataset. Therefore, it can be shown that the proposed method can effectively improve the accuracy of drug virtual screening.

Identification of a small molecule inhibitor of the aminoglycoside 6'-N-acetyltransferase type Ib [AAC(6')-Ib] using mixture-based combinatorial libraries